skip to main content
research-article
Open Access

Smart City Construction and Management by Digital Twins and BIM Big Data in COVID-19 Scenario

Authors Info & Claims
Published:06 October 2022Publication History

Skip Abstract Section

Abstract

With the rapid development of information technology and the spread of Corona Virus Disease 2019 (COVID-19), the government and urban managers are looking for ways to use technology to make the city smarter and safer. Intelligent transportation can play a very important role in the joint prevention. This work expects to explore the building information modeling (BIM) big data (BD) processing method of digital twins (DTs) of Smart City, thus speeding up the construction of Smart City and improve the accuracy of data processing. During construction, DTs build the same digital copy of the smart city. On this basis, BIM designs the building's keel and structure, optimizing various resources and configurations of the building. Regarding the fast data growth in smart cities, a complex data fusion and efficient learning algorithm, namely Multi-Graphics Processing Unit (GPU), is proposed to process the multi-dimensional and complex BD based on the compositive rough set model. The Bayesian network solves the multi-label classification. Each label is regarded as a Bayesian network node. Then, the structural learning approach is adopted to learn the label Bayesian network's structure from data. On the P53-old and the P53-new datasets, the running time of Multi-GPU decreases as the number of GPUs increases, approaching the ideal linear speedup ratio. With the continuous increase of K value, the deterministic information input into the tag BN will be reduced, thus reducing the classification accuracy. When K = 3, MLBN can provide the best data analysis performance. On genbase dataset, the accuracy of MLBN is 0.982 ± 0.013. Through experiments, the BIM BD processing algorithm based on Bayesian Network Structural Learning (BNSL) helps decision-makers use complex data in smart cities efficiently.

Skip 1INTRODUCTION Section

1 INTRODUCTION

COVID-19 swept the world, endangered lives, subverted enterprises, and set off a global economic recession in just a few months. During this period, to maintain connectivity and exchange problems and solutions, smart cities needed a lot of collaborative technologies. The development of smart city, especially for travel, focuses on three key areas: better road data, real-time control system of existing infrastructure, and the intersection of vehicle and fleet technology and infrastructure in the future. COVID-19 has been proved to be a catalyst for smart city change, and modern technology is providing the necessary means to shape the future. It is expected that the journey will become interconnected, have less waiting time, and a more seamless traffic flow. Urban infrastructure that meets active travel needs includes flexible pedestrian-only areas and bicycle lanes, as well as public transport services that can adapt to changing crowd peaks.

For Germany, the direction of industry 4.0 is very accurate, that is, it uses the advantages of the German manufacturing industry to continue to maintain the advantages of national development in the new era. Industry 4.0 includes the transformation from centralized control to decentralized enhanced control, which is of great significance to the construction industry. The massive procurement and transportation of buildings has been subversive in industry 4.0. Under the wave of new infrastructure, loads of internet platforms transform online services into cloud services. At this stage, cloud computing and the Fifth Generation (5G) communication technology can bring tremendous changes to smart cities [13]. Mature urban brains provide the largest collection of scenes, while new infrastructure shall be scene-driven and value-driven. Consequently, smart cities usher in better development in the wave. Cities are huge settlements where humans gather. As an effective field, cities can safely and stably operate the flow of people, logistics, information, resources, and energy; they are the hubs for the life service industry [4]. After nearly a decade of smart city construction, urban sensors are everywhere, smart terminals are publicly deployed, and smart applications based on massive multi-dimensional data analysis continue to appear. In the digital information era, smart city construction is at a critical stage of quantitative change to qualitative change driven by policies in various countries. The design and construction stages of Building Information Modeling (BIM) have been universally accepted [5, 6]. BIM establishes the 3D model of a virtual building project in the computer and employs digital technology to provide this model with a complete information library for the building project consistent with the actual situation. Rich in building engineering information, this 3D model can significantly improve the degree of information integration, thereby providing a platform for engineering information exchange and sharing for the stakeholders of the building engineering project [7]. Architectural multimedia is the dynamic performance of applying multimedia technology to architecture. It is a new digital product widely used in bidding, reporting, and display in construction engineering industry in recent years. Based on BIM technology, architectural multimedia gives full play to the technical advantages of multimedia, and integrates the illustration of plane bidding documents, the authenticity of effect drawings, the guidance of drawings, and the appreciation of video and audio. The comprehensive use of two-dimensional and three-dimensional animation, dynamic demonstration of images and words, special effect editing of video materials, video and audio production technology, and interpretation by professional broadcasters will further guide users with immersive, real, and intuitive attraction and persuasion. BIM multimedia is based on 3D modeling and allows virtual construction of engineering projects before construction, which eliminates many inefficient problems in the construction process and helps to eliminate potential design and construction risks.

As the copy of digital models or physical assets of products, institutions, public infrastructure, and even cities, Digital Twins (DTs) have a broad application scope regarding enhanced service design and urban management [810]. In smart cities, DTs gather information from the building environment continuously via sensors, drones, or mobile devices. Data are collected from the real world using remote communication technology, the Internet of Things (IoT), and sensors to construct the digital copy of the city. Besides IoT, technologies such as Big Data (BD), Artificial Intelligence (AI), cloud computing, and machine learning can also improve the accuracy and dynamics of digital copies, enabling these copies to process and aggregate the historical and the real-time data [11]. In road traffic, the DTs technology can not only realize the virtual mapping of physical entities, but also realize the dynamic monitoring of the life cycle of road infrastructure and the accurate restoration of traffic participants on the road by using various sensors and network communication technologies. Additionally, it can judge and predict the possible traffic events and accident risks according to traffic behavior and analyze the road traffic condition according to the traffic state, to provide an accurate basis for road traffic diagnosis and traffic management decision-making.

The collected data needs to be processed effectively to construct smart city DTs. However, the massive, complex, and changing urban data have brought serious obstacles to data mining. Simultaneously, the data size of Bayesian Networks (BNs) in various applications has increased dramatically. However, BNs trained by traditional approaches cannot meet the increasingly higher accuracy requirements proposed by actual application scenes. Efficient transportation system plays an important supporting role in epidemic prevention and control. Intelligent transportation has formed a relatively mature application system in China. Using information technology and algorithm design, they built an intelligent traffic management model, and realized the seamless connection between patients and medical resources and cross regional coordinated utilization of medical resources through scientific and efficient traffic organization and management.

At present, DTs have gradually penetrated the fields of building, medical treatment, and comprehensive urban governance. BIM can design the keel, structure, and wind-hydro-electricity layouts of the building to optimize various resources and configurations and rehearse the emergency plans. Integrating DTs and BIM to build an efficient smart city is a hot topic worldwide. The study expects to speed up the construction of Smart City and improve the processing accuracy of urban information data. In this process, accurate processing of complex data is a key link. Section 2 summarizes the research status of related technologies. Section 3 firstly describes the Smart City structure based on DTs and secondly combs the methods of establishing BIM three-dimensional information system. On this basis, according to the characteristics of BIM BD, an effective data learning algorithm is proposed. In addition, Bayesian Network Structural Learning (BNSL) is introduced to characterize the correlation between tags, thus improving the classification accuracy of complex data. The research results will help to promote the construction and realization of Smart City supported by BIM. Moreover, the exploration focuses on the breadth, complexity, and uncertainty of BD, and applies BNSL innovatively to the processing of BD, which lays a very solid foundation for the construction and popularization of Smart City DTs.

Skip 2RECENT RELATED WORK Section

2 RECENT RELATED WORK

2.1 Characteristics of DTs in Smart Cities

Three-dimensional simulation and IoT access technologies can build a remote control and device linkage platform in the internet digital space. This platform can reproduce the city's building geographic structure and perceive the city's population heat map in real-time. It recognizes and analyzes images by automatically collecting data returned to the center, thereby predicting and deducing abnormalities and development of the city.

Deren et al. [12] summarized and analyzed several understandings of DTs at this stage from different dimensions. They defined the relationship between DTs and smart cities and analyzed the characteristics of smart cities based on DTs. They focused on the five major applications of smart cities based on DTs: urban operation monitoring, smart grid services, traffic management, public health services, and disaster life-cycle monitoring. Francisco et al. [13] enabled DTs in smart cities and addressed building energy management issues. They combined smart metering data with the new data stream obtained by the infrastructure to manage the smart energy in a larger building area. Fan et al. [14] proposed a smart city disaster DTs example, which could be brought together with AI algorithms to enable people to better understand the network dynamics of complex disaster management.

2.2 BIM+ Multimedia BD

Taking the analyses of all building models and building costs of free buildings in the past as the core, building BD keeps up with the latest building material price network system worldwide to make the quotation of the whole building simpler and more transparent. BIM has triggered the second digital revolution in the entire engineering construction field, and “BIM+BD” will surely become the third digital revolution in the engineering field. Demirdöğen et al. [15] put forward an integrated scheme based on BIM building energy-saving performance simulation and BD analysis. This scheme achieved building facility management as per lean management concepts. Greene et al. [16] employed Big Data Mining (BDM) to gather the knowledge about the performance of present building projects and regarded it as the basis for novel intelligent building design. This approach could guarantee efficient facility management during operation to ensure the sustainability of whole project stages, bringing smarter building design planning and more efficient facility management schemes.

“BIM+BD” will not only advance and upgrade existing technologies but also indirectly affect the production organization and management methods. The entire building design can be completed by one designer efficiently, which will affect people's thinking patterns in the long run.

2.3 BDM and Learning

With the increasingly wide application of green buildings and the modernization of the housing industry, and even the modernization of the whole construction industry, building cities in factories has become a reality, and the era of “construction industry 4.0” has come. Various data in smart cities are expanding at an unprecedented speed, which determines the future development progress of smart cities. Data mining uses intelligent optimization algorithms to extract valid information from massive datasets and transforms the extracted information into a structure that can be used further. Data mining adopts specific machine learning and statistical models to mine the potential meaning in the data to predict the future. Ateş et al. [17] suggested that data mining played a huge part in gathering information through necessary analysis and provided a new perspective for the decision-making process. Şahin and Yurdugül [18] applied BDM and learning to the education field. Information hidden in BD could be discovered through mining algorithms; then, data learning and analysis results could be utilized to optimize the education management system.

2.4 Research Review

The above research status reflects that most of the data of Smart City are collected, processed and analyzed through traditional data mining algorithms. However, the data in the Smart City is highly uncertain, so more advanced technologies are needed to process the data. At present, Bayesian network has become the mainstream technology to study uncertain data. It is considered to be one of the most useful theoretical models to describe and infer uncertain knowledge, but it has not been applied to data analysis in Smart City scenarios. The work will explore new data processing methods based on BNSL to effectively analyze Smart City data and accelerate the development of Smart City.

Skip 3SMART CITY DATA ANALYSIS BASED ON BNSL Section

3 SMART CITY DATA ANALYSIS BASED ON BNSL

3.1 Smart City Transportation System under COVID-19 Scenario

Efficient intelligent transportation system (ITS) is the core of collaborative joint control system for major public health events. Due to the wide range of information points, high timeliness, and strong comprehensiveness required in the process of transportation support, comprehensive transportation operation coordination and construction of emergency command center have become the necessity of transportation support. The information technology and algorithm design are used to build a three-dimensional GIS simulation management model of intelligent transportation and realize the seamless connection between traffic simulation patients and medical resources and cross-regional coordinated utilization of medical resources through scientific and efficient traffic organization and management. The traffic monitoring and auxiliary decision-making system can be used to view the vehicle ranking length and estimated transit time of each quarantine station in real time. Moreover, it can broadcast the serious congestion in real time, and make hourly automatic statistics, thereby assisting the decision-maker in deploying multi-point diversion and dredging at the back end for serious congestion. In the meantime, the system will be linked with the epidemic situation registration system of Chengdu entry channel to divert and regulate the traffic pressure of quarantine stations, to shorten the traffic time of vehicles.

If the epidemic is serious locally, cross-regional medical resources need to be applied cooperatively. First, consider the regional medical resources, epidemic tension, and transportation distance between regions. Transportation efficiency is the key factor, and transportation time is the key index of transportation efficiency. The transportation mode should be selected according to the transportation time, and the expressway, high-speed railway, or air transportation between cities can be considered. If the transportation time is controlled at 1 h, the coverage radius of urban fast track should be controlled at 50∼60 km, the coverage of expressway, high-speed subway, and air transportation can be controlled at about 80 km, 150 km, and 200 km, respectively, to build a three-dimensional comprehensive traffic prevention and control system [19]. Meanwhile, considering the transportation cost, if the number of patients transported is large and the air transportation cost is high, a temporary rescue site can be built in an area with good traffic conditions to maximize the service of the organization. According to the transportation efficiency of ITS, formulate a regional coordination scheme, improve the utilization efficiency of medical facilities, avoid the shortage of medical facilities, and reasonably control the rescue cost. The rapid and accurate information system can accurately judge the trajectory and influence a range of key populations, determine the distribution of affected personnel, quickly control the influence range, realize the timely investigation of an epidemic situation, quickly curb the spread of the epidemic situation, and avoid the rapid spread of the epidemic situation in new areas.

3.2 DTs-based Smart City Construction in the Industrial 4.0 Era

The main feature of the fourth industrial revolution is to make comprehensive use of the “physical system” created by the first and second industrial revolutions and the increasingly complete “information system” brought by the third industrial revolution to realize intelligent production through the integration of the two. The DTs of a city is a virtual entity that maps one-to-one with a physical city and can be manipulated intelligently. It is a complex system integrating technology, business, and data. Its technical system framework includes necessary components such as smart terminals, ubiquitous networks, support platforms, and application systems, and runs through the entire process of urban data collection, aggregation, integration, analysis, and application [20]. Based on DTs, integrating IoT, cloud computing, BD, AI, and other new-generation information technologies can guide and optimize the planning and management of physical cities, improve the services for citizens, and help the smart city construction. The operating brain of smart cities can be constructed based on DTs [2123], as shown in Figure 1. The smart city operating brain can adequately integrate data in the cloud centers with the digital subsystems of various departments to build a multi-port integrated urban command and emergency center. Fundamental data analysis components, such as multi-dimensional analysis and data mining, can minimize government costs and eventually improve urban efficiency.

Fig. 1.

Fig. 1. Structure of smart city operating brain based on DTs.

The public information cloud service platform is a vital link to the infrastructure in the smart city operating brain; its structure is illustrated in Figure 2. This platform employs infrastructure such as servers, networks, and sensor devices to obtain data. Finally, it can provide cloud services for various fields such as smart city management, public security, and tourism. This platform serves as the resource library of the city BD, which can command and monitor the urban operation and efficiently coordinate various departments in different districts.

Fig. 2.

Fig. 2. Structure of public information cloud service platform for smart cities based on DTs.

A smart city integrates the real world and the digital world established based on DTs, IoT, and cloud computing technologies, perceiving, controlling, and providing intelligent services for people and things [24, 25]. Smart city DTs have broad prospects in economic transformation, urban smart management, and public smart services, enabling a coordinated development of humans and nature. Smart cities require a complete spatial information infrastructure to better utilize various applications in them.

3.3 Construction of Smart City BIM 3D Multimedia Information System

Technological innovation is necessary for promoting the digital service industry and better applying “internet + smart city.” In smart city construction, BIM promotes collaboration and information communication management between teams participating in construction projects, which is one of its advantages [2628]. BIM applications can be the product of digital design, the process of cooperative project implementation, and the tool to manage the building life cycle. BIM can be understood as a digital representation of the physical and functional characteristics of building facilities. It can serve as a shared knowledge resource for building facility information and a reliable basis for decision-making in the all life cycle of building facilities, as demonstrated in Figure 3.

Fig. 3.

Fig. 3. The all life cycle of building facility based on BIM.

The information of the model contains all the information of building construction, so it is very easy to get the statistics of quantities, economic and technical indicators and other data, and it is also very easy to make indoor and outdoor building roaming animation. If the path and the direction of the camera are set, the animation can be completed, which truly realizes the three-dimensional design of the architect's brain – three-dimensional architectural information model – real architecture. Architectural visualization allows the architect to virtually build his own design works in advance, and timely correct unforeseen errors in advance. These data files are based on the implementation of the BIM model. The intelligent structural status monitoring system in the BIM 3D information system is the prototype of smart cities in the architecture field [29]. Smart cities also require a relatively complete urban infrastructure, especially a complete design of urban network infrastructure and the ability to carry BD. At the design stage, data in BIM can be extracted through programming for collision inspection, illumination analysis, and emergency evacuation analysis. The advantages of Unity 3D for BIM design are programmability, compatibility, extensibility, and easy operation. After the model is loaded, designers can roam the model, check the problems in the design, and get feedback on the problems interactively. Unity 3D can simulate the construction based on BIM data to determine a reasonable construction guide [3032]. At the construction stage, the construction staff can interact with the designers in the same model via the internet platform developed by Unity 3D, reducing the communication cost and the probability of secondary errors while solving the problems during the construction. Using Unity 3D to display the schemes not only saves a lot of rendering time but also provides interaction. It sets the schemes to a first-person or third-person perspective for roaming; meanwhile, multiple scenes can be provided for interaction as needed.

The visual human-computer interaction (HCI) method of 3D IoT based on BIM and unity 3D includes the following steps: (1) Import the construction CAD drawing into Autodesk Revit to implement the BIM model; (2) Package and export the implemented BIM model with Autodesk Revit in fbx. format, open Unity 3D software, and import the packaged fbx. format file into Unity 3D; (3) After BIM import, add identification objects at the parts that need to interact, add scripts, and bind them to the identification objects; (4) Based on Unity 3D platform, optimize the imported model, adjust the proportion, make the model in the center of the interface, and design the UI interface based on different interactive functions; (5) Add camera components in Unity 3D, bind scripts on the camera, and manipulate the camera to rotate 360°; and (6) The import of IoT data uses MySQL database. Further, HCI includes viewing some data, the 3D model construction, and the real-time image data at the specified position of the 3D model, planning the 3D travel route, etc.

Using a BIM 3D information system to implement the smart city provides a planning scheme at the technical level, as displayed in Figure 4. This scheme fully taps the potential of the BIM 3D information system, combines the advantages of wireless transmission and Local Area Network (LAN) for data transmission, and finally, stores and analyzes the data in the storage center and the analysis center, respectively. Abnormal results will be displayed on the client in real-time.

Fig. 4.

Fig. 4. Structure of public information cloud service platform for smart cities based on DTs.

3.4 Efficient Learning Algorithm for Complex BD

Actual smart city data are diverse in types, including symbolic, numeric, set-value, and missing data [33, 34]. The rough set can obtain the target information without using prior knowledge during data modeling and rule extraction. However, it cannot well fuse the high-dimensional, massive, and complex data [35]. Like other modeling methods, it requires much time for data fusion, but the fusion outcome is not good. Hence, the relation is introduced to propose a complex data fusion and efficient learning algorithm based on the compositive rough set model. An information system containing two or more different types of attributes is called a compositive information system \( GIS = ( {U{\rm{ }}A{\rm{ }}V{\rm{ }}f} ) \), where \( U \) represents a set of non-empty finite objects, \( A = \cup {A_k} \) refers to the union of non-empty attribute sets (\( {A_k} \) is the attribute set of the same data type), \( V = { \cup _{{A_k} \subseteq A}}{V_{{A_k}}}{V_\alpha } \), \( {V_\alpha } \) denotes the value range of attribute \( \alpha \), and \( f \) is the information function. A decision information system simultaneously comprising condition attribute \( C \) and decision attribute \( D \), namely \( A = C \cup D \), is referred to as the compositive decision information system, recorded as \( GIS = ( {U{\rm{ }}C \cup D{\rm{ }}V{\rm{ }}f} ) \). Suppose a given compositive decision information system \( GIS = ( {U{\rm{ }}C \cup D{\rm{ }}V{\rm{ }}f} ) \), \( B \subseteq C \). In that case, \( \Gamma ( B ) = {B_1},{B_2},\ldots ,{B_K} \) is the attribute partition on \( B \). The compositive relation \( {C_B} \) can be defined as: (1) \( \begin{equation} {C_B} = \mathop \cap \limits_{{B_k} = \Gamma \left( B \right)} {R_{{B_k}}}\mathop = \limits^\Delta {R_{{B_1}}} \cap {R_{{B_2}}} \cap ... \cap {R_{{B_K}}} \end{equation} \) (2) \( \begin{equation} {C_B} = \left( {x{\rm{ }}y} \right)U \times U\left( {x{\rm{ }}y} \right) \in \mathop \cap \limits_{{B_k} = \Gamma \left( B \right)} {R_{{B_k}}} \end{equation} \)

In (1) and (2), \( \forall {B_k} \in \Gamma ( B ) \), and \( {R_{{B_k}}} \in U \times U \) describes the indiscernibility relation about \( {B_k} \) on \( U \).

Suppose \( GIS = ( {U{\rm{ }}C \cup D{\rm{ }}V{\rm{ }}f} ) \), \( \forall X \subseteq U \); in that case, the lower and the upper approximation sets of \( X \) about the compositive relation \( {C_B} \) are, respectively, defined as: (3) \( \begin{equation} \underline {{C_B}} = x \in U{C_B}\left( x \right) \subseteq X \end{equation} \) (4) \( \begin{equation} \overline {{C_B}} = x \in U{C_B}\left( x \right) \cap X \ne \emptyset \end{equation} \)

\( U/D = {D_1}{D_2} \) denotes a partition. The lower and the upper approximation sets of decision \( D \) about \( B \) are respectively defined as: (5) \( \begin{equation} \underline {{C_B}} \left( D \right) = \underline {{C_B}} \left( {{D_1}} \right)\underline {{C_B}} \left( {{D_{12}}} \right)...\underline {{C_B}} \left( {{D_m}} \right) \end{equation} \) (6) \( \begin{equation} \overline {{C_B}} \left( D \right) = \overline {{C_B}} \left( {{D_1}} \right)\overline {{C_B}} \left( {{D_{12}}} \right)...\overline {{C_B}} \left( {{D_m}} \right) \end{equation} \)

In (5) and (6), \( \forall j \in 1,2,...m \). (7) \( \begin{equation} \underline {{C_B}} ( {{D_j}}) = x \in U{C_B}\left( x \right) \subseteq {D_j} \end{equation} \) (8) \( \begin{equation} \overline {{C_B}} ( {{D_j}}) = x \in U{C_B}\left( X \right) \cap {D_j} \ne \emptyset \end{equation} \)

The positive region \( PO{S_{{C_B}}}( D ) \) can be defined as: (9) \( \begin{equation} PO{S_{{C_B}}}\left( D \right) = \mathop \cup \limits_{j = 1}^m \underline {{C_B}} ( {{D_j}}) \end{equation} \)

Table 1 presents a compositive information system \( GIS = ( {U A V{\rm{ }}f} ) \). Based on the attribute type, the attribute partition \( \Gamma ( B ) = {B_1},{B_2},{B_3},{B_4} \) on \( B \) can be obtained, where \( {B_1} = {a_1} \), \( {B_2} = {a_2} \), \( {B_3} = {a_3}{a_4} \), and \( {B_4} = {a_5} \). Suppose that \( {R_{{B_1}}} \), \( {R_{{B_2}}} \), \( {R_{{B_3}}} \), and \( {R_{{B_4}}} \) are respectively the equivalence relation R, neighborhood relation N, compatibility relation T, and characteristic relation K about attribute sets \( {B_1} \), \( {B_2} \), \( {B_3} \), and \( {B_4} \), and the neighborhood is \( \delta = 0.15 \). In that case, the compositive relation presented in Table 1 can be constructed.

Table 1.
Ua1a2a3a4a5D
x1y1,20,20,1*
x2y10,20,3?×
x3y00,10,1Small
x4y0,1,20,10,2Small
x5N10,10,3Large
x6n0,20,20,2Large×
xi\( {R_B}_1( {{x_i}} ) = {R_B}_1( {{x_i}} ) \)\( {R_B}_2( {{x_i}} ) = {\Gamma _B}_2( {{x_i}} ) \)
x1x1, x2, x3, x4x1, x2, x4, x5, x6
x2x1, x2, x3, x4x1, x2, x4, x5
x3x1, x2, x3, x4x3, x4, x6
x4x1, x2, x3, x4x1, x2, x3, x4, x5, x6
x5x5, x6x1, x2, x4, x5
x6x5, x6x1, x3, x4, x6
xi\( {R_B}_3( {{x_i}} ) = {N_B}_3( {{x_i}} ) \)\( {R_B}_4( {{x_i}} ) = {K_B}_4( {{x_i}} ) \)
x1x1, x3, x4, x6U
x2x2, x4, x5, x6U
x3x1, x3, x4, x6x1, x3, x4
x4x1, x2, x3, x4, x5, x6x1, x3, x4,
x5x2, x4, x5, x6x1, x5, x6
x6x1, x2, x3, x4, x5, x6x1, x5, x6

Table 1. Compositive Decision Information System and Relation Structure

Data of smart cities usually occupy a large storage space. Thus, a batch processing algorithm is designed to calculate the approximate set of the compositive rough set. For any k, the sub-matrices of the relation matrix are constructed first; then, the sub-matrices of the lower and the upper approximate set matrices are calculated. According to Amdahl's law, the Graphics Processing Unit (GPU) is employed to accelerate the construction of the relation matrix and the calculation of the upper and the approximate set matrices. The decision matrix is constructed using the serial algorithm.

The relation matrix \( {R_B} = {( {{r_{ij}}} )_{n \times n}} \) can be constructed using \( {R_B} = U_B^T \circ {U_B} \). From a coarse-grained perspective, the upper approximation set matrix \( \overline {{C_B}} ( D ) \) depends on the relation matrix \( {R_B} \); from a fine-grained perspective, elements of the upper approximation set matrix do not depend on the elements in all relation matrices. The equation to calculate \( {u_{ij}} \) of the upper approximation set matrix can also be written as: (10) \( \begin{equation} {u_{ij}} = \mathop \vee \limits_{k = 1}^n ( {{r_{ik}} \wedge {d_{kj}}}) = \mathop \vee \limits_{k = 1}^n \left( {\left( {{x_i} \circ {x_k}} \right) \wedge {d_{kj}}} \right) \end{equation} \)

The serial algorithm shall be executed \( T \times n \) times; with sufficient resources, the parallel algorithm only needs to be executed once. Similarly, all \( T \times m \) elements in the sub-matrix \( \overline {{C_B}} ( D )[ {1:T} ] \) of the upper approximation set matrix can be calculated in parallel.

Since the decision matrix and the data matrix are transmitted to the GPU side, the upper and the lower approximation set matrices are opposite, which are sent back to the Central Processing Unit (CPU) side from the GPU side [36, 37]. The asynchronous execution indicates that the transmission can be completed at the same time. The decision matrix and the data matrix are transmitted to all GPUs, and the sub-matrices of the upper and the lower approximation set matrices are transmitted back to the CPU side from different GPU sides.

The basic unit of GPU processing is thread, which then forms thread block and thread grid. Each thread has a dedicated register and local memory. Each block has shared memory, and its internal threads can be accessed. The running thread can access the global memory (GM) of the device. The model of GPU is AMD FirePro™ V5900, a professional graphics card for graphics processing, which has 2GB video memory and 512 stream processors. At the software level, the compilation and development software used by the computing platform is Visual Studio and CodeXL, and the programming language is OpenCL.

After CADU4.0, multiple GPUs can be easily operated by stream processing. Specifically, in the proposed Multi-GPU based algorithm, independent streams are adopted to manage different GPUs. Given a group of GPU devices G, it is assumed that the programming starts from Device 0 and ends at Device G-1. Map the global “Start” tag and size to each Device. Then, the decision matrix and data matrix are transmitted from the CPU to all GPU. After the calculation, the sub-matrices of the upper approximation set matrix and the lower approximation set matrix on each device will be transmitted back to the CPU, and the process is also executed asynchronously. Finally, the CPU side (Host) combines all these sub-matrices into an upper approximation set matrix and a lower approximation set matrix (i.e., the final result).

Multiple Tesla GPU devices and two datasets from the University of California Irvine (UCI) database, namely P53-old and P53-new, are utilized to validate the effectiveness of the proposed Multi-GPU algorithm. The two datasets contain 16,772 and 31,420 samples, respectively. Attributes of the datasets are all numerical. The datasets are still expanding. These datasets are mainly divided into binary classification problems, multi classification problems, and regression fitting problems. UCI dataset provides the main attributes of each dataset. The experimental results can be demonstrated on its dataset according to various algorithms proposed to prove the rationality of the proposed algorithm. The 1-norm, 2-norm, and infinite norm are chosen; the neighborhood radius \( \delta \) takes 0.3, 0.4, 0.5, 0.6, 0.7, and 0.8, respectively. The distance between the two attributes is calculated first; then, the relation matrix can be constructed. In the experiment, the distance between two attributes is calculated by traversing all the attributes. A judgment criterion is set to speed up this process: if the current distance is greater than \( \delta \), the traversal will be stopped. This makes only one thread running while other threads have completed, and the entire Warp is still active.

3.5 Multi-Label Classification based on BNSL

BIM establishes the 3D model of a virtual building project in the computer. In the meantime, it employs the digital technology to provide this model with a complete information library for the building project consistent with the actual situation. This information library contains not only geometric information, professional attributes, and status information describing building objects but also status information of non-component objects. Rich in building engineering information, this 3D model can significantly improve the degree of information integration, thereby providing a platform for engineering information exchange and sharing for the stakeholders of the building engineering project. Combining with other digital technologies, users can simulate the state and changes of the building in the real world using engineering information in BIM. Hence, the stakeholders can comprehensively analyze and evaluate the success or failure of the entire engineering project before the construction is completed. In BIM BD, multi-label classification is more practical; nonetheless, it will make the classification more complicated because of the complexity caused by the increased label categories and the correlation between labels [3840]. In this case, BN is adopted for multi-label classification, and the algorithm is recorded as MLBN. Each label is considered a BN node; then, structural learning is applied to learn the structure of the label BN from the data.

The Bayesian theorem connects the prior probability and the posterior probability of an event. Suppose that the joint probability distribution density of random variables \( x \) and \( \theta \) is \( p( {x,\theta } ) \); in that case. their edge densities are \( p( x ) \) and \( p( \theta ) \), respectively. If \( x \) is the observation vector and \( \theta \) is the unknown parameter vector, the estimation of the unknown parameter vector will be obtained through the observation vector. The Bayesian theorem can be expressed as: (11) \( \begin{equation} p\left( {\theta |x} \right) = \frac{{\pi \left( \theta \right)p\left( {x|\theta } \right)}}{{p\left( x \right)}} = \frac{{\pi \left( \theta \right)p\left( {x|\theta } \right)}}{{\int{{\pi \left( \theta \right)p\left( {x|\theta } \right)}}}}d\theta \end{equation} \)

In (11), \( \pi ( \theta ) \) refers to the prior distribution of \( \theta \). Traditional parameter estimation models only obtain information from sample data, while BN's estimation of unknown parameter vectors combines their prior information and sample information.

Mutual Information (MI) is an important knowledge of information theory. The entropy \( H( X ) \) of variable X can be expressed as: (12) \( \begin{equation} H\left( X \right) = - \sum\limits_{i = 1}^n {P\left( {{x_i}} \right)} \log \left( {{x_i}} \right) \end{equation} \)

Entropy measures the degree of uncertainty of random variables; the greater the entropy value, the greater the uncertainty.

The joint entropy \( H( {X,Y} ) \) for two random variables X and Y can be expressed as: (13) \( \begin{equation} H\left( {X,Y} \right) = - \sum\limits_{i = 1}^n {\sum\limits_{j = 1}^m {P( {{x_i},{y_j}})} \log ( {{x_i},{y_j}})} \end{equation} \)

The scoring function evaluates the posterior probability based on the known prior probability distribution knowledge. The posterior probability is the probability of the network structure G conditional on the data D, expressed as: (14) \( \begin{equation} P\left( {G|D} \right) = \frac{{P\left( {GD} \right)}}{{P\left( D \right)}} = \frac{{P\left( G \right)P\left( {D|G} \right)}}{{P\left( D \right)}} \end{equation} \)

The logarithm of Equation (14) is taken for the convenience of calculation, and the following equation can be obtained: (15) \( \begin{equation} \log P\left( {G,D} \right) = \log \left( {P\left( G \right)P\left( {D|G} \right)} \right) \end{equation} \)

Scoring equations of BN are: (16) \( \begin{align} &{g_{CH}}\left( {G,D} \right) = \log P\left( G \right) \nonumber\\ &\quad +\sum\limits_{i = 1}^n {\sum\limits_{j = 1}^{{q_i}} {\left[ {\log \left( {\frac{{\left( {r_i} - 1 \right)!}}{{\left( {N_{ij}} + {r_i} - 1 \right)!}}} \right) + \sum\limits_{k = 1}^{{r_i}} {\log ( {{N_{ijk}}!})} } \right]} } \end{align} \) (17) \( \begin{equation} g\left( {{X_i},fa\left( {{X_i}} \right)} \right) = \prod\limits_{j = 1}^{{q_i}} {\frac{{\left( {{r_i} - 1} \right)!}}{{( {{N_{ij}} + {r_i} - 1})!}}} \prod\limits_{k = 1}^{{r_i}} {{N_{ijk}}!} \end{equation} \)

In (16) and (17), \( {r_i} \) and \( {q_i} \) represent the number of states that the i-th variable \( {X_i} \) and its parent node \( fa( {{X_i}} ) \) can take, respectively, and \( {N_{ijk}} \) signifies the sample size in the data when the state of the variable \( {X_i} \) is k and the state of \( fa( {{X_i}} ) \) is j.

Based on the K2-CH score, a sequence with a larger score value (chain structure) is searched in the space formed by variables. The better sequence is input into the K2 algorithm for BNSL. This staged learning reduces the computational cost. (18) \( \begin{equation} {f_{Chain - K}} = K2 - C{H_{Ordering}} \end{equation} \) (19) \( \begin{equation} {f_{Chain - K}} = \sum\limits_{i = 2}^n {\log \left( {\left( {\prod\limits_{j = 1}^{{q_i}} {\frac{{\left( {{r_i} - 1} \right)!}}{{\left( {{N_{ij}} + {r_i} - 1} \right)!}}\prod\limits_{k = 1}^{{r_i}} {{N_{ijk}}!} } } \right)} \right)} \end{equation} \)

Since MI represents the correlation between variables, the greater the MI value, the stronger the correlation. A chain structure is constructed by taking \( {f_{Chain - M}} \) as the objective function, and the following equations can be obtained: (20) \( \begin{equation} {f_{Chain - M}} = M{I_{Ordering}} \end{equation} \) (21) \( \begin{equation} {f_{Chain - M}} = \sum\limits_{i = 2}^n {\log \left( {MI\left( {{X_i},fa\left( {{X_i}} \right)} \right)} \right)} \end{equation} \)

The empirical Bayesian estimation combines BN with the classical method. After the sample's edge density is obtained, the prior distribution can be determined according to the following equation: (22) \( \begin{equation} p\left( x \right) = \int_{{ - \infty }}^{{ + \infty }}{{\pi \left( \theta \right)}}p\left( {x|\theta } \right)d\theta \end{equation} \)

Each label in the label set \( L = \{ {{l_1},{l_2},\ldots ,{l_n}} \} \) is treated as a BN node. The structural learning algorithm is practiced to learn the corresponding network structure. After the label's network structure is learned, the parameter learning method is adopted to learn the Conditional Probability Table (CPT) of the label node. Given the training data \( D \) and the model structure \( G \), the maximum likelihood estimation selects the optimal parameters through the model parameter \( \Theta G \) and the maximum likelihood of the training data. Suppose that node \( {V_i} \)have \( {r_i} \) values in total; in that case, the father node set \( fa( {{V_i}} ) \) total \( {q_i} \) value combinations. The parameter for node \( {V_i} \) in BN is: (23) \( \begin{equation} {\theta _{ijk}} = P\left( {{V_i} = k|fa\left( {{V_i}} \right) = j} \right) \end{equation} \)

\( \Theta G \) denotes the vector represented by all \( {\theta _{ijk}} \) in the network. The likelihood degree of a given \( \Theta \) is the conditional probability \( P( {D,G|\Theta } ) \), which can be described as: (24) \( \begin{equation} L\left( {\Theta |D,G} \right) = P\left( {D,G|\Theta } \right) \end{equation} \)

\( {\Theta ^*} \) is the value that maximizes the likelihood and is also the optimal network parameter, expressed as: (25) \( \begin{equation} {\Theta ^*} = \mathop {\arg \max }\limits_\Theta L\left( {\Theta |D,G} \right) \end{equation} \)

Suppose that the sample \( D = ( {{D_1},{D_2}, \ldots ,{D_m}} ) \) is a complete dataset with m independent and identical distributions; in that case, the log-likelihood function can be denoted as: (26) \( \begin{equation} l\left( {\Theta |D,G} \right) = \log \left( {\Theta |D,G} \right) = \sum\limits_{s = 1}^m {\log P\left( {{D_s},G|\Theta } \right)} \end{equation} \)

To maximize Equation (28), the value of \( {\theta _{ijk}} \) shall be: (27) \( \begin{equation} {\theta _{ijk}} = \left\{ \begin{array}{@{}l@{}} \frac{{{m_{ijk}}}}{{{\theta _{ij}}}},{m_{ij}} > 0\\[4pt] \frac{1}{{{r_i}}}{\rm{ }},{m_{ij}} > 0 \end{array} \right. \end{equation} \)

Incorporating the idea of K neighborhood into MLBN can get the possible label \( \mathop Y\limits^ \wedge ( T ) \) of the test set. The deterministic information \( \mathop {{y_1}}\limits^ \wedge ( T ) = 1 \) and \( \mathop {{y_4}}\limits^ \wedge ( T ) = 0 \) in \( \mathop Y\limits^ \wedge ( T ) \) is entered into label BN as evidence.

The deterministic label information \( \mathop E\limits^ \wedge ( T ) \) in \( \mathop Y\limits^ \wedge ( T ) \) is entered into the BN structure corresponding to the label as evidence infer the most likely label class \( {Y^*}( T ) \). In BN, the global label can be predicted by inputting evidence, which is expressed as: (28) \( \begin{equation} {Y^*}\left( T \right) = \mathop {\arg \max }\limits_Y P( {Y = y|\mathop E\limits^ \wedge \left( T \right) = \mathop e\limits^ \wedge \left( T \right)}) \end{equation} \)

Excessive nodes may complicate BN reasoning. The Joint Tree (JT) reasoning algorithm can integrate evidence label information and improve the reasoning efficiency. Figure 5 demonstrates the algorithm flow of the proposed MLBN.

Fig. 5.

Fig. 5. Algorithm flow of MLBN.

The time complexity of MLBN is mainly divided into two parts. First, for each object, the worst time complexity is \( O( {\log m} ) \); Therefore, the complexity of scanning all data at once is \( O( {n\log m} ) \). Second, the mapping and division into decision matrix can be completed by scanning the constructed tree once, and the complexity is \( O( n ) \). Therefore, the complexity of the construction is \( O( {{n^2} \times B} ) \). When using depth first search to correct the structure diagram, it is generally necessary to explore the general elements in the structure diagram matrix. In this way, for the structure with node number of \( n \), the time complexity of structure correction is \( O( {{n^2}} ) \).

Six different multi-label classification datasets from the Mulan open-source database (The number of attributes is 1,449 and the number of tags is 45) are adopted to evaluate the performance of MLBN: emotions, image, yeast, genbase, recreation, and education. The proposed structural learning algorithm is practiced to learn the BN structure. Accuracy and F1-measure are indicators to evaluate the performance of MLBN. Firstly, Bayesian network is used to construct the corresponding Bayesian network structure of tags, and parameter learning is used to learn the conditional probability distribution table between nodes (tags). For each test set, find its corresponding K neighborhoods, and then input them into the Bayesian network as evidence according to the same label category in its neighborhood, and obtain the maximum possible state distribution of all labels through reasoning algorithm. The value of K in MLBN can affect its performance. In this experiment, the K value is determined to be between 1 and 10, and the algorithm's performance is tested under different K values, respectively.

In addition, the accuracy and F1 value of MLBN algorithm, multi-label classification k-nearest neighbor (ML-KNN), and ML-RBF algorithm are compared on different data sets. ML-KNN is an algorithm for solving multi-label classification problems by combining k-nearest neighbor algorithm with Bayesian theorem. Its basic idea can be summarized as follows: statistics label category information contained in k-nearest neighbor samples of test samples, and use the maximum a posteriori probability (MAP) criterion to predict the category of samples to be classified. The neurons (basis functions) of the radial basis function (RBF) hidden layer are associated with the prototype vector in the first layer, and each output neuron corresponds to a possible class. RBF neural network is usually trained in two stages, in which the basic function is learned by cluster analysis of training examples, and the second layer weight is optimized by solving linear problems. The input of ML-RBF neural network corresponds to d-dimensional feature vector. The parameter selection in ML-RBF is ratio = 0.01, mu = 1.

Skip 4RESULTS AND DISCUSSIONS Section

4 RESULTS AND DISCUSSIONS

4.1 Effectiveness of Multi-GPU Algorithm in Complex BD Processing

Figures 6 and 7 illustrate the results of the average running time and speedup ratio of the Multi-GPU algorithm on different datasets. As the value of \( \delta \) increases, the running time of the Multi-GPU algorithm gradually increases in all three metrics. On these two datasets, the running time of the Multi-GPU algorithm decreases as the number of GPUs increases. In particular, the ideal linear speedup ratio can be obtained in the infinite norm experiment. The above results prove the effectiveness of the proposed Multi-GPU algorithm. A GPU based algorithm is proposed for parallel computing approximate sets to deal with massive and high-dimensional data. Furthermore, an algorithm is given for parallel computing approximate sets based on Multi-GPU, which achieves excellent speedup in the test of this section. Compared with the approximate set calculation algorithm based on GPU, the performance of the approximate set calculation algorithm based on Multi-GPU has reached a higher level, and the acceleration ratio is up to 335 times.

Fig. 6.

Fig. 6. Running time and speedup ratio on the P53-old dataset.

Fig. 7.

Fig. 7. Running time and speedup ratio on the P53-new dataset.

4.2 Multi-Label Data Classification Performance of MLBN

Figure 8 presents changes in the evaluation indicators of the MLBN algorithm under six different classification datasets. When K = 1, MLBN provides abysmal classification performance because the selected neighborhood of the test set has only one sample at this time, so that the label information of each node input into the label BN is determined. However, the correlation between labels is not considered in this case. As the K value goes up continuously, the deterministic information input into the label BN gets reduced, lowering the classification accuracy. Overall, MLBN can provide the best performance when K = 3. Furthermore, the accuracy and F1 value of MLBN algorithm are compared with other algorithms in different data sets. The specific results are shown in Table 2.

Fig. 8.

Fig. 8. Classification performance of MLBN on six datasets.

Table 2.
Data setMLBNML-KNNML-RBF
AccuracyF1AccuracyF1AccuracyF1
genbase0.982 ± 0.0130.983 ± 0.0080.972 ± 0.0050.978 ± 0.0110.970 ± 0.0040.962 ± 0.008
emotions0.583 ± 0.0110.644 ± 0.0170.541 ± 0.0120.630 ± 0.0130.545 ± 0.0150.630 ± 0.013
yeast0.464 ± 0.0020.621 ± 0.0090.510 ± 0.0090.609 ± 0.0140.505 ± 0.0200.604 ± 0.018
image0.432 ± 0.0070.463 ± 0.0100.324 ± 0.0140.345 ± 0.0070.268 ± 0.0140.309 ± 0.013
education0.288 ± 0.0010.372 ± 0.0110.234 ± 0.0120.236 ± 0.0110.271 ± 0.0140.250 ± 0.013
recreation0.364 ± 0.0130.334 ± 0.0050.202 ± 0.0080.206 ± 0.0100.158 ± 0.0120.167 ± 0.020

Table 2. Accuracy and F1 Value of Different Algorithms in Different Data Sets

MLBN inputs the deterministic information in the neighborhood label as evidence into the label BN, thereby transforming the label estimation problem into the maximum likelihood interpretation in BN. Moreover, it adopts the JT algorithm to reduce the reasoning complexity. In MLBN, by constructing the Bayesian network model of tags, each tag is transformed into nodes, and the directed edge and node parameters are used to describe the correlation between tags. The labeled Bayesian network structure is learned from the data through the structure learning method proposed above, and then the maximum likelihood estimation method is used to learn the network parameters. In the process of label estimation of the test set, by combining the neighborhood samples of the test set, the deterministic information in the neighborhood label is input into the labeled Bayesian network as evidence. This operation can transform the label estimation problem into the maximum possible interpretation problem in the Bayesian network for solution, and the joint tree reasoning algorithm is used to reduce the reasoning complexity. The simulation results show that the proposed MLBN algorithm can effectively improve the performance of multi label classification.

Skip 5CONCLUSIONS Section

5 CONCLUSIONS

COVID-19 has its distinct characteristics compared with SARS the Avian Flu epidemic and a remarkable speed of transmission. Most of the achievements of smart city construction failed to respond quickly and play a key role in COVID-19's prevention and control and resumption of production. From IT era to DT era, BD, IoT and other technologies are gradually innovating the infrastructure of human social and economic activities, and will complete the in-depth information transformation of the original physical infrastructure. The arrival of the industry 4.0 era has undoubtedly become the most eye-catching wave in the current global business environment. The building industry has the greatest amount of data and the largest industrial scale. Thanks to DTs, urban services can be launched and provided faster than ever. The continuous development of smart cities will also expand the use scope of DTs. Seamless matching between digital models and physical devices can harvest the operating data of the device monitoring system in real-time for fault prediction and timely maintenance. As a complete information model, BIM can connect data, procedures, and resources at different stages of the life cycle of a building project, providing cities with corresponding digital models based on image scanning. It can comprehensively describe engineering objects and provide real-time engineering data that can be automatically calculated, consulted, fused, and split, which can be employed by all participants of construction projects. BIM is an inevitable product of the BD era. As the source code of the building industry, BIM can process the primary data at the project level; more importantly, its biggest advantage is the capacity to carry massive project data.

BIM BD of smart city DTs is investigated in the present work. The compositive relation is introduced for high-dimensional, massive data. A complex data fusion and efficient learning algorithm (Multi-GPU) is proposed based on the compositive rough set model. MLBN is put forward for complex data, which can learn the label Bayesian network's structure from data. The Multi-GPU algorithm is simulated on the P53-old and the P53-new dataset; basically, the ideal linear speedup ratio can be attained. Compared with the previous studies, the innovation is to expand the rough set model and propose a composite rough set model. The model can deal with multiple data types simultaneously, which provides a new method for complex data fusion. The MLBN algorithm is tested on different datasets to validate its classification performance. Results demonstrate that this algorithm can improve the performance of multi-label classification effectively. Based on structural learning, Bayesian networks in multi-label classification are considered, which is of practical significance for processing smart city BD. The work mainly focuses on the magnanimity, complexity, and uncertainty of BD, but there is no time for in-depth analysis in terms of multi-source heterogeneity. In addition to the characteristics of multi-source heterogeneity, large volume, and high complexity, BD in practice also faces the problems of scarce labeled data. In the next research, the research results of deep learning, multimodal learning and semi supervised learning will be combined to meet the needs of data representation in different BD applications.

REFERENCES

  1. [1] Sarkar M., Banerjee S., Badr Y., and Sangaiah A. K.. 2017. Configuring a trusted cloud service model for smart city exploration using hybrid intelligence. International Journal of Ambient Computing and Intelligence (IJACI) 8, 3 (2017), 121.Google ScholarGoogle ScholarCross RefCross Ref
  2. [2] Li J.. 2020. Resource optimization scheduling and allocation for hierarchical distributed cloud service system in smart city. Future Generation Computer Systems 107 (2020), 247256.Google ScholarGoogle ScholarDigital LibraryDigital Library
  3. [3] Jegadeesan S., Azees M., Kumar P. M., Manogaran G., Chilamkurti N., Varatharajan R., and Hsu C.. 2019. An efficient anonymous mutual authentication technique for providing secure communication in mobile cloud computing for smart city applications. Sustainable Cities and Society 49 (2019), 101522.Google ScholarGoogle ScholarCross RefCross Ref
  4. [4] Maftei S. S.. 2020. Smart city Cluj, from provincial hotspot to transnational hub: The adventures of a would-be post-industrial city in Romania. Journal of Cultural Management and Cultural Policy/Zeitschrift für Kulturmanagement und Kulturpolitik 6, 1 (2020), 143158.Google ScholarGoogle ScholarCross RefCross Ref
  5. [5] Yamamura S., Fan L., and Suzuki Y.. 2017. Assessment of urban energy performance through integration of BIM and GIS for smart city planning. Procedia Engineering 180 (2017), 14621472.Google ScholarGoogle ScholarCross RefCross Ref
  6. [6] El-Hallaq M. A., Alastal A. I., and Salha R. A.. 2019. Enhancing sustainable development through web based 3D smart city model using GIS and BIM. Case study: Sheikh Hamad City. Journal of Geographic Information System 11, 03 (2019), 321.Google ScholarGoogle ScholarCross RefCross Ref
  7. [7] Arcuri N., De Ruggiero M., Salvo F., and Zinno R.. 2020. Automated valuation methods through the cost approach in a BIM and GIS integration framework for smart city appraisals. Sustainability 12, 18 (2020), 7546.Google ScholarGoogle ScholarCross RefCross Ref
  8. [8] Ham Y. and Kim J.. 2020. Participatory sensing and digital twin city: Updating virtual city models for enhanced risk-informed decision-making. Journal of Management in Engineering 36, 3 (2020), 04020005.Google ScholarGoogle ScholarCross RefCross Ref
  9. [9] Fan C., Jiang Y., and Mostafavi A.. 2020. Social sensing in disaster city digital twin: Integrated textual–visual–geo framework for situational awareness during built environment disruptions. Journal of Management in Engineering 36, 3 (2020), 04020002.Google ScholarGoogle ScholarCross RefCross Ref
  10. [10] Liu M., Fang S., Dong H., and Xu C.. 2021. Review of digital twin about concepts, technologies, and industrial applications. Journal of Manufacturing Systems 58 (2021), 346361.Google ScholarGoogle Scholar
  11. [11] Syafrudin M., Alfian G., Fitriyani N., and Rhee J.. 2018. Performance analysis of IoT-based sensor, big data processing, and machine learning model for real-time monitoring system in automotive manufacturing. Sensors 18, 9 (2018), 2946.Google ScholarGoogle ScholarCross RefCross Ref
  12. [12] Deren L., Wenbo Y., and Zhenfeng S.. 2021. Smart city based on digital twins. Computational Urban Science 1, 1 (2021), 111.Google ScholarGoogle ScholarCross RefCross Ref
  13. [13] Francisco A., Mohammadi N., and Taylor J. E.. 2020. Smart city digital twin–enabled energy management: Toward real-time urban building energy benchmarking. Journal of Management in Engineering 36, 2 (2020), 04019045.Google ScholarGoogle ScholarCross RefCross Ref
  14. [14] Fan C., Zhang C., Yahja A., and Mostafavi A.. 2021. Disaster city digital twin: A vision for integrating artificial and human intelligence for disaster management. International Journal of Information Management 56 (2021), 102049.Google ScholarGoogle ScholarDigital LibraryDigital Library
  15. [15] Demirdöğen G., Işık Z., and Arayici Y.. 2020. Lean management framework for healthcare facilities integrating BIM, BEPS and Big Data Analytics. Sustainability 12, 17 (2020), 7061.Google ScholarGoogle Scholar
  16. [16] Greene M. C., Clarke-Hagan D., and Curran M.. 2020. Achieving smarter buildings and more efficient facilities management: The implementation of big data. International Journal of Digital Innovation in the Built Environment (IJDIBE) 9, 2 (2020), 116.Google ScholarGoogle ScholarCross RefCross Ref
  17. [17] Ateş E. C., Bostanci G. E., and Sedar M. S. G.. 2020. Big data, data mining, machine learning, and deep learning concepts in crime data. Journal of Penal Law and Criminology 8, 2 (2020), 293319.Google ScholarGoogle Scholar
  18. [18] Şahin M. and Yurdugül H.. 2020. Educational data mining and learning analytics: Past, present and future. Bartın Üniversitesi Eğitim Fakültesi Dergisi 9, 1 (2020), 121131.Google ScholarGoogle Scholar
  19. [19] Zheng S. and Kahn M. E.. 2013. China's bullet trains facilitate market integration and mitigate the cost of megacity growth. Proceedings of the National Academy of Sciences, 110, 14 (2013), E1248E1253.Google ScholarGoogle ScholarCross RefCross Ref
  20. [20] Sepasgozar S. M. E.. 2021. Differentiating digital twin from digital shadow: Elucidating a paradigm shift to expedite a smart, sustainable built environment. Buildings 11, 4 (2021), 151.Google ScholarGoogle Scholar
  21. [21] Sato M., Fukuyama Y., Iizaka T., and Matsui T.. 2019. Total optimization of energy networks in a smart city by multi-population global-best modified brain storm optimization with migration. Algorithms 12, 1 (2019), 15.Google ScholarGoogle Scholar
  22. [22] Nikitas A., Michalakopoulou K., Njoya E. T., and Karampatzakis D.. 2020. Artificial intelligence, transport and the smart city: Definitions and dimensions of a new mobility era. Sustainability 12, 7 (2020), 2789.Google ScholarGoogle ScholarCross RefCross Ref
  23. [23] Arai K., Fukuyama Y., Iizaka T., and Matsui T.. 2019. Optimal operational planning of energy plants considering uncertainty of renewable energy outputs by global-best brain storm optimization. IFAC-PapersOnLine 52, 4 (2019), 153158.Google ScholarGoogle ScholarCross RefCross Ref
  24. [24] Austin M., Delgoshaei P., Coelho M., and Heidarinejad M.. 2020. Architecting smart city digital twins: Combined semantic model and machine learning approach. Journal of Management in Engineering 36, 4 (2020), 04020026.Google ScholarGoogle ScholarCross RefCross Ref
  25. [25] Min Q., Lu Y., Liu Z., Su C., and Wang B.. 2019. Machine learning based digital twin framework for production optimization in petrochemical industry. International Journal of Information Management 49 (2019), 502519.Google ScholarGoogle ScholarDigital LibraryDigital Library
  26. [26] Liu Z., Chi Z., Osmani M., and Demian P.. 2021. Blockchain and building information management (BIM) for sustainable building development within the context of smart cities. Sustainability 13, 4 (2021), 2090.Google ScholarGoogle ScholarCross RefCross Ref
  27. [27] Alastal A. I., Salha R. A., and El-Hallaq M. A.. 2019. The Reality of Gaza Strip cities towards the smart city's concept. A case study: Khan Younis City. Current Urban Studies 7, 1 (2019), 143155.Google ScholarGoogle ScholarCross RefCross Ref
  28. [28] Tang S., Shelden D. R., Eastman C. M., Pishdad-Bozorgi P., and Gao X.. 2019. A review of building information modeling (BIM) and the Internet of Things (IoT) devices integration: Present status and future trends. Automation in Construction 101 (2019), 127139.Google ScholarGoogle ScholarCross RefCross Ref
  29. [29] Zhu J., Tan Y., Wang X., and Wu P.. 2021. BIM/GIS integration for web GIS-based bridge management. Annals of GIS 2021 27, 1 (2021), 99109.Google ScholarGoogle Scholar
  30. [30] Graham K., Pybus C., Arellano N., Doherty J., Chow L., Fai S., and Grunt T.. 2019. Defining geometry levels for optimizing BIM for VR: Insights from traditional architectural media. Technology| Architecture + Design 3, 2 (2019), 234244.Google ScholarGoogle Scholar
  31. [31] Ma Y. P.. 2021. Extending 3D-GIS district models and BIM-based building models into computer gaming environment for better workflow of cultural heritage conservation. Applied Sciences 11, 5 (2021), 2101.Google ScholarGoogle ScholarCross RefCross Ref
  32. [32] Chen H., Hou L., Zhang G. K., and Moon S.. 2021. Development of BIM, IoT and AR/VR technologies for fire safety and upskilling. Automation in Construction 125 (2021), 103631.Google ScholarGoogle ScholarCross RefCross Ref
  33. [33] Huang H., Lu Z., Peng R., Feng Z., Xuan X., Hung P. C. K., and Huang S.. 2019. Efficiently querying large process model repositories in smart city cloud workflow systems based on quantitative ordering relations. Information Sciences 495 (2019), 100115.Google ScholarGoogle ScholarDigital LibraryDigital Library
  34. [34] Wang H., Wang H., Zhou H., and Zhu T.. 2018. Modeling and optimization for hydraulic performance design in multi-source district heating with fluctuating renewables. Energy Conversion and Management 156 (2018), 113129.Google ScholarGoogle ScholarCross RefCross Ref
  35. [35] Yan R., Zhang F., Rao X., Lv Z., Li J., Zhang L., Liang S., Li Y., Ren F., Zheng C., and Liang J.. 2021. Richer fusion network for breast cancer classification based on multimodal data. BMC Medical Informatics and Decision Making 21, 1 (2019), 115.Google ScholarGoogle Scholar
  36. [36] Jing S., Li G., Zeng K., Pan W., and Liu C.. 2018. Efficient parallel algorithm for computing rough set approximation on GPU. Soft Computing 22, 22 (2019), 75537569.Google ScholarGoogle ScholarDigital LibraryDigital Library
  37. [37] Yamazaki I., Tomov S., and Dongarra J.. 2015. Mixed-precision Cholesky QR factorization and its case studies on multicore CPU with multiple GPUs. SIAM Journal on Scientific Computing 37, 3 (2015), C307C330.Google ScholarGoogle ScholarCross RefCross Ref
  38. [38] Luo F., Guo W., Yu Y., and Chen G.. 2017. A multi-label classification algorithm based on kernel extreme learning machine. Neurocomputing 260 (2017), 313320.Google ScholarGoogle ScholarDigital LibraryDigital Library
  39. [39] Zhang J., Li C., Sun Z., Luo Z., Zhou C., and Li S.. 2019. Towards a unified multi-source-based optimization framework for multi-label learning. Applied Soft Computing 76 (2019), 425435.Google ScholarGoogle ScholarDigital LibraryDigital Library
  40. [40] Sinha S., Bhatnagar V., and Bansal A.. 2018. Multi-label Naïve Bayes classifier for identification of top destination and issues to accost by tourism sector. Journal of Global Information Management (JGIM) 26, 3 (2018), 3753.Google ScholarGoogle ScholarCross RefCross Ref

Index Terms

  1. Smart City Construction and Management by Digital Twins and BIM Big Data in COVID-19 Scenario

          Recommendations

          Comments

          Login options

          Check if you have access through your login credentials or your institution to get full access on this article.

          Sign in

          Full Access

          • Published in

            cover image ACM Transactions on Multimedia Computing, Communications, and Applications
            ACM Transactions on Multimedia Computing, Communications, and Applications  Volume 18, Issue 2s
            June 2022
            383 pages
            ISSN:1551-6857
            EISSN:1551-6865
            DOI:10.1145/3561949
            • Editor:
            • Abdulmotaleb El Saddik
            Issue’s Table of Contents

            Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. Copyrights for components of this work owned by others than ACM must be honored. Abstracting with credit is permitted. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. Request permissions from [email protected].

            Publisher

            Association for Computing Machinery

            New York, NY, United States

            Publication History

            • Published: 6 October 2022
            • Online AM: 4 April 2022
            • Accepted: 28 March 2022
            • Revised: 11 February 2022
            • Received: 23 October 2021
            Published in tomm Volume 18, Issue 2s

            Permissions

            Request permissions about this article.

            Request Permissions

            Check for updates

            Qualifiers

            • research-article
            • Refereed

          PDF Format

          View or Download as a PDF file.

          PDF

          eReader

          View online with eReader.

          eReader

          HTML Format

          View this article in HTML Format .

          View HTML Format
          About Cookies On This Site

          We use cookies to ensure that we give you the best experience on our website.

          Learn more

          Got it!