Urban space resource optimal configuration method based on urban information model

Abstract

The invention provides an urban space resource optimal configuration method based on an urban information model, and relates to the field of space resource configuration, and the method comprises the steps: S1, data acquisition and preprocessing, S2, dynamic association feature extraction, S3, space-time cooperative supply and demand matching, S4, dynamic demand prediction, and S5, facility layout optimization. According to the method, on the basis of a city information model, iterative optimization of city space resources such as infrastructures in the whole life cycle of planning, construction and management is achieved, the problems that in the construction process or after the infrastructures are constructed, due to uncontrollable factors of city development, the requirements are reduced, and public resources are redundant are effectively solved, and the economic benefits of the infrastructures are improved. The utilization rate of urban space is effectively improved, and the planning and construction cost of infrastructures is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of space resource configuration, and in particular to a method for optimizing the configuration of urban space resources based on an urban information model. Background Art

[0002] City Information Modeling (CIM) is a three-dimensional spatial information model based on digital technology that integrates information on all elements and the entire life cycle of a city. It is the digital foundation that supports the construction of smart cities and refined urban governance. CIM builds a "digital city" that is synchronized with the city by integrating multi-dimensional information such as geographic spatial information, building information modeling (BIM), the Internet of Things (IoT), and government data. It provides intelligent support for urban planning, construction, management, and services, and manages urban operations in an integrated and systematic manner.

[0003] In the process of urban construction, infrastructure planning is the foundation for supporting the sustainable economic and social development, people's livelihood needs (green, livable, etc.) and ecological security of a region or city. It achieves orderly construction of a region or city through systematic and forward-looking arrangements for the layout, scale, construction sequence and operation mechanism of various basic engineering facilities (such as transportation, energy, water conservancy, communications, environmental protection, safety, medical care, etc.). It directly affects the development quality, operation efficiency and quality of life of residents in the region or city. However, due to the advanced nature of infrastructure planning, after planning and during construction, due to sudden changes in the social and economic structure (for example, unexpected population mobility, drastic adjustments in industrial structure, etc.), accelerated technological iteration (leading to the problem of "unbuilt and backward" infrastructure), sudden situations of natural environment and ecological constraints (for example, sudden earthquakes, floods, landslides and other problems during construction, resulting in project suspension or rerouting), uncontrollable risks in construction and supply chains, etc., there is a temporal and spatial mismatch between the "demand chain" and the "supply chain", resulting in the risk of "oversupply" or "backwardness" of advanced planned infrastructure. The completed infrastructure loses its original planned function, resulting in problems such as waste of resources and increased construction costs. Summary of the Invention

[0004] In response to the problems existing in the above-mentioned existing technologies, the purpose of the present invention is to provide a method for optimizing the configuration of urban space resources based on the urban information model. Based on the urban information model, this method realizes the iterative optimization of urban space resources such as infrastructure in the entire life cycle of planning, construction, and management, and effectively avoids the problems of reduced infrastructure demand and redundancy of public resources caused by uncontrollable factors of urban development after planning and during the construction process of infrastructure, thereby improving the utilization rate of urban space and reducing the planning and construction costs of infrastructure.

[0005] The purpose of the present invention is achieved through the following technical solutions: A method for optimizing the configuration of urban space resources based on an urban information model, comprising: Step S1, data acquisition and preprocessing: Based on the City Information Model (CIM), static and dynamic data in the city are acquired, and the data is preprocessed for spatiotemporal dynamic alignment to obtain multivariate heterogeneous data; Step S2, dynamic correlation feature extraction: convert multivariate heterogeneous data into spatiotemporal features to provide robust and semantically rich input for subsequent supply and demand matching, dynamic demand forecasting, and layout optimization; Step S3, spatiotemporal coordinated supply and demand matching: The spatiotemporal potential energy field is used to characterize the supply and demand relationship between facilities and planned areas. That is, first, facility weights and time decay factors are assigned according to facility types to obtain the supply potential energy of the corresponding type of facilities. Then, the regional demand factor and the spatiotemporal feature coding in step S2 are used to obtain the demand potential energy of the corresponding facilities. Finally, the supply and demand relationship is obtained through the supply potential energy and the demand potential energy. Step S4: Dynamic demand forecasting: Obtaining future facility demand forecasts to provide accurate and causally interpretable forecast results for layout optimization; Step S5: Facility layout optimization: Dynamic optimization of facility layout is achieved by using proximal strategy optimization.

[0006] Based on further optimization of the above solution, the static data includes infrastructure vectors, planning maps, etc.; dynamic data includes pedestrian flow, traffic, POI (Point of Interest) activity (a core indicator for measuring the dynamic performance of map points of interest in specific scenarios, which is usually a comprehensive evaluation based on multi-dimensional data such as pedestrian flow, user interaction, temporal and spatial characteristics), historical demand, social media popularity, festival calendar, traffic congestion index, etc.

[0007] Based on the further optimization of the above solution, the "preprocessing of data for spatiotemporal dynamic alignment" in step S1 includes temporal dynamic alignment and spatial adaptive alignment; Time dynamic alignment: For dynamic data, it makes the timestamps of all dynamic data consistent and eliminates the impact of data collection lag, including: Traffic congestion index T co (t) As the core of time dynamic alignment, the dynamic lag function is obtained:

[0008] Where: express k Class data at time t hysteresis; express kThe historical average lag of class data; ; Use the dynamic hysteresis function to correct the timestamps of all dynamic data:

[0009] Where: Indicates the minimum time granularity; t k express k The original timestamp of the class data; Spatial adaptive alignment: Use seats of infrastructure vectors (such as hospitals, shopping malls, etc.) as data source locations s j , and divide the grid units according to the land use nature of the planning map (such as commercial area, residential area, etc.) g i ; For aligned dynamic data, by data type j Assign its attenuation coefficient , and obtain the spatial weights:

[0010] Where: IOU represents the spatial intersection and union ratio; All data sources (including static data and dynamic data) are spatially weighted to obtain grid-level fusion data that is consistent in time and space, that is, multi-heterogeneous data. G i (t) :

[0011] Where: Data j (t) Indicates the j Class data at time t The original observation data.

[0012] Based on the further optimization of the above scheme, the step S2 introduces a dynamic neighborhood spatiotemporal autoencoder to convert multivariate heterogeneous data into spatiotemporal features, including an encoder and a decoder: Encoder, used to capture dynamic neighborhood associations: Based on real-time mutual information I(i,j,t) Screening Moment t With Grid i Strongly connected neighborhood N(i,t) :

[0013] Where: Indicates time t The maximum mutual information in the whole area is obtained; Real-time mutual information is a grid i and j The KL divergence of the joint distribution and independent distribution of the feature sequence in the local time window is:

[0014]

[0015] Where: S and D They represent the category labels after the discretization of features, which are the bridge from continuous features to discrete probabilities and are used to solve the computational difficulties of mutual information for continuous data. p ij (S,D) Representation Grid i In category S , Grid j In category D The joint probability of n ij (S,D) Represents feature pairs (S,D) The actual observation count of the grid i data S and grid j belong D The number of samples; p i (S) 、 p j (D) Represents the grid i Alone in category S The marginal probability of the grid j In category D The marginal probability of n i (S) Representation characteristics S The edge observation count of the grid i Alone in category S The number of samples, n j (D) Representation characteristics D The edge observation count of the grid j Alone in category D The number of samples; N Indicates the total sample size; Based on dynamic neighborhood N(i,t) Complete spatiotemporal feature encoding:

[0016] Where: z i,tThe feature vector representing the encoder output; Encoder represents the encoder function; Representing a dynamic neighborhood N(i,t) All grids in j The fused data set at time t is used to capture the grid i Real-time interaction with a strongly connected grid; Representation Grid i in the past L The fused data sequence of time steps is used to capture the historical time series trend of grid i; L Indicates the length of the time window, that is, the number of historical time steps, which is used to control the amount of historical data included in the encoder; Decoder, used to verify the validity of spatiotemporal features and strengthen neighborhood associations: Dynamic neighborhood constraints are added to the reconstruction error to ensure that spatiotemporal features contain temporary correlations. The loss function is:

[0017] Where: Decoder represents the decoder function; is the neighborhood constraint weight.

[0018] Based on the further optimization of the above scheme, in step S2, when converting multivariate heterogeneous data into spatiotemporal features, the Dirichlet prior + pseudo-count method is also used to solve the problem of misjudgment of sparse data (such as sparse data in late night and remote areas); First, by adding pseudo counts , smooth the probability estimates to obtain joint probability smoothing and edge probability smoothing respectively: Joint probability smoothing:

[0019] Edge probability smoothing:

[0020] Where: K Indicates the number of feature categories; Afterwards, the mutual information is calculated: .

[0021] Based on the further optimization of the above solution, the step S3 of "using the spatiotemporal potential energy field to depict the supply and demand relationship between the facilities and the planned land" is specifically as follows: First, assign facility weights based on facility type w k and the time decay factor T s (k) , and obtain the supply potential energy of the corresponding type of facilities:

[0022] Where: C k express k the size of the class data (i.e., facilities); express k Class data (i.e. facilities) in t Real-time utilization at the moment; x Represents the spatial position variable, that is, the target position; x k express k The actual location of the class data (i.e., facility); t Represents the time variable, that is, the current moment; t k express k Peak service hours for the data type (i.e., facility); represents the spatial attenuation coefficient; By regional demand factor B(x) Obtain the demand potential of the corresponding facility (the regional demand factor is defined according to the regionality of the corresponding facility):

[0023] Where: w q represents the demand weight vector of the corresponding facility; Get the grid i At the moment t Supply and demand matching M(i,t) :

[0024] Where: Represents the total penalty coefficient; t reach (i,k) Representation Grid i To the facility k The space-time reachable time; T 0 indicates the reachable time threshold.

[0025] Based on further optimization of the above solution, the dynamic demand forecasting in step S4 is specifically as follows: Introducing causal gates into the LSTM (Long Short-Term Memory) model to distinguish the causal impact of exogenous variables such as policy releases, emergencies, social media buzz, and festivals; First, the causal variables are decomposed into endogenous variables according to their specific types X end (t) With exogenous variables X exo(t) ,endogenous variables include time-dependent variables such as historical demand and passenger flow, and exogenous variables include causal intervention variables such as policy releases, emergencies, social media popularity, and festivals; Then, a causal gate is introduced into the LSTM model to modify the cell state to incorporate the intervention effect: Input Gate:

[0026] Where: Represents the Sigmiod activation function; W i Represents the input weight, connected to the current moment input x t The weight matrix to the input gate; x t Indicates the current moment input; U i Represents the hidden state weight, connected to the previous hidden state h t-1 The weight matrix to the input gate; h t-1 Indicates the hidden state at the previous moment; V i Represents the cell state weight, connected to the cell state at the previous moment c t-1 The weight matrix to the input gate; c t-1 Represents the cell state at the previous moment (the core carrier of LSTM to store historical information); b i represents the input gate bias; Forget Gate:

[0027] Where: W f 、 U f 、 V f Represent the weight matrices from input, hidden state, cell state to forget gate respectively; b f represents the forget gate bias; Candidate status:

[0028] Where: tanh() represents the Tanh activation function; W c 、 U c Represent the weight matrices from input and hidden state to candidate state respectively; b c represents the candidate state bias; Causally correcting cell states:

[0029] Where: represents element-wise product; Indicates causal strength; represents the change in cell state caused by the intervention; Output gate:

[0030] Where: W o 、 U o 、 V o Represent the weight matrices from input, hidden state, current cell state to output gate respectively; b o represents the forget gate bias; Output status:

[0031] Causal strength is obtained through MLP (Multilayer Perceptron), which consists of an input layer, two hidden layers, and an output layer. The layers are fully connected and the activation function is ReLU:

[0032] The PC (Peter-Clark) algorithm is used to construct a causal diagram of exogenous variables to obtain the changes in cell states caused by intervention. :

[0033] Where: do() Indicates the budget for intervention; Represents the causal coefficient obtained from data mining, used to measure variables k marginal impact on demand; Finally, output the future time Demand forecast:

[0034] Where: W h Represents the weight matrix input to the output layer; b h represents the bias of the output layer; represents the predicted volatility coefficient.

[0035] Based on the further optimization of the above solution, the “facility layout optimization” in step S5 is specifically as follows: First, define the state space St and action space At:

[0036] Where: G i (t)} represents dynamically integrated multi-dimensional heterogeneous data, reflecting the real-time supply and demand relationship, and is output by step S1; JS k (t)} represents the construction progress, constraining the “temporal feasibility” of the decision, which is obtained based on the actual construction progress; Expressing the future The demand forecast is output by step S4; t build Indicates construction timing; C Indicates the size of the facility; C max Indicates the resource ceiling (i.e., the upper limit of the infrastructure budget for the corresponding region); A O represents the feasible action space, i.e., the set of all facility layout decisions that satisfy resource constraints; Then, we set the reward function to distinguish the importance of facility areas and achieve dynamic matching:

[0037] Where: B(x) i represents the regional demand factor (used to measure the priority of resource matching); Respectively represent the corresponding weight coefficients; It represents the average matching degree when the facility is built, which is obtained by the supply-demand matching degree in step S3. M(i, t) Obtain, that is, use step S3 to predict the supply and demand matching degree when it is built M(i,t+t build ) , then for all grids i , and obtain the average matching degree according to the regional demand factor:

[0038] represents redundancy (i.e., the proportion of time and space when facility utilization is below the threshold); Finally, through the policy network Output the layout decision probability, and then iteratively update the network parameters to achieve layout optimization (PPO, Proximal Policy Optimization):

[0039] Where: Indicates the n Parameters before the update; represents the learning rate (obtained through experimental data); E Represents the expectation of the empirical data distribution of "state-action-advantage"; Y t represents the advantage function, which guides the direction of policy update; clip() Represents a cropping operation; represents the clipping factor (hyperparameter).

[0040] Based on the further optimization of the above scheme, the method for obtaining redundancy is: first, the facility k The position coordinates are taken as the center and the radius is generated. r k The circular area is taken as the service range of the facility; then, the causal gate is introduced into the LSTM model in step S4 to predict the service range of the facility. i The requirements of a grid at the time of construction ; The grid g i Geometric scope and facilities k Perform spatial intersection analysis on the service scope and obtain the intersection area ratio IOU(g i ,k) , preset intersection threshold, if the intersection area is greater than the intersection threshold, then the demand is summarized:

[0041] Afterwards, get the predicted utilization:

[0042] Where: t yy Indicates the actual operating hours of the facility; t yc Indicates the duration of forecast application in the facility planning process; Preset utilization thresholds and statistically predict utilization The time and space proportion that is less than the utilization threshold is used to obtain redundancy.

[0043] The following are the technical effects of the solution of the present invention: This invention transforms multi-dimensional, heterogeneous data (static and dynamic) into high-quality spatiotemporal features through dynamic correlation feature extraction. This effectively addresses issues such as lag, decay, and sparseness in the original data, providing robust and semantically rich input for subsequent supply-demand matching and layout optimization. Subsequently, through spatiotemporal coordinated supply-demand matching, not only does it accurately depict supply-demand relationships and improve resource allocation efficiency, but it also effectively corrects the mismatch and avoids resource mismatches. This in turn supports dynamic layout optimization, quantifies the value of spatial resources, and forms efficient and accurate long-term planning. Dynamic demand forecasting not only accurately captures demand drivers and improves forecast accuracy, but also quantifies forecast uncertainty, making layout decision-making risk manageable. This provides accurate, robust, and interpretable demand forecasts for subsequent layout optimization. Finally, through facility layout optimization, it not only ensures regional equity and optimizes resource utilization, but also effectively responds to urban emergencies and long-term trends, thereby enhancing the dynamic responsiveness of the entire model. The present invention shifts resource allocation from "experience-driven" to "data-model-driven" through a combination of steps: data acquisition and preprocessing - dynamic correlation feature extraction - spatiotemporal coordinated supply and demand matching - dynamic demand prediction - facility layout optimization, thereby achieving spatiotemporal precision, type adaptation, and dynamic elastic optimization of urban space resources, and realizing iterative optimization of urban space resources throughout their entire life cycle of planning, construction, and management. This avoids the situation where, after planning and during the construction of infrastructure, the city's demand for facilities is reduced due to time lags and sudden factors, leading to problems such as resource waste and planning incompatibility. This improves the spatiotemporal matching accuracy of resource allocation with urban operation needs and reduces the redundancy of public resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 This is a structural diagram of the method for collaborative optimization of urban space resources configuration in an embodiment of the present invention. DETAILED DESCRIPTION

[0045] The technical solutions in the embodiments of the present invention will be described clearly and completely below. In the following description, for the purpose of illustration rather than limitation, specific details such as specific system structures and technologies are proposed to facilitate a thorough understanding of the embodiments of the present invention.

[0046] Example 1: A method for optimizing the configuration of urban space resources based on an urban information model, comprising: Step S1, data acquisition and preprocessing: Based on the city information model (CIM), static data and dynamic data in the city are obtained. Static data include infrastructure vectors, planning maps, etc. (which can be obtained through surveying and 3D modeling, for example: using oblique photography, lidar, etc. to collect urban geographic spatial data and generate high-precision 3D models; at the same time, in collaboration with natural resources, planning, housing and construction departments, auxiliary satellite image data is used to obtain administrative divisions, land use planning, transportation network and other data); dynamic data include pedestrian flow, traffic, POI activity (a core indicator to measure the dynamic performance of map points of interest in specific scenarios, which is usually combined with multi-dimensional data such as pedestrian flow, user interaction, time and space characteristics for comprehensive evaluation), historical demand (through the deployment of flow sensors and cameras on roads and public transportation stations to collect pedestrian and vehicle flow data in real time; through the export of historical demand and POI from government information systems such as census and traffic management systems activity, etc.), social media popularity, festival calendar, traffic congestion index (social media popularity information and festival calendar information can be crawled through web crawler technology; traffic congestion index can be obtained through the real-time traffic data of the traffic management department and the deployment of geomagnetic sensors, radar detectors, etc. on key roads), etc.

[0047] The data is pre-processed for spatiotemporal dynamic alignment to obtain multivariate heterogeneous data. Spatiotemporal dynamic alignment includes time dynamic alignment and spatial adaptive alignment. Time dynamic alignment: For dynamic data, it makes the timestamps of all dynamic data consistent and eliminates the impact of data collection lag, including: Traffic congestion index T co (t) As the core of time dynamic alignment, the dynamic lag function is obtained:

[0048] Where: express k Class data at time t hysteresis; express k The historical average lag of a type of data (obtained by statistically averaging historical data. For example, for the historical average lag of traffic data, 1,000 historical traffic accidents can be collected. The actual time of the historical traffic accident or the actual time of the start of congestion can be obtained as the event time. The time when the historical traffic accident sensor or camera and other instruments uploaded the corresponding traffic accident is simultaneously obtained as the collection time. The difference between the collection time and the event time is the lag of the corresponding traffic accident. The average lag of 1,000 historical traffic accidents is the historical average lag. This method can realize the delay correction of the collection chain, thereby accurately quantifying the historical lag characteristics of each type of data and providing key parameters for spatiotemporal causal alignment). (0 means unimpeded traffic, 1 means severe congestion); Use the dynamic hysteresis function to correct the timestamps of all dynamic data:

[0049] Where: Indicates the minimum time granularity (usually 4 to 6 minutes); t k express k The original timestamp of the class data; For example, for each type of data (such as passenger flow, traffic, POI activity, historical demand, social media popularity, festival calendar, traffic congestion index, etc.), the delay on its collection chain is obtained through a dynamic hysteresis function. (For example: people flow data is collected in real time, POI activity is reported in real time, historical demand is collected without delay, and social media popularity is crawled in real time, i.e. ,but , k Any one of the following categories: pedestrian flow, POI activity, historical demand, and social media popularity; for traffic flow, its impact on the traffic congestion index is obtained based on the historical average lag, which is the lag effect, that is, traffic flow is reflected after traffic congestion, then their corresponding is a positive number, and the timestamp is obtained according to the corresponding historical average lag. For the festival calendar, the impact of the festival on the traffic congestion index is obtained according to the historical average lag as the advance impact. For example, the impact of the National Day holiday on the traffic congestion index is reflected 3 days in advance, so their corresponding is a negative number, and the corresponding timestamp is obtained by calculating the historical average lag of the three days before the National Day holiday). Spatial adaptive alignment: Use seats of infrastructure vectors (such as hospitals, shopping malls, etc.) as data source locations s j , and divide the grid units according to the land use nature of the planning map (such as commercial area, residential area, etc.) g i ; For aligned dynamic data, by data type j Assign its attenuation coefficient (The attenuation coefficient is obtained based on the actual dynamic data type and empirical data; for example, if the short distance of the flow of people is significantly affected, then ; POI data has a stable long-distance impact, then ), and obtain the spatial weight:

[0050] Where: IOU represents the spatial intersection and union ratio; All data sources (including static data and dynamic data) are spatially weighted to obtain grid-level fusion data that is consistent in time and space, that is, multi-heterogeneous data. G i (t) :

[0051] Where: Data j (t) Indicates the j Class data at time t The original observation data.

[0052] Step S2, dynamic correlation feature extraction: convert multivariate heterogeneous data into spatiotemporal features to provide robust and semantically rich input for subsequent supply and demand matching, dynamic demand forecasting, and layout optimization; Including encoder and decoder: Encoder, used to capture dynamic neighborhood associations: Based on real-time mutual information I(i,j,t) Screening Moment t With Grid i Strongly connected neighborhood N(i,t) :

[0053] Where: Indicates time t Maximize mutual information across the entire region (ensuring that only strongly connected grids are included, such as large venues and their surrounding transportation hubs); Real-time mutual information is a grid i and j The KL divergence of the joint distribution and independent distribution of the feature sequence in the local time window is:

[0054]

[0055] Where: S and D They represent the category labels after the discretization of features, which are the bridge from continuous features to discrete probabilities and are used to solve the computational difficulties of mutual information for continuous data. p ij (S,D) Representation Grid i In category S , Grid j In category D The joint probability of n ij (S,D) Represents feature pairs (S,D)The actual observation count of the grid i data S and grid j belong D The number of samples; p i (S) 、 p j (D) Represents the grid i Alone in category S The marginal probability of the grid j In category D The marginal probability of n i (S) Representation characteristics S The edge observation count of the grid i Alone in category S The number of samples, n j (D) Representation characteristics D The edge observation count of the grid j Alone in category D The number of samples; N Indicates the total sample size; Based on dynamic neighborhood N(i,t) Complete spatiotemporal feature encoding:

[0056] Where: z i,t The feature vector representing the encoder output; Encoder Represents the encoder function (usually a neural network such as LSTM, CNN, Transformer, etc.); Representing a dynamic neighborhood N(i,t) All grids in j The fused data set at time t is used to capture the grid i Real-time interaction with a strongly connected grid; Representation Grid i in the past L The fused data sequence of time steps is used to capture the historical time series trend of grid i; L Indicates the length of the time window, that is, the number of historical time steps, which is used to control the amount of historical data included in the encoder; Decoder, used to verify the validity of spatiotemporal features and strengthen neighborhood associations: Dynamic neighborhood constraints are added to the reconstruction error to ensure that spatiotemporal features contain temporary correlations. The loss function is:

[0057] Where: Decoder represents the decoder function; is the neighborhood constraint weight (usually 0.1 to 0.3).

[0058] In the process of converting multivariate heterogeneous data into spatiotemporal features, the Dirichlet prior + pseudo-count method is also used to solve the problem of misjudgment of sparse data (such as sparse data in late night and remote areas); First, by adding pseudo counts (usually 0.1), smooth the probability estimate to obtain joint probability smoothing and edge probability smoothing respectively: Joint probability smoothing:

[0059] Edge probability smoothing:

[0060] Where: K Indicates the number of feature categories; Afterwards, the mutual information is calculated: ; J(S,D) By introducing pseudo-counts and solving the problem of misjudgment of sparse data, the purpose is to correct the real-time mutual information.

[0061] Step S3, spatiotemporal coordinated supply and demand matching: The spatiotemporal potential field is used to characterize the supply and demand relationship between facilities and planned areas. That is, first, facility weights and time decay factors are assigned according to facility types to obtain the supply potential of the corresponding type of facilities. Then, the regional demand factor and the spatiotemporal feature coding in step S2 are used to obtain the demand potential of the corresponding facilities. Finally, the supply and demand relationship is obtained through the supply potential and demand potential. Specifically: First, assign facility weights based on facility type w k and the time decay factor T s (k) (For example, the priority of a hospital is higher than that of a shopping mall, and the hospital needs to respond 24 hours a day and has a slow time decay, while the shopping mall has a concentrated peak and a fast time decay, so the facility weight of the hospital can be w k1 =1.5, the time attenuation factor can be T s (k 1 ) =24h, and the facility weight of the mall can be w k2 =1, the time decay factor can be T s (k 2 ) =6h), and obtain the supply potential energy of the corresponding type of facilities:

[0062] Where: C k express k the size of the class data (i.e., facilities); express k Class data (i.e. facilities) in t Real-time utilization at the moment (obtained by comparing the actual amount of resources used with the total amount of available resources); x Represents the spatial position variable, that is, the target position; x k express k The actual location of the class data (i.e., facility); t Represents the time variable, that is, the current moment; t k express k Peak service hours for the data type (i.e., facility); represents the spatial attenuation coefficient (obtained through experimental data); By regional demand factor B(x) Obtain the demand potential of the corresponding facilities (the regional demand factor is defined according to the regionality of the corresponding facilities. For example, if the demand for facilities in the central city is higher than that in the suburbs, the demand in the central city is higher than that in the suburbs. B(x 1 )=1.2 , suburban B (x 2 )=0.8 ):

[0063] Where: w q Represents the demand weight vector of the corresponding facility (combined with the multi-source characteristics of urban spatiotemporal data and the corresponding facility type, obtained by statistically analyzing the historical demand data of the corresponding facility); Get the grid i At the moment t Supply and demand matching M(i,t) :

[0064] Where: Indicates the total penalty coefficient (usually 0.1); t reach (i,k) Representation Grid i To the facility k The spatial and temporal reachable time (available through real-time traffic data); T 0 indicates the reachable time threshold.

[0065] Step S4: Dynamic demand forecasting: Obtaining future facility demand forecasts to provide accurate and causally interpretable forecast results for layout optimization; specifically: Introducing causal gates into the LSTM (Long Short-Term Memory) model to distinguish the causal impact of exogenous variables such as policy releases, emergencies, social media buzz, and festivals (for example, sudden changes in demand caused by new district planning); First, the causal variables are decomposed into endogenous variables according to their specific types X end (t) With exogenous variables X exo (t) ,endogenous variables include time-dependent variables such as historical demand and passenger flow, and exogenous variables include causal intervention variables such as policy releases, emergencies, social media popularity, and festivals; Then, a causal gate is introduced into the LSTM model to modify the cell state to incorporate the intervention effect: Input Gate:

[0066] Where: Represents the Sigmiod activation function (output range [0,1]); W i Represents the input weight, connected to the current moment input x t The weight matrix to the input gate (obtained through training parameters); x t Indicates the current moment input; U i Represents the hidden state weight, connected to the previous hidden state h t-1 The weight matrix to the input gate (obtained through training parameters); h t-1 Indicates the hidden state at the previous moment; V i Represents the cell state weight, connected to the cell state at the previous moment c t-1 The weight matrix to the input gate (obtained through training parameters); c t-1 Represents the cell state at the previous moment (the core carrier of LSTM to store historical information); b i represents the input gate bias (obtained through training parameters); Forget Gate:

[0067] Where: W f 、U f 、 V f Represent the weight matrices from input, hidden state, cell state to forget gate respectively; b f represents the forget gate bias; Candidate status:

[0068] Where: tanh() represents the Tanh activation function (output range [-1,1]); W c 、 U c Represent the weight matrices from input and hidden state to candidate state respectively; b c represents the candidate state bias; Causally correcting cell states:

[0069] Where: represents element-wise product; Indicates causal strength; represents the change in cell state caused by the intervention; Output gate:

[0070] Where: W o 、 U o 、 V o Represent the weight matrices from input, hidden state, current cell state to output gate respectively; b o represents the forget gate bias; Output status:

[0071] Causal strength is obtained through MLP (Multilayer Perceptron), which consists of an input layer, two hidden layers, and an output layer. The layers are fully connected and the activation function is ReLU:

[0072] The PC (Peter-Clark) algorithm is used to construct a causal diagram of exogenous variables to obtain the changes in cell states caused by intervention. :

[0073] Where: do() Indicates the budget for intervention; Represents the causal coefficient obtained from data mining, used to measure variablesk marginal impact on demand (obtained using actual data); Finally, output the future time Demand forecast:

[0074] Where: W h Represents the weight matrix input to the output layer; b h represents the bias of the output layer; Represents the predicted fluctuation coefficient (obtained through past experimental data).

[0075] Step S5, facility layout optimization: Dynamic optimization of facility layout is achieved by using proximal strategy optimization, specifically: First, define the state space St and action space At :

[0076] Where: G i (t)} represents dynamically integrated multi-dimensional heterogeneous data, reflecting the real-time supply and demand relationship, and is output by step S1; JS k (t)} represents the construction progress, constraining the “temporal feasibility” of the decision, which is obtained based on the actual construction progress; Expressing the future The demand forecast is output by step S4; t build Indicates construction timing; C Indicates the size of the facility; C max Indicates the resource ceiling (i.e., the upper limit of the infrastructure budget for the corresponding region); A O represents the feasible action space, i.e., the set of all facility layout decisions that satisfy resource constraints; x Indicates the location of the facility; Then, we set the reward function to distinguish the importance of facility areas and achieve dynamic matching:

[0077] Where: B(x) i Represents the regional demand factor (used to measure the priority of resource matching, such as the central city is higher than the suburbs); Respectively represent the corresponding weight coefficients (obtained through experimental data); It represents the average matching degree when the facility is built, which is obtained by the supply-demand matching degree in step S3. M(i, t) Obtain, that is, use step S3 to predict the supply and demand matching degree when it is built M(i,t+t build ) , then for all grids i , and obtain the average matching degree according to the regional demand factor:

[0078] Represents redundancy (i.e., the proportion of time and space when the facility utilization rate is lower than the threshold), which is obtained by first k The position coordinates are taken as the center and the radius is generated. r k The circular area is taken as the service range of the facility; then, the causal gate is introduced into the LSTM model in step S4 to predict the service range of the facility. i The requirements of a grid at the time of construction ; The grid g i Geometric scope and facilities k Perform spatial intersection analysis on the service scope and obtain the intersection area ratio IOU(g i ,k) , preset the intersection threshold (such as 0.5). If the intersection area is greater than the intersection threshold, the demand is summarized:

[0079] Afterwards, get the predicted utilization:

[0080] Where: t yy Indicates the actual operating hours of the facility; t yc Indicates the duration of forecast application in the facility planning process; Preset utilization rate thresholds (set according to facility type and experimental data statistics, for example: 0.8 for schools, 0.7 for hospitals, and 0.5 for shopping malls), and statistically predict utilization rates The time and space proportion that is less than the utilization threshold is used to obtain redundancy.

[0081] Finally, through the policy network Output the layout decision probability, and then iteratively update the network parameters to achieve layout optimization (PPO, Proximal Policy Optimization):

[0082] Where: Indicates the n Parameters before the update; represents the learning rate (obtained through experimental data); E represents the expectation of the empirical data distribution of "state-action-advantage" (obtained by taking a batch of samples from the training data and calculating the average); Y t represents the advantage function, which guides the direction of policy update; clip() Represents a cropping operation; represents the clipping coefficient (hyperparameter, usually 0.1 to 0.3);

[0083] Where: Q(St,At) is the action value function (is the state St Select Action At Expected cumulative reward of ); V(St) is the state value function (is the state St The advantage function guides the direction of policy updates: a positive advantage encourages the action, while a negative advantage inhibits the action.

[0084] Example 2: As another preferred embodiment of the technical solution of the present invention, based on the solution of Example 1 above, iterative optimization is further performed, that is, dynamic adaptation to the stage of urban development is achieved through dynamic error feedback, adaptive parameter update and staged triggering layout; Dynamic error feedback: Collect actual operational data of completed facilities through City Information Modeling (CIM) D real (s) , obtain actual operation data and predicted values The error between:

[0085] Where: Indicates the attenuation coefficient (usually between (0, 1), set according to actual conditions); Adaptive FTRL (Follow The Regularized Leader) update:

[0086] Where: N Indicates the amount of accumulated data (generally N = t , that is, at the time t the number of samples); Indicates dynamic L1 regularization. The less data (e.g., new areas), the stronger the regularization. represents dynamic L2 regularization. The more data (i.e., mature urban areas), the weaker the regularization.

[0087] Where: 、 represents the initial regularization coefficient; Phased trigger layout: First, principal component analysis (PCA) is used to integrate multi-dimensional indicators to construct a comprehensive urban development index; the multi-dimensional indicators include population growth rate, GDP growth rate, infrastructure investment rate, and employment rate; each indicator is normalized to [0,1], and its corresponding principal component is obtained through PCA dimensionality reduction. PC i (t) ; Obtain the city's comprehensive development index through the main city scores of each indicator ZH(t) :

[0088] Where: Indicates the i The principal component weight coefficients of the indicators (obtained through experimental data); Then, the mismatch entropy threshold is obtained by the city comprehensive development index H 0 (t) :

[0089] Where: H 00 represents the stable development threshold (obtained through experimental data), i.e., the static benchmark of mismatch tolerance; Get the actual misfit entropy of the grid H(t) :

[0090] Where: N w Indicates the total number of city grids; represents the time decay factor ( t - t i The larger the value, the lower the weight of historical mismatch, and the current mismatch is given priority); When the actual mismatch entropy at K consecutive moments H(t) Greater than the mismatch entropy threshold H 0 at time (t), Trigger layout reoptimization:

[0091] Where: Indicates that the strategy network in step S5 is adopted; S(t + K) Indicates update status; A(t + K) Indicates the action of updating; Indicates the updated parameters.

[0092] Closed-loop logic for iterative optimization: Data collection and error calculation: real-time collection of demand data D real (s) , calculate the dynamic error e(t) , weakening the impact of obsolete data; Model parameter update: based on e(t) and data volume N , adaptive FTRL adjusts parameters to balance underfitting / overfitting; Trigger condition judgment: Based on the city's comprehensive development index ZH(t) , adjust the mismatch entropy threshold H 0 (t) ,continuous K If the target is exceeded once, re-optimization will be triggered; Layout optimization execution: based on the new state S(t + K) and new parameters , generate a new layout plan A(t + K) ; Iterative loop: After the new layout is implemented, repeat the above steps to achieve a closed-loop evolution of "data → model → decision → data".

[0093] Example 3: A system for optimizing urban infrastructure configuration includes a data acquisition and preprocessing module, a dynamic correlation feature extraction module, a spatiotemporal coordinated supply and demand matching module, a dynamic demand prediction module, a facility layout optimization module, and an iterative optimization module, wherein each module adopts the corresponding method in Example 1 and Example 2 respectively.

Claims

1. A method for optimizing the allocation of urban space resources based on an urban information model, characterized by: include: Step S1, data acquisition and preprocessing: Based on the city information model, static data and dynamic data in the city are acquired, and the data are preprocessed for spatiotemporal dynamic alignment to obtain multivariate heterogeneous data; Step S2, dynamic correlation feature extraction: convert multivariate heterogeneous data into spatiotemporal features to provide robust and semantically rich input for subsequent supply and demand matching, dynamic demand forecasting, and layout optimization; Step S3, spatiotemporal coordinated supply and demand matching: The spatiotemporal potential field is used to characterize the supply and demand relationship between facilities and planned areas. That is, first, facility weights and time decay factors are assigned according to facility types to obtain the supply potential of the corresponding type of facilities. Then, the regional demand factor and the spatiotemporal feature coding in step S2 are used to obtain the demand potential of the corresponding facilities. Finally, the supply and demand relationship is obtained through the supply potential and demand potential. Step S4: Dynamic demand forecasting: Obtaining future facility demand forecasts to provide accurate and causally interpretable forecast results for layout optimization; Step S5: Facility layout optimization: Dynamic optimization of facility layout is achieved by using proximal strategy optimization.

2. The method for optimizing the allocation of urban space resources based on an urban information model according to claim 1, characterized in that: The static data includes infrastructure vectors and planning maps; the dynamic data includes passenger flow, traffic, POI activity, historical demand, social media popularity, festival calendar, and traffic congestion index.

3. The method for optimizing the allocation of urban space resources based on an urban information model according to claim 2, characterized in that: The "preprocessing of data for spatiotemporal dynamic alignment" in step S1 includes temporal dynamic alignment and spatial adaptive alignment; Time dynamic alignment: For dynamic data, it makes the timestamps of all dynamic data consistent and eliminates the impact of data collection lag, including: Traffic congestion index T co (t) As the core of time dynamic alignment, the dynamic lag function is obtained: Where: express k Class data at time t hysteresis; express k The historical average lag of class data; ; Use the dynamic hysteresis function to correct the timestamps of all dynamic data: Where: Indicates the minimum time granularity; t k express k The original timestamp of the class data; Spatially adaptive alignment: using the seat of the infrastructure vector as the data source location s j , and divide the grid units according to the land use characteristics of the planning map g i ; For aligned dynamic data, by data type j Assign its attenuation coefficient , and obtain the spatial weights: Where: IOU represents the spatial intersection and union ratio; All data sources are spatially weighted to obtain grid-level fusion data that is consistent in time and space, i.e. multivariate heterogeneous data. G i (t) : Where: Data j (t) Indicates the j Class data at time t The original observation data.

4. The method for optimizing the allocation of urban space resources based on an urban information model according to claim 3, characterized in that: In step S2, a dynamic neighborhood spatiotemporal autoencoder is introduced to convert multivariate heterogeneous data into spatiotemporal features, including an encoder and a decoder: Encoder, used to capture dynamic neighborhood associations: Based on real-time mutual information I(i,j,t) Screening Moment t With Grid i Strongly connected neighborhood N(i,t) : Where: Indicates time t The maximum mutual information in the whole area is obtained; Real-time mutual information is a grid i and j The KL divergence of the joint distribution and independent distribution of the feature sequence in the local time window is: Where: S and D They represent the category labels after the discretization of features, which are the bridge from continuous features to discrete probabilities and are used to solve the computational difficulties of mutual information for continuous data. p ij (S,D) Representation Grid i In category S , Grid j In category D The joint probability of n ij (S,D) Represents feature pairs (S,D) The actual observation count of the grid i data S and grid j belong D The number of samples; p i (S) 、 p j (D) Represents the grid i Alone in category S The marginal probability of the grid j In category D The marginal probability of n i (S) Representation characteristics S The edge observation count of the grid i Alone in category S The number of samples, n j (D) Representation characteristics D The edge observation count of the grid j Alone in category D The number of samples; N Indicates the total sample size; Based on dynamic neighborhood N(i,t) Complete spatiotemporal feature encoding: Where: z i,t The feature vector representing the encoder output; Encoder represents the encoder function; Representing a dynamic neighborhood N(i,t) All grids in j The fused data set at time t is used to capture the grid i Real-time interaction with a strongly connected grid; Representation Grid i in the past L The fused data sequence of time steps is used to capture the historical time series trend of grid i; L Indicates the length of the time window, that is, the number of historical time steps, which is used to control the amount of historical data included in the encoder; Decoder, used to verify the validity of spatiotemporal features and strengthen neighborhood associations: Dynamic neighborhood constraints are added to the reconstruction error to ensure that spatiotemporal features contain temporary correlations. The loss function is: Where: Decoder represents the decoder function; is the neighborhood constraint weight.

5. The method for optimizing the allocation of urban space resources based on the urban information model according to claim 4, characterized in that: The specific steps of "using the spatiotemporal potential energy field to depict the supply and demand relationship between facilities and planned areas" in step S3 are: First, assign facility weights based on facility type w k and the time decay factor T s (k) , and obtain the supply potential energy of the corresponding type of facilities: Where: C k express k The size of the class data; express k Class data in t Real-time utilization at the moment; x Represents the spatial position variable, that is, the target position; x k express k The actual location of the class data; t Represents the time variable, that is, the current moment; t k express k Peak service hours for this type of data; represents the spatial attenuation coefficient; By regional demand factor B(x) Obtain the required potential energy of the corresponding facility: Where: w q represents the demand weight vector of the corresponding facility; Get the grid i At the moment t Supply and demand matching M(i,t) : Where: Represents the total penalty coefficient; t reach (i,k) Representation Grid i To the facility k The space-time reachable time; T 0 indicates the reachable time threshold.

6. The method for optimizing the allocation of urban space resources based on an urban information model according to claim 4 or 5, characterized in that: The dynamic demand forecasting step S4 is specifically as follows: Introducing causal gates into the LSTM model to distinguish the causal impact of exogenous variables such as policy releases, emergencies, social media popularity, and festivals; First, the causal variables are decomposed into endogenous variables according to their specific types X end (t) With exogenous variables X exo (t) ,endogenous variables include time-dependent variables of historical demand and passenger flow, and exogenous variables include causal intervention variables of policy releases, emergencies, social media popularity, and festival days; Then, a causal gate is introduced into the LSTM model to modify the cell state to incorporate the intervention effect: Input Gate: Where: Represents the Sigmiod activation function; W i Represents the input weight, connected to the current moment input x t The weight matrix to the input gate; x t Indicates the current moment input; U i Represents the hidden state weight, connected to the previous hidden state h t-1 The weight matrix to the input gate; h t-1 Indicates the hidden state at the previous moment; V i Represents the cell state weight, connected to the cell state at the previous moment c t-1 The weight matrix to the input gate; c t-1 Indicates the cell state at the previous moment; b i represents the input gate bias; Forget Gate: Where: W f 、 U f 、 V f Represent the weight matrices from input, hidden state, cell state to forget gate respectively; b f represents the forget gate bias; Candidate status: Where: tanh() represents the Tanh activation function; W c 、 U c Represent the weight matrices from input and hidden state to candidate state respectively; b c represents the candidate state bias; Causally correcting cell states: Where: represents element-wise product; Indicates causal strength; represents the change in cell state caused by the intervention; Output gate: Where: W o 、 U o 、 V o Represent the weight matrices from input, hidden state, current cell state to output gate respectively; b o represents the forget gate bias; Output status: The causal strength is obtained through MLP, which consists of an input layer, two hidden layers, and an output layer. The layers are fully connected, and the activation function uses ReLU: The PC (Peter-Clark) algorithm is used to construct a causal diagram of exogenous variables to obtain the changes in cell states caused by intervention. : Where: do() Indicates the budget for intervention; Represents the causal coefficient obtained from data mining, used to measure variables k marginal impact on demand; Finally, output the future time Demand forecast: Where: W h Represents the weight matrix input to the output layer; b h represents the bias of the output layer; represents the predicted volatility coefficient.

7. The method for optimizing the allocation of urban space resources based on the urban information model according to claim 6, characterized in that: The "facility layout optimization" in step S5 is specifically as follows: First, define the state space St and action space At : Where: G i (t) } represents dynamically integrated multi-dimensional heterogeneous data, reflecting the real-time supply and demand relationship, and is output by step S1; JS k (t) } represents the construction progress and constrains the "temporal feasibility" of the decision, which is obtained based on the actual construction progress; Expressing the future The demand forecast is output by step S4; t build Indicates construction timing; C Indicates the size of the facility; C max Indicates resource upper limit; A O represents the feasible action space, i.e., the set of all facility layout decisions that satisfy resource constraints; Then, we set the reward function to distinguish the importance of facility areas and achieve dynamic matching: Where: B(x) i represents the regional demand factor; Respectively represent the corresponding weight coefficients; It represents the average matching degree when the facility is built, which is obtained by the supply-demand matching degree in step S3. M(i,t) Obtain, that is, use step S3 to predict the supply and demand matching degree when it is built M(i,t+t build ) , then for all grids i , and obtain the average matching degree according to the regional demand factor: Indicates redundancy; Finally, through the policy network Output the layout decision probability, and then iteratively update the network parameters to achieve layout optimization: Where: Indicates the n Parameters before the update; represents the learning rate; E Represents the expectation of the empirical data distribution of "state-action-advantage"; Y t represents the advantage function, which guides the direction of policy update; clip() Represents a cropping operation; Represents the crop factor.

Images