Urban resident daily activity space identification diagnosis and optimization method based on multi-agent supply and demand deduction

Abstract

The invention discloses an urban resident daily activity space identification diagnosis and optimization method based on multi-agent supply and demand deduction. Comprising the following steps of town space and resident travel basic data integration, daily activity measurement and hotspot space identification, daily activity-travel decision network construction, multi-agent supply and demand simulation and space matching diagnosis, daily activity space optimization scheme generation and deduction, and optimization scheme interactive output feedback. According to multi-agent supply and demand deduction, accurate identification, scientific diagnosis and intelligent optimization of daily activity space of urban residents are realized, and the efficiency of urban space resource allocation and accurate treatment can be remarkably improved.

Description

Technical Field

[0001] This invention belongs to the technical field, and in particular relates to a method for identifying, diagnosing and optimizing the daily activity space of urban residents based on multi-agent supply and demand extrapolation. Background Technology

[0002] With the acceleration of urbanization, residents' daily activity spaces are becoming increasingly diversified and dynamic, encompassing multiple functions such as residence, employment, consumption, leisure, and services. Residents' travel chains traverse different spatial units and facilities, and their spatiotemporal distribution directly impacts the efficiency of public resource allocation and quality of life. However, existing research and planning methods largely rely on demographics or land use indicators, lacking in-depth insights into residents' dynamic behaviors. While mobile phone signaling data, LBS trajectories, and POI information have been used in recent years for stop point identification and hotspot analysis, these methods mostly remain at the level of activity space description, failing to provide a systematic matching diagnosis in conjunction with facility supply capacity. Furthermore, although multi-agent simulation has been applied in traffic flow or emergency scenarios, it has not yet been effectively integrated to address the multidimensional patterns and spatial optimization of residents' daily activities. In urban planning practice, facility layout optimization still primarily focuses on static placement and capacity expansion, lacking supply-demand analysis and interactive optimization from a dynamic, holistic perspective, making it difficult to diagnose and correct imbalances in the supply and demand of public service facilities in a timely manner. Summary of the Invention

[0003] Purpose of the invention: The purpose of this invention is to provide a method for identifying, diagnosing, and optimizing the daily activity space of urban residents based on multi-agent supply and demand extrapolation, which significantly improves the efficiency of urban spatial resource allocation and precise governance.

[0004] Technical solution: To achieve the above objectives, the method for identifying, diagnosing, and optimizing the daily activity space of urban residents as described in this invention includes the following steps: (1) Obtain spatial data of the target area, divide the target area into fishing net units, obtain mobile LBS trajectory point data and three-dimensional real scene image data of the current spatiotemporal flow of the target area, perform spatial registration and projection conversion and integrate them into each fishing net unit to form a basic dataset of urban space and residents' travel. (2) Cluster the mobile LBS trajectory points and extract the potential stop point set from the original mobile LBS trajectory points; identify the stop point type based on the stop point ID, the stop point type includes residential point, employment point and flexible activity point; use kernel density estimation to identify hotspot space of the spatial distribution of various activity points; (3) Based on residence, employment and flexible activity points, generate residents’ historical activity-travel chain, calculate the time-segmented transfer probability, and introduce the logarithm of the transfer probability, spatial impedance factor and facility attractiveness into multiple Logit utility functions to correct the transfer probability. Finally, construct a daily activity-travel decision network composed of resident type layer, activity characteristic layer and spatial unit layer. (4) Establish a resident intelligent agent, determine the activities of the resident intelligent agent at different times and the corresponding activity target fishing net unit based on the daily activity-travel decision network, obtain the facility demand index and facility supply index of the fishing net unit, construct the matching degree index through the supply and demand ratio difference and coupling coordination degree, and diagnose the supply and demand imbalance of daily activity space from the dimensions of space, time and type respectively. (5) Taking the diagnosed supply and demand mismatch areas and facility types as the core optimization objects, with the goal of maximizing the average supply and demand matching degree of the whole region, construct an optimization scheme library that includes stock adjustment and incremental planning, conduct scheme deduction and optimization, screen the generated candidate schemes step by step, and output the optimal scheme for daily activity space. (6) Optimize the interactive output feedback of the scheme.

[0005] Optionally, step (1) specifically includes the following steps: (1.1) Obtain vector data of the street boundaries, POI functional facilities, and road traffic network of the target area from the local planning department; divide the area data of the target area into fishing net units; obtain mobile LBS trajectory point data of the current spatiotemporal flow of the population; collect three-dimensional real-scene image data of the target plot; (1.2) After spatial registration and projection transformation, the obtained data is integrated into each fishing net unit to form a basic dataset of urban space and residents' travel.

[0006] Optionally, step (2) specifically includes the following steps: (2.1) The ST-DBSCAN clustering algorithm is used to cluster the mobile phone LBS trajectory points, and the potential stop point set is extracted from the original mobile phone LBS trajectory points. For any two trajectory points and When spatial constraints are met and time constraints At that time, trajectory point Belongs to trajectory point spatiotemporal neighborhood point set When the trajectory point spatiotemporal neighborhood point set When the number of points contained is not less than the minimum point threshold MinPts, the trajectory points Using this as the core, trajectory points that satisfy the spatiotemporal constraints are aggregated into the same cluster of rest points, where... The Euclidean distance between two points. Spatial threshold, This is a time threshold; (2.2) Identify the type of stop point based on the ID of the stop point. Among them, high-frequency stop points located in residential land or commercial and residential land and that meet the characteristics of nighttime activities are identified as residential points. The second highest frequency stop point in non-residential areas is identified as employment points. The remaining stop points are identified as flexible activity points. A buffer zone of a certain distance is set for the stop points. The number of POIs of each type in the buffer zone is counted. According to the POI type, the activity points are further subdivided into consumption type, leisure type and service type. Among them, consumption type is matched with commercial POI, leisure type is matched with cultural and entertainment POI, and service type is matched with education and medical POI. (2.3) Kernel density estimation is used to identify hotspot spaces in the spatial distribution of various activity points. Using fishing net units as statistical units, the kernel density values ​​of residential points, employment points and various types of flexible activity points are counted respectively. The heat of various activity points is divided into three levels: "high, medium and low" by the natural discontinuity method. Areas with a quantity level of "high" are identified as high-density hotspot spaces. The distribution of high-density hotspot spaces is visualized in the geographic information system.

[0007] Optionally, step (3) specifically includes the following steps: (3.1) Based on the residential points, employment points and three types of flexible activity points (consumption, leisure and service) obtained in step (2), the residents' historical activity-travel chain is generated in sequence, the 24-hour activities are divided into multiple time periods, the time period transition probability is calculated using the first-order Markov chain model and smoothed by Laplace. (3.2) The logarithm of the transition probability is introduced into a multinomial Logit utility function along with the spatial impedance factor and facility attractiveness. The impedance factor is based on an exponential decay function of road network distance or travel time, and the attractiveness is obtained by weighting the number, level, and pedestrian flow intensity of POIs. The transition probability is then corrected by combining this with residents' travel preference variables and normalizing using Softmax. As an a priori utility term, it enters the multinomial Logit utility function along with the spatial impedance factor and facility attractiveness: , in From Class point to The shortest path distance of the road network for points of the same type. The facility's attractiveness is calculated by weighting the number of Points of Interest (POIs), their tier, and historical foot traffic intensity. These are the residents' preference variables regarding travel distance, stay duration, and facility type; The space impedance coefficient, The facility attractiveness coefficient, These are the coefficients of the preference characteristic variables; Corrected transition probability The result is obtained through the Softmax function: , in, With state number Consistency is ensured to guarantee complete coverage of residents' potential activity relocation directions; (3.3) Based on the modeling of resident type preferences and the correction of time-based activity transition probabilities, a daily activity-travel decision network is constructed, consisting of a resident type layer, an activity characteristic layer, and a spatial unit layer. The network uses the corrected transition probability as the edge weight to characterize the transition from the current activity at different time periods. Transfer to target activity The selection bias supports the simulation of residents' activity destinations and spatiotemporal distribution in the extrapolation; among them, the resident type layer records the group's gender, age, and income characteristics, and the activity characteristic layer records the type attributes of residence, employment location, and various flexible activity locations, serving as a set of activity transition states. The carrier; the spatial unit layer records the spatial environmental characteristics of different fishing net units.

[0008] Optionally, the specific formula for calculating the time-segmented transition probability using a first-order Markov chain model in step (3.1) is as follows: , in For time period From the inside Class point to Number of transitions for class points The Laplace smoothing parameter is... For the set of activity transition states, This represents the number of states.

[0009] Optionally, step (4) specifically includes the following steps: (4.1) Establish an equal number of resident intelligent agents based on the resident composition within the target area; extract POI functional facility type and geographic location information from the urban space and resident travel basic dataset, and associate them with the corresponding space using the fishing net unit as the supply statistical unit; determine the activities of resident intelligent agents at different times and the corresponding activity target fishing net units based on the daily activity-travel decision network; (4.2) Statistically analyze the decision results of all residents' intelligent agents at all times and in all quantities. The facility demand index of each fishing net unit is obtained by normalizing the number of times each type of facility is selected in each fishing net unit and the number of residents' intelligent agents, so as to realize the simulation of service facility demand. At the same time, the supply capacity of service facilities in each fishing net unit is summarized and normalized according to the facility type to obtain the facility supply index. (4.3) A matching degree index is constructed by using the supply-demand ratio difference and coupling coordination degree to calculate the ratio difference between the facility demand index and the facility supply index, reflecting the degree of overall imbalance; then, the coupling coordination coefficient between the facility demand index and the facility supply index reflects the local adaptation level; the supply-demand matching degree of each fishing net unit and each facility type is obtained by weighting the supply-demand ratio difference and coupling coordination degree, and the supply-demand imbalance of daily activity space is diagnosed from the spatial, temporal and type dimensions respectively; the spatial dimension refers to drawing a heat map of the distribution of fishing net units with a supply-demand matching degree lower than the average at the whole domain level to identify the hot spot clusters of mismatch; the temporal dimension refers to dividing by time period Calculate the supply and demand matching degree of each unit and analyze the supply and demand fluctuation differences during morning and evening peak hours and off-peak hours; the type dimension refers to the average supply and demand matching degree of various facilities, and screen the facility types with the most severe mismatch.

[0010] Optionally, the facility demand index in step (4.2) refers to the index within the research scope. The decision-making results of each resident intelligent agent throughout the entire time period are statistically analyzed, duplicate selections are excluded, and the results of each fishing net unit are calculated. Various types Facilities demand index : , in, For each fishing net unit Time period Select Go The number of residents' smart entities in similar facilities; The total number of intelligent agents among residents within the research scope.

[0011] Optionally, the facility supply index in step (4.2) refers to the index for each fishing net unit. Various types The facility supply capacity is weighted and aggregated, and the facility supply index is obtained after unifying the dimensions. : , in, For facilities Supply capacity; for The capacity conversion factor for different types of facilities is used to unify the capacity dimensions of different facility types. The target range is the entire domain, and the denominator is the entire domain. The maximum supply of such facilities is used to normalize the supply index within the [0,1] interval.

[0012] Optionally, step (5) specifically includes the following steps: (5.1) Taking the supply-demand mismatch areas and facility types diagnosed in step (4) as the core optimization objects, and aiming to maximize the average supply-demand matching degree of the whole region, an optimization scheme library including stock adjustment and incremental planning is constructed. Stock adjustment includes functional transformation and capacity expansion measures, which directly affect the supply capacity or spatial resistance of existing facilities; incremental planning includes site addition and level improvement measures, which improve the supply level by increasing the number of facilities or enhancing attractiveness. Each scheme is preset with optimization cost and potential impact coefficient. (5.2) The reinforcement learning PPO algorithm is used for scheme deduction and optimization: the average supply and demand matching degree and the proportion of optimization cost are used as the state space, and the specific measures in the scheme library are used as the action space. A reward function that satisfies the matching degree improvement and cost control is designed. The strategy is updated stably by limiting the difference between the new and old strategies. During the training process, starting from the initial supply and demand state, the scheme is selected iteratively and the implementation effect is simulated. The strategy is adjusted according to the reward function until convergence. At the same time, multiple simulations are performed to evaluate the stability of the scheme. (5.3) The generated candidate solutions are screened step by step to check Pareto optimality, spatial constraints and scene adaptability, and the optimal solution for daily activity space is output. The step-by-step screening of the generated candidate solutions refers to generating multiple sets of candidate solutions by training the converged PPO strategy, and then screening the Pareto optimal solution set according to the following process: In Pareto optimality, any solution in the solution set... There are no other solutions. Satisfying the average supply and demand matching degree across the entire region And optimize costs Meanwhile, within spatial constraints, the site selection for incremental facilities should be located within the planning-permitted construction area and meet planning management regulations.

[0013] Optionally, step (6) specifically includes the following steps: (6.1) Connect the supply and demand diagnosis results of step (4) and the optimal solution of step (5) to the Dashboard visualization platform and display them on an 8K resolution LCD screen: based on the three-dimensional image of urban space, overlay the heat map of supply and demand matching degree, the bar chart of facility type mismatch and the time series curve of time matching degree. (6.2) Human-machine collaboration is achieved through interconnected interactive devices: Users can immerse themselves in the cloud sand table through an 8K resolution VR all-in-one device, review the supply and demand diagnosis and solution deduction process, and mark the facility adjustment areas under different schemes by color; at the same time, parameter interaction is supported, and users can manually modify the budget threshold and facility capacity limit parameters, and the system calls the PPO algorithm to update the scheme in real time. (6.3) After the user confirms the solution through VR interaction, a PDF report, GIS spatial data and Word solution description are generated.

[0014] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: (1) This invention proposes a method for identifying, diagnosing and optimizing the daily activity space of urban residents by integrating multi-source data, behavioral modeling and intelligent optimization. By introducing multi-agent supply and demand simulation, it can realistically reproduce the activity chain of residents and their travel patterns, and dynamically match and optimize the supply and demand relationship of facilities, providing scientific support for urban governance and public service planning, thereby realizing the transformation from "static configuration" to "dynamic optimization". (2) Based on the supply and demand deduction of multiple agents, the present invention can realize the accurate identification, scientific diagnosis and intelligent optimization of the daily activity space of urban residents, and significantly improve the efficiency of urban spatial resource allocation and precise governance; (3) This invention constructs a multi-layer travel decision network that combines Markov chain model and Logit utility function, combining residents’ migration probability with spatial impedance factor, facility attractiveness and individual preferences. It breaks through the traditional single flow prediction or experience judgment mode, and can accurately depict the dynamic activity chain and spatial distribution pattern of residents in the 24-hour cycle, providing a scientific basis for facility demand simulation. (4) This invention quantifies the supply index and demand index and constructs a matching index to reveal the imbalance characteristics of daily activity space from the dimensions of space, time and type, thus achieving a more refined and operational supply and demand matching diagnosis. (5) This invention introduces reinforcement learning PPO algorithm to build an optimization scheme library for stock adjustment and incremental planning, realizes dynamic deduction and automatic optimization of facility layout and capacity configuration, shortens the traditional planning process that relies on human experience and has a long cycle to an intelligent optimization process that can converge quickly in multiple iterations, and combined with VR interactive display, realizes the interpretability and efficient output of optimization schemes. Attached Figure Description

[0015] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a schematic diagram of the operation of a multi-layer decision network according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the interactive output feedback process in an embodiment of the present invention. Detailed Implementation

[0016] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0017] like Figure 1 As shown, the present invention provides a method for spatial identification, diagnosis, and optimization of daily activities of urban residents based on multi-agent supply and demand extrapolation, comprising the following steps: (1) Integration of basic data on urban space and residents’ travel: acquire spatial data of the target area, divide the target area into fishing net units, acquire mobile LBS trajectory point data and three-dimensional real scene image data of the current spatiotemporal flow of the target area, and integrate them into each fishing net unit after spatial registration and projection conversion to form a basic dataset of urban space and residents’ travel.

[0018] Step (1) specifically includes the following steps: (1.1) Obtain vector data of the street boundaries, POI functional facilities, and road traffic network of the target area from the local planning department; divide the area data of the target area into 100m×100m fishing net units; obtain mobile phone LBS trajectory point data of the current spatiotemporal flow of the population from the local telecommunications bureau; use a surveying drone equipped with a 16K resolution camera module to collect three-dimensional real-scene image data of the target plot; (1.2) After spatial registration and projection transformation, the obtained data is integrated into each fishing net unit to form a basic dataset of urban space and residents' travel.

[0019] (2) Daily activity measurement and hotspot space identification: Cluster the mobile phone LBS trajectory points and extract the potential stop point set from the original mobile phone LBS trajectory points; identify the stop point type based on the stop point ID, which includes residential points, employment points and flexible activity points; use kernel density estimation to identify hotspot space for the spatial distribution of various activity points.

[0020] Step (2) specifically includes the following steps: (2.1) The ST-DBSCAN clustering algorithm is used to cluster the mobile phone LBS trajectory points, and the potential stop point set is extracted from the original mobile phone LBS trajectory points. For any two trajectory points and When spatial constraints are met and time constraints At that time, trajectory point Belongs to trajectory point spatiotemporal neighborhood point set When the trajectory point spatiotemporal neighborhood point set When the number of points contained is not less than the minimum point threshold MinPts, the trajectory points Using this as the core, trajectory points that satisfy the spatiotemporal constraints are aggregated into the same cluster of rest points, where... The Euclidean distance between two points. Spatial threshold, This is a time threshold; (2.2) Identify the type of stop point based on the ID of the stop point. Among them, high-frequency stop points located in residential land or commercial and residential land and that meet the characteristics of nighttime activities are identified as residential points. The second highest frequency stop point in non-residential areas is identified as employment points. The remaining stop points are identified as flexible activity points. A 50m buffer zone is set for the stop points. The number of POIs of each type in the buffer zone is counted. According to the POI type, the activity points are further subdivided into consumption type, leisure type and service type. Among them, consumption type is matched with commercial POI, leisure type is matched with cultural and entertainment POI, and service type is matched with education and medical POI. (2.3) Kernel density estimation is used to identify hotspot spaces in the spatial distribution of various activity points. Using fishing net units as statistical units, the kernel density values ​​of residential points, employment points and various types of flexible activity points are counted respectively. The heat of various activity points is divided into three levels: "high, medium and low" by the natural discontinuity method. Areas with a quantity level of "high" are identified as high-density hotspot spaces. The distribution of high-density hotspot spaces is visualized in the geographic information system.

[0021] (3) Construction of Daily Activity-Travel Decision Network: Based on residence, employment, and flexible activity points, a resident's historical activity-travel chain is generated, and the time-segmented transfer probability is calculated. The logarithm of the transfer probability, along with the spatial impedance factor and facility attractiveness, is introduced into a multinomial Logit utility function to correct the transfer probability. Finally, a daily activity-travel decision network consisting of a resident type layer, an activity characteristic layer, and a spatial unit layer is constructed, such as... Figure 2 As shown.

[0022] Step (3) specifically includes the following steps: (3.1) Based on the residential points, employment points and three types of flexible activity points (consumption, leisure and service) obtained in step (2), the residents' historical activity-travel chain is generated in sequence. The 24-hour activities are divided into 12 time periods with a 2-hour interval. The time period transition probability is calculated using the first-order Markov chain model and smoothed by Laplace. The specific formula for calculating the time-segmented transition probability using a first-order Markov chain model is as follows: , in For time period From the inside Class point to Number of transitions for class points The Laplace smoothing parameter is... For the set of activity transition states, The number of states; (3.2) The logarithm of the transition probability is introduced into a multinomial Logit utility function along with the spatial impedance factor and facility attractiveness. The impedance factor is based on an exponential decay function of road network distance or travel time, and the attractiveness is obtained by weighting the number, level, and pedestrian flow intensity of POIs. The transition probability is then corrected by combining this with residents' travel preference variables and normalizing using Softmax. As an a priori utility term, it enters the multinomial Logit utility function along with the spatial impedance factor and facility attractiveness: , in The shortest path distance in the road network. The facility's attractiveness is calculated by weighting the number of Points of Interest (POIs), their tier, and historical foot traffic intensity. These are the residents' preference variables regarding travel distance, stay duration, and facility type; The space impedance coefficient, The facility attractiveness coefficient, These are the coefficients of the preference characteristic variables; Corrected transition probability The result is obtained through the Softmax function:

[0023] in, With state number Consistency is ensured to guarantee complete coverage of residents' potential activity relocation directions; (3.3) Based on the modeling of resident type preferences and the correction of time-based activity transition probabilities, a daily activity-travel decision network is constructed, consisting of a resident type layer, an activity characteristic layer, and a spatial unit layer. The network uses the corrected transition probability as the edge weight to characterize the transition from the current activity at different time periods. Transfer to target activity The selection bias supports the simulation of residents' activity destinations and spatiotemporal distribution in the extrapolation; among them, the resident type layer records the group's gender, age, and income characteristics, and the activity characteristic layer records the type attributes of residence, employment location, and various flexible activity locations, serving as a set of activity transition states. The carrier; the spatial unit layer records the spatial environmental characteristics of different fishing net units.

[0024] (4) Multi-agent supply and demand simulation and spatial matching diagnosis: Resident agents are established, and the activities and corresponding target fishing net units of resident agents in different time periods are determined based on the daily activity-travel decision network. The facility demand index and facility supply index of the fishing net unit are obtained. The matching degree index is constructed by the supply and demand ratio difference and coupling coordination degree. The imbalance between supply and demand in daily activity space is diagnosed from the dimensions of space, time and type.

[0025] Step (4) specifically includes the following steps: (4.1) Establish an equal number of resident intelligent agents based on the composition of residents within the target area; extract POI functional facility type and geographic location information from the urban space and resident travel basic dataset, and associate them with the corresponding space using 100m fishing net units as supply statistical units; determine the activities of resident intelligent agents at different times and the corresponding activity target fishing net units based on the daily activity-travel decision network.

[0026] (4.2) Statistically analyze the decision-making results of all residents' intelligent agents at all times and in all quantities. The facility demand index of each fishing net unit is obtained by normalizing the number of times each type of facility is selected in each fishing net unit and combining it with the number of residents' intelligent agents, so as to realize the simulation of service facility demand. The facility demand index refers to the demand for facilities within the research scope. The decision-making results of each resident intelligent agent throughout the entire time period are statistically analyzed, duplicate selections are excluded, and the results of each fishing net unit are calculated. Various types Facilities demand index : , in, For each fishing net unit Time period Select Go The number of residents' smart entities in similar facilities; The total number of intelligent agents among residents within the research scope.

[0027] At the same time, the supply capacity of service facilities within each fishing net unit is summarized and normalized according to facility type to obtain the facility supply index. The facility supply index refers to the supply index for each fishing net unit. Various types The facility supply capacity is weighted and aggregated, and the facility supply index is obtained after unifying the dimensions. : , in, For facilities Supply capacity, commercial and cultural entertainment POIs take business area To supply capacity, the number of educational POIs is determined by the number of degree places. To determine supply capacity, the number of beds is taken as the medical POI. To supply capacity; for The capacity conversion factor for different types of facilities is used to unify the capacity dimensions of different facility types. The target range is the entire domain, and the denominator is the entire domain. The maximum supply of such facilities is used to normalize the supply index within the [0,1] interval.

[0028] (4.3) A matching degree index is constructed by using the supply-demand ratio difference and coupling coordination degree to calculate the ratio difference between the facility demand index and the facility supply index, reflecting the degree of overall imbalance; then, the coupling coordination coefficient between the facility demand index and the facility supply index reflects the local adaptation level; the supply-demand matching degree of each fishing net unit and each facility type is obtained by weighting the supply-demand ratio difference and coupling coordination degree, and the supply-demand imbalance of daily activity space is diagnosed from the spatial, temporal and type dimensions respectively; the spatial dimension refers to drawing a heat map of the distribution of fishing net units with a supply-demand matching degree lower than the average at the whole domain level to identify the hot spot clusters of mismatch; the temporal dimension refers to dividing by time period Calculate the supply and demand matching degree of each unit, analyze the supply and demand fluctuation differences during morning and evening peak hours and off-peak hours. Morning and evening peak hours refer to 6:00-10:00 am and 4:00-8:00 pm. The type dimension refers to the average supply and demand matching degree of various facilities, and screen the facility types with the most severe mismatch.

[0029] (5) Generation and deduction of daily activity space optimization schemes: Taking the diagnosed supply and demand mismatch areas and facility types as the core optimization objects, and aiming to maximize the average supply and demand matching degree of the whole area, we will build an optimization scheme library that includes stock adjustment and incremental planning, carry out scheme deduction and optimization, screen the generated candidate schemes step by step, and output the optimal scheme for daily activity space.

[0030] Step (5) specifically includes the following steps: (5.1) Taking the supply-demand mismatch areas and facility types diagnosed in step (4) as the core optimization objects, and aiming to maximize the average supply-demand matching degree of the whole region, an optimization scheme library including stock adjustment and incremental planning is constructed. Stock adjustment includes functional transformation and capacity expansion measures, which directly affect the supply capacity or spatial resistance of existing facilities. Incremental planning includes new locations and grade improvement measures, which improve the supply level by increasing the number of facilities or enhancing attractiveness. Each scheme is preset with optimization cost and potential impact coefficient.

[0031] (5.2) The reinforcement learning PPO algorithm is used for scheme deduction and optimization: the average supply and demand matching degree and the proportion of optimization cost are used as the state space, and the specific measures in the scheme library are used as the action space. A reward function that satisfies the matching degree improvement and cost control is designed. The strategy is updated stably by limiting the difference between the new and old strategies. During the training process, starting from the initial supply and demand state, the scheme is selected iteratively and the implementation effect is simulated. The strategy is adjusted according to the reward function until convergence. At the same time, multiple simulations are performed to evaluate the stability of the scheme.

[0032] (5.3) The generated candidate solutions are screened step by step to check Pareto optimality, spatial constraints, and scene adaptability, and the optimal solution for daily activities is output. The step-by-step screening of the generated candidate solutions refers to generating multiple sets of candidate solutions by training the converged PPO strategy, and screening the Pareto optimal solution set according to the following process: In Pareto optimality, any solution in the solution set is the optimal solution for Pareto optimality. There are no other solutions. Satisfying the average supply and demand matching degree across the entire region And optimize costs Meanwhile, within spatial constraints, the site selection for incremental facilities should be located within the planning-permitted construction area and meet planning management regulations.

[0033] (6) Optimize the interactive output feedback of the scheme.

[0034] like Figure 3 As shown, step (6) specifically includes the following steps: (6.1) Connect the supply and demand diagnosis results of step (4) and the optimal solution of step (5) to the Dashboard visualization platform and display them on an 8K resolution LCD screen: based on the three-dimensional image of urban space, overlay the heat map of supply and demand matching degree, the bar chart of facility type mismatch and the time series curve of time matching degree. (6.2) Human-machine collaboration is achieved through interconnected interactive devices: Users can immerse themselves in the cloud sand table through an 8K resolution VR all-in-one device, review the supply and demand diagnosis and solution deduction process, and mark the facility adjustment areas under different schemes by color; at the same time, parameter interaction is supported, and users can manually modify the budget threshold and facility capacity limit parameters, and the system calls the PPO algorithm to update the scheme in real time. (6.3) After the user confirms the solution through VR interaction, a PDF report, GIS spatial data and Word solution description are generated.

Claims

1. A method for spatial identification, diagnosis, and optimization of daily activities of urban residents based on multi-agent supply and demand extrapolation, characterized in that, Includes the following steps: (1) Obtain spatial data of the target area, divide the target area into fishing net units, obtain mobile LBS trajectory point data and three-dimensional real scene image data of the current spatiotemporal flow of the target area, perform spatial registration and projection conversion and integrate them into each fishing net unit to form a basic dataset of urban space and residents' travel. (2) Cluster the mobile LBS trajectory points and extract the potential stop point set from the original mobile LBS trajectory points; identify the stop point type based on the stop point ID, the stop point type includes residential point, employment point and flexible activity point; use kernel density estimation to identify hotspot space of the spatial distribution of various activity points; (3) Based on residence, employment and flexible activity points, generate residents’ historical activity-travel chain, calculate the time-segmented transfer probability, and introduce the logarithm of the transfer probability, spatial impedance factor and facility attractiveness into multiple Logit utility functions to correct the transfer probability. Finally, construct a daily activity-travel decision network composed of resident type layer, activity characteristic layer and spatial unit layer. (4) Establish a resident intelligent agent, determine the activities of the resident intelligent agent at different times and the corresponding activity target fishing net unit based on the daily activity-travel decision network, obtain the facility demand index and facility supply index of the fishing net unit, construct the matching degree index through the supply and demand ratio difference and coupling coordination degree, and diagnose the supply and demand imbalance of daily activity space from the dimensions of space, time and type respectively. (5) Taking the diagnosed supply and demand mismatch areas and facility types as the core optimization objects, with the goal of maximizing the average supply and demand matching degree of the whole region, construct an optimization scheme library that includes stock adjustment and incremental planning, conduct scheme deduction and optimization, screen the generated candidate schemes step by step, and output the optimal scheme for daily activity space. (6) Optimize the interactive output feedback of the scheme.

2. The method for spatial identification, diagnosis, and optimization of urban residents' daily activities based on multi-agent supply and demand extrapolation as described in claim 1, characterized in that, Step (1) specifically includes the following steps: (1.1) Obtain vector data of the street boundaries, POI functional facilities, and road traffic network of the target area from the local planning department; divide the area data of the target area into fishing net units; obtain mobile LBS trajectory point data of the current spatiotemporal flow of the population; collect three-dimensional real-scene image data of the target plot; (1.2) After spatial registration and projection transformation, the obtained data is integrated into each fishing net unit to form a basic dataset of urban space and residents' travel.

3. The method for spatial identification, diagnosis, and optimization of urban residents' daily activities based on multi-agent supply and demand extrapolation according to claim 2, characterized in that, Step (2) specifically includes the following steps: (2.1) The ST-DBSCAN clustering algorithm is used to cluster the mobile phone LBS trajectory points, and the potential stop point set is extracted from the original mobile phone LBS trajectory points. For any two trajectory points and When spatial constraints are met and time constraints At that time, trajectory point Belongs to trajectory point spatiotemporal neighborhood point set When the trajectory point spatiotemporal neighborhood point set When the number of points contained is not less than the minimum point threshold MinPts, the trajectory points Using this as the core, trajectory points that satisfy the spatiotemporal constraints are aggregated into the same cluster of rest points, where... The Euclidean distance between two points. Spatial threshold, This is a time threshold; (2.2) Identify the type of stop point based on the ID of the stop point. Among them, high-frequency stop points located in residential land or commercial and residential land and that meet the characteristics of nighttime activities are identified as residential points. The second highest frequency stop point in non-residential areas is identified as employment points. The remaining stop points are identified as flexible activity points. A buffer zone of a certain distance is set for the stop points. The number of POIs of each type in the buffer zone is counted. According to the POI type, the activity points are further subdivided into consumption type, leisure type and service type. Among them, consumption type is matched with commercial POI, leisure type is matched with cultural and entertainment POI, and service type is matched with education and medical POI. (2.3) Kernel density estimation is used to identify hotspot spaces in the spatial distribution of various activity points. Using fishing net units as statistical units, the kernel density values ​​of residential points, employment points and various types of flexible activity points are counted respectively. The heat of various activity points is divided into three levels: "high, medium and low" by the natural discontinuity method. Areas with a quantity level of "high" are identified as high-density hotspot spaces. The distribution of high-density hotspot spaces is visualized in the geographic information system.

4. The method for spatial identification, diagnosis, and optimization of urban residents' daily activities based on multi-agent supply and demand extrapolation according to claim 3, characterized in that, Step (3) specifically includes the following steps: (3.1) Based on the residential points, employment points and three types of flexible activity points (consumption, leisure and service) obtained in step (2), the residents' historical activity-travel chain is generated in sequence, the 24-hour activities are divided into multiple time periods, the time period transition probability is calculated using the first-order Markov chain model and smoothed by Laplace. (3.2) The logarithm of the transition probability is introduced into a multinomial Logit utility function along with the spatial impedance factor and facility attractiveness. The impedance factor is based on an exponential decay function of road network distance or travel time, and the attractiveness is obtained by weighting the number, level, and pedestrian flow intensity of POIs. The transition probability is then corrected by combining this with residents' travel preference variables and normalizing using Softmax. As an a priori utility term, it enters the multinomial Logit utility function along with the spatial impedance factor and facility attractiveness: ， in From Class point to The shortest path distance of the road network for points of the same type. The facility's attractiveness is calculated by weighting the number of Points of Interest (POIs), their tier, and historical foot traffic intensity. These are the residents' preference variables regarding travel distance, stay duration, and facility type; The space impedance coefficient, The facility attractiveness coefficient, These are the coefficients of the preference characteristic variables; Corrected transition probability The result is obtained through the Softmax function: ， in, With state number Consistency is ensured to guarantee complete coverage of residents' potential activity relocation directions; (3.3) Based on the modeling of resident type preferences and the correction of time-based activity transition probabilities, a daily activity-travel decision network is constructed, consisting of a resident type layer, an activity characteristic layer, and a spatial unit layer. The network uses the corrected transition probability as the edge weight to characterize the transition from the current activity at different time periods. Transfer to target activity The selection bias supports the simulation of residents' activity destinations and spatiotemporal distribution in the extrapolation; among them, the resident type layer records the group's gender, age, and income characteristics, and the activity characteristic layer records the type attributes of residence, employment location, and various flexible activity locations, serving as a set of activity transition states. The carrier; the spatial unit layer records the spatial environmental characteristics of different fishing net units.

5. The method for spatial identification, diagnosis, and optimization of urban residents' daily activities based on multi-agent supply and demand extrapolation according to claim 4, characterized in that, The specific formula for calculating the time-segmented transition probability using a first-order Markov chain model in step (3.1) is as follows: ， in For time period From the inside Class point to Number of transitions for class points The Laplace smoothing parameter is... For the set of activity transition states, This represents the number of states.

6. The method for spatial identification, diagnosis, and optimization of urban residents' daily activities based on multi-agent supply and demand inference as described in claim 5, characterized in that, Step (4) specifically includes the following steps: (4.1) Establish an equal number of resident intelligent agents based on the resident composition within the target area; extract POI functional facility type and geographic location information from the urban space and resident travel basic dataset, and associate them with the corresponding space using the fishing net unit as the supply statistical unit; determine the activities of resident intelligent agents at different times and the corresponding activity target fishing net units based on the daily activity-travel decision network; (4.2) Statistically analyze the decision results of all residents' intelligent agents at all times and in all quantities. The facility demand index of each fishing net unit is obtained by normalizing the number of times each type of facility is selected in each fishing net unit and the number of residents' intelligent agents, so as to realize the simulation of service facility demand. At the same time, the supply capacity of service facilities in each fishing net unit is summarized and normalized according to the facility type to obtain the facility supply index. (4.3) A matching degree index is constructed by using the supply-demand ratio difference and coupling coordination degree to calculate the ratio difference between the facility demand index and the facility supply index, reflecting the degree of overall imbalance; then, the coupling coordination coefficient between the facility demand index and the facility supply index reflects the local adaptation level; the supply-demand matching degree of each fishing net unit and each facility type is obtained by weighting the supply-demand ratio difference and coupling coordination degree, and the supply-demand imbalance of daily activity space is diagnosed from the spatial, temporal and type dimensions respectively; the spatial dimension refers to drawing a heat map of the distribution of fishing net units with a supply-demand matching degree lower than the average at the whole domain level to identify the hot spot clusters of mismatch; the temporal dimension refers to dividing by time period Calculate the supply and demand matching degree of each unit and analyze the supply and demand fluctuation differences during morning and evening peak hours and off-peak hours; the type dimension refers to the average supply and demand matching degree of various facilities, and screen the facility types with the most severe mismatch.

7. The method for spatial identification, diagnosis, and optimization of urban residents' daily activities based on multi-agent supply and demand inference as described in claim 6, characterized in that, In step (4.2), the facility demand index refers to the index within the research scope. The decision-making results of each resident intelligent agent throughout the entire time period are statistically analyzed, duplicate selections are excluded, and the results of each fishing net unit are calculated. Various types Facilities demand index : ， in, For each fishing net unit Time period Select Go The number of smart entities in similar facilities; The total number of intelligent agents among residents within the research scope.

8. The method for spatial identification, diagnosis, and optimization of urban residents' daily activities based on multi-agent supply and demand inference as described in claim 7, characterized in that, The facility supply index mentioned in step (4.2) refers to the supply index for each fishing net unit. Various types The facility supply capacity is weighted and aggregated, and the facility supply index is obtained after unifying the dimensions. : ， in, For facilities Supply capacity; for The capacity conversion factor for different types of facilities is used to unify the capacity dimensions of different facility types. The target range is the entire domain, and the denominator is the entire domain. The maximum supply of such facilities is used to normalize the supply index within the [0,1] interval.

9. The method for spatial identification, diagnosis, and optimization of urban residents' daily activities based on multi-agent supply and demand extrapolation according to claim 1, characterized in that, Step (5) specifically includes the following steps: (5.1) Taking the supply-demand mismatch areas and facility types diagnosed in step (4) as the core optimization objects, and aiming to maximize the average supply-demand matching degree of the whole region, an optimization scheme library including stock adjustment and incremental planning is constructed. Stock adjustment includes functional transformation and capacity expansion measures, which directly affect the supply capacity or spatial resistance of existing facilities; incremental planning includes site addition and level improvement measures, which improve the supply level by increasing the number of facilities or enhancing attractiveness. Each scheme is preset with optimization cost and potential impact coefficient. (5.2) The reinforcement learning PPO algorithm is used for scheme deduction and optimization: the average supply and demand matching degree and the proportion of optimization cost are used as the state space, and the specific measures in the scheme library are used as the action space. A reward function that satisfies the matching degree improvement and cost control is designed. The strategy is updated stably by limiting the difference between the new and old strategies. During the training process, starting from the initial supply and demand state, the scheme is selected iteratively and the implementation effect is simulated. The strategy is adjusted according to the reward function until convergence. At the same time, multiple simulations are performed to evaluate the stability of the scheme. (5.3) The generated candidate solutions are screened step by step to check Pareto optimality, spatial constraints and scene adaptability, and the optimal solution for daily activity space is output. The step-by-step screening of the generated candidate solutions refers to generating multiple sets of candidate solutions by training the converged PPO strategy, and then screening the Pareto optimal solution set according to the following process: In Pareto optimality, any solution in the solution set... There are no other solutions. Satisfying the average supply and demand matching degree across the entire region And optimize costs Meanwhile, within spatial constraints, the site selection for incremental facilities should be located within the planning-permitted construction area and meet planning management regulations.

10. The method for spatial identification, diagnosis, and optimization of urban residents' daily activities based on multi-agent supply and demand inference as described in claim 1, characterized in that, Step (6) specifically includes the following steps: (6.1) Connect the supply and demand diagnosis results of step (4) and the optimal solution of step (5) to the Dashboard visualization platform and display them on an 8K resolution LCD screen: based on the three-dimensional image of urban space, overlay the heat map of supply and demand matching degree, the bar chart of facility type mismatch and the time series curve of time matching degree. (6.2) Human-machine collaboration is achieved through interconnected interactive devices: Users can immerse themselves in the cloud sand table through an 8K resolution VR all-in-one device, review the supply and demand diagnosis and solution deduction process, and mark the facility adjustment areas under different schemes by color; at the same time, parameter interaction is supported, and users can manually modify the budget threshold and facility capacity limit parameters, and the system calls the PPO algorithm to update the scheme in real time. (6.3) After the user confirms the solution through VR interaction, a PDF report, GIS spatial data and Word solution description are generated.

Images