Intelligent construction method of urban simulation model based on activity flow and complex built environment

Abstract

The application discloses an intelligent construction method of a city simulation model based on activity flow and a complex built environment, relates to the technical field of data processing, and comprises the following steps: constructing a social domain state in layers according to activity observation values, constructing a physical domain state in layers according to environmental attribute values, and splicing the social domain state and the physical domain state into a joint state representation; estimating spatial activity physical behavior parameters of each grid according to the social domain state and the physical domain state, and constructing physical conservation and spatial constraints according to the spatial activity physical behavior parameters; taking the joint state representation as an initial condition, constructing an initial city simulation model of the collaborative evolution of activity states and environmental states by superimposing a physical guiding term of a driving vector field and the spatial activity physical behavior parameters, and solving a city state trajectory; the application improves the representation capability of a city simulation on the dynamic interaction process of crowd activities and a built environment, and further provides a reliable quantitative basis for real decision-making in complex scenarios such as extreme weather and sudden public events.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to an intelligent construction method for urban simulation models based on activity flow and complex built environments. Background Technology

[0002] Urban simulation models are core tools supporting urban planning, traffic management, and emergency response decision-making. Early urban simulations primarily used cellular automata and multi-agent models, simulating urban spatial evolution through pre-defined behavioral rules. With the development of big data technology, the widespread acquisition of multi-source spatiotemporal data such as mobile phone signaling and GPS trajectories has driven the rise of data-driven simulation methods. Deep learning technologies such as graph neural networks and spatiotemporal attention mechanisms have further enhanced the models' ability to represent complex urban systems. In recent years, physical information neural networks have embedded domain prior knowledge into the model training process as constraints, providing a new technical path for the interpretability and physical consistency of urban simulations.

[0003] However, existing urban simulation methods typically model human activities and the built environment separately, failing to effectively reflect their mutual influence during the simulation process. In actual urban operation, changes in human activity directly affect the state of the urban physical environment, while changes in the environmental state, in turn, adjust human behavior. This two-way dynamic relationship is a fundamental law of urban system evolution. Existing methods lack the ability to model this two-way interaction mechanism, preventing urban managers from predicting evacuation routes, areas exacerbated by the heat island effect, or the spread of traffic congestion in advance through simulations when facing complex scenarios such as extreme weather, large-scale events, or sudden public incidents. This leads to real-world decision-making errors such as delayed emergency response and inappropriate resource allocation. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides an intelligent construction method for urban simulation models based on activity flow and complex built environment to solve the problem that the dynamic interaction between activity flow and built environment cannot be effectively represented due to the independent modeling of social domain and physical domain.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides an intelligent construction method for urban simulation models based on activity flow and complex built environments, comprising: S1: Acquire urban activity flow data and built environment data, align and map them to the grid according to a unified spatiotemporal benchmark, and obtain the activity observation values ​​and environmental attribute values ​​of each grid. S2, construct the social domain state in layers based on activity observation values, construct the physical domain state in layers based on environmental attribute values, and concatenate the social domain state and physical domain state into a joint state representation; S3 estimates the spatial activity physical behavior parameters of each grid based on the social domain state and the physical domain state, and constructs physical conservation and spatial constraints accordingly. S4, taking the joint state representation as the initial condition, constructs an initial urban simulation model of the co-evolution of activity state and environmental state by superimposing the driving vector field and the physical guidance term of the physical behavior parameters of spatial activities, and solves the urban state trajectory. S5 calculates the loss based on the urban state trajectory, calibrates the driving vector field parameters and spatial activity physical behavior parameters, and outputs the calibrated urban simulation model.

[0007] Preferably, the method for obtaining the activity observation values ​​and environmental attribute values ​​includes: Mobile phone signaling, GPS trajectory, public transportation travel, and POI visit popularity data are collected as activity flow data; road network topology, building footprint, land cover, land use, infrastructure distribution, surface temperature, and air quality data are collected as built environment data. The activity flow data and built environment data were uniformly converted to a geographic coordinate system, standard time, and fixed time step, and divided into grids; the mobile phone signaling was clustered according to trajectory density and dwell time to extract dwell points and filter drift noise; Activity flow data is statistically summarized by grid to obtain activity observation values ​​for each grid; built environment data is spatially merged by grid to calculate road network density, building coverage, number of POIs and category entropy values ​​for each grid to obtain environmental attribute values ​​for each grid.

[0008] Preferably, the method for constructing the social domain state in a hierarchical manner includes: Individual-level analysis of activity observations is performed using anonymous users as units to extract the spatiotemporal behavioral characteristics of each user and obtain the micro-social domain state; By statistically aggregating the micro-social domain state using grids as the basic unit, the activity distribution and flow characteristics of each grid are obtained, thus yielding the meso-social domain state. By taking cities as units, a global weighted aggregation of the state of the meso-level social domain is obtained to obtain the overall activity pattern characteristics of cities and the state of the macro-level social domain.

[0009] Preferably, the method for constructing the physical domain state in a hierarchical manner includes: By taking a single building or street segment as a unit, the environmental attribute values ​​are analyzed at the element level to extract the spatial physical characteristics of each building and street, and the micro-physical domain state is obtained. By spatially aggregating the microscopic physical domain state using grids as the basic unit, the comprehensive characteristics of the built environment of each grid are obtained, thus yielding the mesoscopic physical domain state. By taking cities as units, the state of the meso-physical domain is globally aggregated to obtain the overall built environment characteristics of cities and the state of the macro-physical domain. The meso-level social domain state and the meso-level physical domain state of each grid are concatenated to form a joint state representation.

[0010] Preferably, the method for obtaining the physical behavior parameters of the space activity includes: Based on the activity flow data, the recovery process of the activity in each grid over time is fitted into an exponential decay form to obtain the activity recovery rate parameter of each grid. Based on the building area, road network density, and infrastructure distribution area in the mesophysical domain of each grid, calculate the spatial carrying capacity parameters of each grid. Based on public transportation travel data and GPS trajectory data, the relationship between inter-grid travel volume and spatial distance is fitted into a gravity attenuation form to obtain the spatial interaction attenuation parameters between grids. The activity recovery rate parameter, spatial carrying capacity parameter, and spatial interaction attenuation parameter are summarized into spatial activity physical behavior parameters for each grid.

[0011] Preferably, the method for constructing physical conservation and spatial constraints includes: Based on the continuous nature of activity over time, an activity conservation constraint is constructed by relating the net inflow rate at the boundary to the changes in activity at each grid, serving as a physical conservation constraint. Environmental capacity constraints are constructed based on the comparison between activity density and spatial carrying capacity parameters of each grid, and distance attenuation constraints are constructed based on the deviation between actual travel volume and spatial distance gravity attenuation relationship between grids, serving as spatial constraints.

[0012] Preferably, the method for constructing the initial urban simulation model includes: Using the joint state representation as the initial condition, time evolution equations for the activity state and the environment state are constructed respectively, and the vector field is constructed through the driving vector field parameterized by the multi-layer fully connected structure. Based on the physical behavior parameters of the space activity, the physical guiding terms of the activity state equation and the physical guiding terms of the environment state equation are constructed. The driving vector field is added to the corresponding physical guiding terms to establish the coupling relationship between the activity state and the environment state. Based on the coupling relationship, a two-way information transmission channel is established at three scales: micro, meso, and macro. This includes an emergent aggregation channel that transmits micro-level individual behavioral decisions to meso and macro levels through statistical aggregation, and a feedback modulation channel that transmits macro-level states down to modulate micro-level individual decisions after decomposing them according to grid weights. By integrating coupling relationships with two-way information transmission channels, an initial urban simulation model is constructed to demonstrate the cross-scale co-evolution of activity states and environmental states.

[0013] Preferably, the method for solving the urban state trajectory includes: Starting with the initial activity state and the initial environmental state, the initial city simulation model is continuously solved within a set time interval, and the evolution trajectory of the activity state and environmental state at any time point is output to obtain the city state trajectory.

[0014] Preferably, the method for calculating loss based on urban state trajectory includes: The activity status and environmental status in the urban state trajectory are compared with the actual observation data to obtain the data fitting loss; Based on the degree to which the urban state trajectory adheres to physical conservation and spatial constraints, the activity conservation deviation, environmental capacity deviation, and distance decay deviation are obtained, and the physical deviation loss is obtained by summarizing them. The scale consistency loss is obtained by considering the deviation between the aggregated states at each scale obtained from the emergent aggregation channel and the direct evolution states at the corresponding scale.

[0015] Preferably, the method for outputting the calibrated urban simulation model includes: The total loss is obtained by weighted summing of the data fitting loss, physical bias loss, and scale consistency loss. Based on the gradient of the total loss with respect to the driving vector field parameters and the physical behavior parameters of the space activity, the driving vector field parameters and the physical behavior parameters of the space activity are iteratively updated until the total loss converges. The converged driving vector field parameters and spatial activity physical behavior parameters are encapsulated to output a calibrated urban simulation model.

[0016] In a second aspect, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, it implements any step of the intelligent construction method for urban simulation models based on activity flow and complex built environments as described in the first aspect of the present invention.

[0017] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the intelligent construction method for urban simulation models based on activity flow and complex built environments as described in the first aspect of the present invention.

[0018] The beneficial effects of this invention are as follows: by constructing a hierarchical state of social domain and physical domain and splicing it into a joint state representation, and by superimposing physical guiding terms and driving vector fields to construct an initial urban simulation model of the co-evolution of activity state and environmental state, the invention realizes bidirectional information transmission between the two at the micro, meso and macro scales, thereby improving the ability of urban simulation to represent the dynamic interaction process between human activities and the built environment, and providing a more reliable quantitative basis for real-world decision-making such as human evacuation, heat island warning and traffic control in complex scenarios such as extreme weather and public emergencies.

[0019] Furthermore, by weighting and summing the data fitting loss, physical bias loss, and scale consistency loss based on the urban state trajectory to obtain the total loss, and iteratively calibrating the model parameters, the simulation results simultaneously meet the requirements of data fitting accuracy and physical conservation constraints, thereby improving the physical credibility and simulation stability of the urban simulation model in long-term simulation scenarios. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a flowchart of the intelligent construction method for urban simulation models based on activity flow and complex built environments in this invention.

[0022] Figure 2 This is a flowchart illustrating the formation of physical behavior parameters for space activities in this invention.

[0023] Figure 3 This is a flowchart of the process for constructing the initial city simulation model in this invention.

[0024] Figure 4 This is a flowchart of the output calibrated urban simulation model in this invention. Detailed Implementation

[0025] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0026] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0027] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0028] Reference Figure 1 , Figure 2 , Figure 3 and Figure 4 This is one embodiment of the present invention, which provides an intelligent construction method for urban simulation models based on activity flow and complex built environments, including the following steps: Methods for obtaining activity observations and environmental attribute values ​​include: The sources of activity stream data include four categories: The first category is mobile signaling data, which obtains anonymized user location records from mobile operators. Each record contains a user anonymity identifier, a timestamp, and base station latitude and longitude coordinates. For example, the time granularity is 15 minutes. The second category is GPS trajectory data, which is obtained from ride-hailing platforms or shared bicycle operators. Each record contains an order identifier, a timestamp, and latitude and longitude coordinates. For example, the time granularity is at the second level. The third category is public transportation travel data, which is obtained from the city's bus and subway card-swiping system. Each record includes anonymized card number, card swiping time, and station identifier. The fourth category is POI access popularity data, which is obtained from Internet location service providers. Each record includes the POI identifier, time period, and number of visits. The sources of built environment data include four categories: The first category is road network topology and building footprint data, which are obtained from open source map platforms and include road nodes and edges, road class, number of lanes, building outlines and building heights; The second category is land cover data, which obtains multi-temporal land cover and urban morphology information from remote sensing satellite imagery; The third category consists of land use classification data and infrastructure distribution data, which are obtained from urban planning departments. The fourth category is surface temperature and air quality data, which are obtained from hourly observation records from meteorological stations and environmental monitoring stations.

[0029] Spatially, all data were uniformly converted to the WGS-84 geographic coordinate system, and the study area was divided using a uniform grid as the basic spatial unit. For example, the grid size was 500 meters × 500 meters. In terms of time, all data is uniformly converted to standard time and discretized with a fixed time step, for example, a time step of 30 minutes.

[0030] Using each user's continuous location records as input, cluster analysis is performed on continuous location points according to trajectory density and dwell time. Location points with small location changes within the same time period and whose duration meets the threshold requirement are grouped into a dwell point. For example, the dwell time threshold is 10 minutes, and records of brief location jumps caused by base station signal switching are filtered out.

[0031] For activity flow data, each grid is used as a statistical unit. The mobile phone signaling dwell point records, GPS trajectory data, public transportation travel data and POI access popularity data falling within the grid range are statistically summarized according to the time step. The number of people entering, leaving and staying in each grid in each time step is calculated to form the activity observation value of each grid.

[0032] For built environment data, each grid is used as a spatial unit. Road network topology, building footprint, land cover, land use classification, and infrastructure distribution data are mapped to the corresponding grid through spatial merging. The road network density, building cover, number of POIs, category entropy, and surface temperature and air quality observations for each grid are calculated to form the environmental attribute values ​​for each grid. For example, if a grid contains 12 catering POIs, 8 commercial POIs, 3 medical POIs, and 5 educational POIs, the POI category entropy value for that grid is calculated based on the proportion of each type of POI to quantify the functional mixing degree of that grid.

[0033] Existing urban simulation methods typically model human activity data and built environment data as independent single-domain variables, lacking the ability to hierarchically organize and correlate social behavioral characteristics and physical spatial characteristics within a unified state space. This results in the inability to effectively capture the interaction between human activities and the built environment during subsequent simulations. Therefore, this invention constructs hierarchical state variable systems for the social and physical domains respectively and concatenates them into a joint state representation, providing structured initial conditions for modeling the co-evolution of activity and environmental states. The specific steps are as follows: Methods for constructing social domain states in a hierarchical manner include: Based on the activity observations of each grid and with each anonymous user as the basic unit, individual-level analysis is performed on the mobile phone signaling station records, GPS trajectory data and public transportation travel data in the activity flow data. Spatiotemporal behavioral characteristics such as the current location grid, current activity type, previous destination grid and travel mode of each user in each time step are extracted to form the micro-social domain state. The activity type is inferred based on the distribution of POI categories and dwell time in the user's grid. For example, if a user stays in a grid with residential POIs as the dominant function for more than 6 hours in a certain time step, the activity type of the user in the current time step is inferred to be "staying at home". If a user appears alternately between two grids in multiple consecutive time steps, the user's travel mode is determined based on public transportation travel data or GPS trajectory data between the two grids.

[0034] Using the micro-social domain state as input and each grid as the basic unit, the micro-social domain state of all anonymous users falling into the grid is statistically aggregated to calculate the total population density, population distribution of each activity type, net inflow rate and net outflow rate of the grid in the current time step, thus forming the meso-social domain state. The total population density is the total number of users staying in the grid at the current time step divided by the grid area; the distribution of users by activity type is the statistical number of users in the grid for various activity types such as home, work, commuting, shopping, and entertainment; the net inflow rate is the difference between the number of users entering and leaving the grid at the current time step divided by the time step length, and the net outflow rate is the difference between the number of users leaving and entering the grid at the current time step divided by the time step length (or directly take the opposite of the net inflow rate).

[0035] Using the meso-level social domain state of each grid as input, the meso-level social domain state of all grids is globally weighted and aggregated to calculate the total urban population, the population ratio of each functional area of ​​the city, and the overall urban activity intensity index, thus forming the macro-level social domain state. The city's total population is the sum of the total number of users currently at the current time step across all grids; the population proportion of each functional area is the ratio of the sum of the activity densities of each functionally labeled grid to the total activity volume of the city; the overall urban activity intensity index is the weighted average of the absolute values ​​of the net inflow rates of all grids, with the weights determined based on the area of ​​each grid. For example, if a city is divided into 800 grids, and the grids are categorized into residential, commercial, industrial, and public service functions based on POI distribution, the sum of the activity densities of commercial functional grids accounts for 35% of the total urban activity, residential functional grids 45%, public service functional grids 12%, and industrial functional grids 8%, with the sum of the population proportions of each functional area being 100%.

[0036] Methods for constructing physical domain states in a hierarchical manner include: Based on the environmental attribute values ​​of each grid, and taking a single building or street segment as the basic unit, element-level analysis is performed on the road network topology, building footprint, land cover, and infrastructure distribution data. This extracts the building height, surface area, and building material type for each building, as well as the street width and tree cover rate for each street segment, forming a micro-physical domain state. For example, if a building has a height of 24 meters, a base area of ​​800 square meters, and is constructed of concrete, then the micro-physical domain state of that building includes the aforementioned three spatial physical characteristics.

[0037] Using the microphysical domain state as input and each grid as the basic unit, the microphysical domain states of all buildings and street segments falling within the grid range are spatially aggregated, and the average building height, building volume ratio, road network density, green space coverage, surface temperature and air quality index of the grid are calculated to form the mesophysical domain state. Among them, the building volume ratio is the sum of the total building area of ​​all buildings in the grid divided by the grid area; the green coverage ratio is the green area in the grid divided by the grid area; the road network density is the total paved area of ​​all roads in the grid (or the total area occupied by roads) divided by the grid area; the surface temperature and air quality index are taken from the hourly observation records of meteorological stations and environmental monitoring stations and mapped to the corresponding grid through spatial interpolation methods.

[0038] Using the mesoscopic physical domain state of each grid as input, the mesoscopic physical domain state of all grids is globally aggregated to calculate the city's average surface temperature, urban heat island intensity, city's average air quality index, and the city's overall green space ratio, thus forming the macroscopic physical domain state. The urban average surface temperature is the area-weighted average of the surface temperatures of all grids. Specifically, it is calculated as follows: the effective area of ​​each grid within the urban area is calculated as the proportion of the total effective area of ​​all grids, which is then used as the area weight of that grid. The surface temperature of each grid is multiplied by its corresponding area weight and summed to obtain the urban average surface temperature. The urban heat island intensity is the difference between the average surface temperature of the urban built-up area and the average surface temperature of the urban suburbs. The overall urban green space ratio is the sum of the green space areas of all grids divided by the total area of ​​the urban built-up area.

[0039] Taking each grid as a unit, the meso-level social domain state vector and the meso-level physical domain state vector of the grid are sequentially concatenated according to the feature dimensions. That is, the features such as total population density, population distribution of each activity type, net inflow rate and net outflow rate in the meso-level social domain state vector are sequentially arranged with the features such as average building height, building volume ratio, road network density, green space coverage, surface temperature and air quality index in the meso-level physical domain state vector, and merged into a higher-dimensional feature vector to form the joint state representation of the grid. For example, if the meso-level social domain state vector dimension of a certain grid is 5 and the meso-level physical domain state vector dimension is 6, then the dimension of the joint state representation vector of the grid after concatenation is 11. The joint state representation simultaneously includes the activity distribution and flow characteristics of the grid as well as the comprehensive characteristics of the built environment.

[0040] Through the above-mentioned hierarchical construction and splicing operations, the social domain state and the physical domain state are fully expressed at the micro, meso and macro scales, respectively. The joint state representation of each grid simultaneously includes the distribution and flow characteristics of population activities in the social domain and the built environment spatial physical characteristics of the physical domain, realizing the structured association between the social domain and the physical domain in the same state space.

[0041] Methods for obtaining physical behavior parameters of space activities include: Based on the activity observations of each grid, historical activity flow data before and after a sudden event or abnormal disturbance in the city are selected as the fitting input. For example, continuous activity observations of three months before and three months after the disturbance are selected. For each grid, using the time-varying sequence of the grid's activity after the disturbance as input, the activity recovery process is fitted into an exponential decay form. The specific fitting method is as follows: The average normal activity of the grid before the disturbance As the activity baseline, calculate the number of seconds after the disturbance occurs. Actual activity at each time step With activity baseline The amount of activity deviation (or unrecovered amount) ; Establish deviation over time step The exponential decay model of change: ; in, It is the maximum activity deviation at the initial moment of the disturbance. This is the parameter representing the activity recovery rate of the grid to be estimated; The sum of squared residuals between the actual observed deviations at each time step and the predicted values ​​of the fitted model is minimized using the least squares method or a nonlinear fitting algorithm (such as the LM algorithm). ; The estimated value obtained by solving That is, it serves as the active recovery rate parameter for this grid; in, The base of the natural logarithm (Euler's number) is approximately 2.71828. The optimization objective is to adjust the parameters to be solved. and Find the parameter solution that minimizes the sum of squared residuals by taking the values ​​of ; For example, the active baseline of a grid before the disturbance occurs. =100% If the activity of a certain grid is 40% of the normal level at the first time step after the disturbance (i.e., a deviation of 60%), recovers to 65% at the sixth time step (a deviation of 35%), and recovers to 92% at the 24th time step (a deviation of 8%), then the activity recovery rate parameter of the grid can be obtained by fitting the above recovery sequence. The larger the value of the activity recovery rate parameter, the faster the activity of the grid recovers.

[0042] Based on the mesoscopic physical domain state of each grid, the spatial carrying capacity parameters of each grid are calculated according to the total building area, road network density and infrastructure distribution area of ​​each grid. The specific calculation method is as follows: multiply the total building area, road network area and public service facility area of ​​each grid by the corresponding weight coefficient and then sum them to obtain the spatial carrying capacity parameter of the grid. The weight coefficient is obtained by fitting the regression relationship between historical activity observations and physical environment characteristics. The total building area is calculated by multiplying the building footprint area by the average number of floors; the road network area is calculated by multiplying the road network density by the grid area; and the public service facility area is extracted from the infrastructure distribution data. For example, if the total building area of ​​a grid is 50,000 square meters, the road network area is 8,000 square meters, and the public service facility area is 5,000 square meters, with corresponding weighting coefficients of 0.6, 0.3, and 0.1 respectively, then the spatial carrying capacity parameter of the grid is 50,000×0.6+8,000×0.3+5,000×0.1=32,900. The larger the spatial carrying capacity parameter value, the higher the maximum activity density that the grid can support.

[0043] Based on public transportation travel data and GPS trajectory data, the historical travel volume between each grid pair is statistically analyzed, and the spatial distance between the center points of each grid is calculated. The relationship between the travel volume between grids and the spatial distance is fitted into a gravity decay form. The specific fitting method is as follows: using the actual travel volume between each grid pair as the dependent variable, and the product of the population size of the two grids and the negative exponential function of the spatial distance between the grid center points as the independent variable, a gravity attenuation model is established: ; Note: This expression uses the exponential gravity decay model commonly used in the field of spatial interaction; in, For grid With grid The actual number of trips between them; and They are grids and grid Population size (which can be taken from the total number of users staying in each grid or the number of permanent residents). For grid With grid Spatial distance between center points It is a negative exponential function of spatial distance; The proportional adjustment coefficient is an example of a constant in the range of 0.05-0.20 (specifically calculated by summing the total daily average travel volume of the entire region by dividing the product of the population size of each grid pair and the product of the negative exponential function of spatial distance). is the base of the natural logarithm; This refers to the spatial interaction decay parameter to be estimated; Furthermore, by taking the natural logarithm of both sides of the above gravity decay model, it is transformed into a linear regression equation: ; in, It is the natural logarithm operator; Summarized as follows: ; in, = constant term; With each grid pair As the dependent variable, the spatial distance between grids As the independent variable, ordinary least squares (OLS) is used to perform a univariate linear regression fit on all grids within the global or local region to minimize the estimated sum of squared residuals: ; The absolute value of the negative value of the regression coefficient obtained by solving is the spatial interaction decay parameter. (Alternatively, the parameters can be obtained by directly using the nonlinear least squares method or the Levenberg-Marquardt algorithm to perform curve fitting on the original nonlinear gravity model.) ); Spatial interaction attenuation parameters This reflects the rate at which the number of rows decreases as the distance between grids increases; the larger the value of the spatial interaction attenuation parameter, the faster the number of rows decreases as the distance increases. For example, if the spatial distance between the center points of two grids When the distance increases from 1 km to 2 km (distance increment is 1 km), the fitted spatial interaction decay parameter... When =0.5, because The travel volume between two grids decreases to about 37% compared to grid pairs that are 1 kilometer apart.

[0044] Taking each grid as a unit, the activity recovery rate parameter and spatial carrying capacity parameter corresponding to the grid are bound one-to-one with the grid identifier as the index. Then, the spatial interaction attenuation parameters between all grid pairs are associated and bound with the combination of the starting grid identifier and the ending grid identifier as the index. Finally, the three types of parameters are aligned according to the grid identifier and merged into the same parameter set to form the spatial activity physical behavior parameter set of each grid.

[0045] Existing urban simulation methods typically rely solely on data-driven statistical fitting when embedding physical laws into models. They lack the ability to embed urban activity conservation laws, spatial carrying capacity constraints, and distance decay laws into neural differential equations as explicit physical guiding terms. This leads to physically unreasonable anomalous emergence or out-of-bounds distributions of activity density during simulation. Therefore, this invention constructs physical conservation and spatial constraints and transforms them into physical guiding terms embedded in the time evolution equations of activity and environmental states, establishing explicit coupling between data-driven approaches and physical priors. The specific steps are as follows: Methods for constructing physical conservation and spatial constraints include: Based on the state of the social domain in each grid, activity conservation constraints are constructed according to the continuity of urban activity during the time evolution process. The physical meaning of the conservation of activity is: in the absence of large-scale external population migration, the total activity in the urban area remains constant during the time evolution. That is, the increase in activity in a certain grid must come from the decrease in activity in adjacent grids, and the change in activity in any grid at any time step is equal to the net inflow at the boundary of that grid. The specific construction method is as follows: the difference between the activity amount of each grid in the current time step and the activity amount in the previous time step is taken as the activity change of the grid. The difference between the number of people entering and leaving the grid in the current time step is taken as the net inflow at the boundary. The deviation between the sum of the activity changes of all grids and the sum of the net inflow at the boundary is calculated. The ratio of this deviation to the initial total activity of the city is taken as the quantitative indicator of the activity conservation constraint.

[0046] For environmental capacity constraints, based on the activity density in the meso-social domain state of each grid and the spatial carrying capacity parameter of each grid, the activity density and spatial carrying capacity parameter of each grid at the current time step are compared grid by grid. The ratio of the excess amount of activity density exceeding the spatial carrying capacity parameter to the spatial carrying capacity parameter is calculated as the quantitative index of environmental capacity constraints. When the activity density does not exceed the spatial carrying capacity parameter, the quantitative index is zero. For example, if the spatial carrying capacity parameter of a certain grid corresponds to a maximum activity density of 800 people per square kilometer, and the actual activity density of the grid at the current time step is 850 people per square kilometer, then the environmental capacity constraint quantification index of the grid is (850-800) / 800=6.25%.

[0047] For distance decay constraints, based on the actual travel volume between each grid pair obtained from public transportation travel data and GPS trajectory data statistics, and the spatial interaction decay parameters, the theoretical travel volume between each grid pair is calculated according to the gravity decay form (the calculation method is: multiply the proportional adjustment coefficient, the product of the population size of the two grids, and the negative exponential function value of the spatial distance between the center points of the two grids to obtain the theoretical travel volume of the grid pair). The ratio of the absolute value of the difference between the actual and theoretical travel volumes of each grid pair to the product of the population sizes of the two grids is used as a quantitative indicator of the distance decay constraint. For example, if the product of the population sizes of a grid pair is 50,000, the spatial distance between the grid center points is 3 kilometers, and the spatial interaction decay parameter is 0.5, then the theoretical travel volume of this grid pair is proportional to the product of 50,000 and the negative exponential function value. The absolute value of the difference between the actual travel volume of this grid pair and the theoretical travel volume calculated based on the above parameters is divided by the product of the population sizes of the two grids to obtain the quantitative indicator of the distance decay constraint for this grid pair. The closer the quantitative indicator value of the distance decay constraint is to zero, the higher the degree of agreement between the actual travel distribution and the gravity decay law between the grid pairs, and the better the distance decay constraint is satisfied.

[0048] The methods for constructing initial urban simulation models include: Using the joint state representation of each grid as the initial condition, and based on the relationship between the activity observations and environmental attribute values ​​of each grid over time, the time evolution equations of the activity state and the environmental state are constructed respectively.

[0049] The activity state time evolution equation uses the activity observations of each grid at the current time step, including the activity density, inflow rate, and outflow rate of each grid, arranged in grid identifier order to form an activity state vector. The derivative of the activity state vector with respect to time is the dependent variable, and the concatenation result of the activity state vector at the current time step and the environmental state vector composed of the surface temperature, air quality index, and green coverage rate of each grid arranged in grid identifier order is the independent variable. The right-hand side of the equation is formed by adding the driving vector field and the physical guidance term. The driving vector field is responsible for capturing the nonlinear evolution law driven by data, and the physical guidance term is responsible for applying physical conservation and spatial constraints.

[0050] The environmental state time evolution equation uses the environmental attribute values ​​of each grid at the current time step, including the surface temperature, air quality index and green coverage of each grid, arranged in the order of grid identification to form an environmental state vector. The derivative of the environmental state vector with respect to time is the dependent variable, and the concatenation result of the environmental state vector and the activity state vector at the current time step is the independent variable. Similarly, the right-hand side of the equation is formed by adding the driving vector field and the physical guidance term. The two equations are interconnected through coupling terms in their respective physical guidance terms. The external driving term of the active state time evolution equation takes the environmental state vector as input, and the active driving term of the environmental state time evolution equation takes the active state vector as input, forming a two-way coupling structure.

[0051] For the temporal evolution equations of the activity state and the environment state, a driving vector field is constructed using a multi-layer fully connected structure parameterized driving vector field. Taking the joint state representation vector of each grid as input, after layer-by-layer linear transformation and nonlinear activation processing, the output layer maps it to the rate of change of the activity state or the rate of change of the environment state at the current time step. For example, the activity state driving vector field consists of a 4-layer fully connected structure, with each layer containing 256 neurons. Adjacent layers are connected by nonlinear activation functions. The learnable parameters of the driving vector field are calibrated during subsequent loss calculation and iterative update.

[0052] Based on the set of physical behavior parameters of spatial activities in each grid, physical guiding terms for the activity state time evolution equation and the environmental state time evolution equation are constructed respectively. The physical guiding terms for the activity state time evolution equation include three terms: the first term is the activity decay guiding term, which is constructed as follows: taking the activity recovery rate parameter and the spatial carrying capacity parameter of each grid as input, the ratio of the current activity density to the maximum activity density corresponding to the spatial carrying capacity parameter is calculated for each grid. The product of the activity recovery rate parameter and the current activity density is multiplied by the complement of this ratio to obtain the value of the activity decay guiding term for that grid. When the grid activity density is lower than the maximum activity density corresponding to the spatial carrying capacity parameter, the activity decay guiding term drives the grid activity density to regress to the activity baseline according to the exponential decay law. When the grid activity density is close to the maximum activity density corresponding to the spatial carrying capacity parameter, the activity decay guiding term automatically inhibits further growth of the activity density.

[0053] The second term is the spatial diffusion guiding term. The construction process of the spatial diffusion guiding term is as follows: taking the spatial interaction attenuation parameter between each grid pair and the spatial distance between grids as input, calculate the activity density difference between each grid and all its adjacent grids. Multiply the activity density difference between each adjacent grid by the product of the corresponding spatial interaction attenuation parameter and the negative exponential function of the inter-grid distance, and then sum them to obtain the value of the spatial diffusion guiding term for that grid. The spatial diffusion guiding term drives the activity to diffuse from high-density grids to low-density grids, and the diffusion intensity decreases as the spatial distance between grids increases.

[0054] The third term is the external driving term, which is constructed as follows: Using the changes in surface temperature and air quality index of each grid as output by the environmental state time evolution equation at the current time step as input, for each grid, the changes in surface temperature and air quality index are multiplied by the corresponding environmental impact weight coefficients and then summed. The negative value of the summation result is used as the value of the external driving term for that grid. The environmental impact weight coefficients reflect the relative inhibitory strength of surface temperature and air quality changes on activity inflows. The values ​​are derived from the fitting results of the regression relationship between historical activity observations and environmental state changes. For example, the surface temperature change weight coefficient is 0.6, and the air quality index change weight coefficient is 0.4. When the surface temperature of a grid increases or the air quality index deteriorates, the external driving term generates a negative driving force to inhibit the activity inflow of that grid.

[0055] The physical guide term of the environmental state time evolution equation includes two terms: the first term is the environmental recovery guide term, which is constructed as follows: the difference between the environmental state observation value at the current time step of each grid and the natural background state baseline value of that grid is used as input. This difference is multiplied by the environmental recovery rate coefficient and then negative, to obtain the value of the environmental recovery guide term for that grid. The natural background state baseline value is taken from the average of historical observation records of meteorological stations and environmental monitoring stations during periods without human activity interference. The environmental recovery rate coefficient is derived from the fitting results of the historical environmental state recovery process. The environmental recovery guide term drives the environmental state of each grid to regress to the natural background state when there is no external disturbance.

[0056] The second term is the activity-driven term, which is constructed as follows: Using the activity density of each grid at the current time step as input, the activity density is multiplied by the activity-environment coupling strength coefficient to obtain the activity-driven term value for that grid. The activity-environment coupling strength coefficient reflects the driving strength of a unit activity density on the environmental state. Its value is derived from the fitting results of the regression relationship between historical activity density and environmental state changes. For example, if historical data for a certain grid shows that for every 100 people per square kilometer increase in activity density, the average surface temperature rises by 0.3 degrees Celsius, then the activity-environment coupling strength coefficient for that grid is obtained by fitting this regression relationship. The activity state driving vector field is added to the physical guidance term of the activity state time evolution equation, and the environmental state driving vector field is added to the physical guidance term of the environmental state time evolution equation to establish the coupling relationship between the activity state and the environmental state.

[0057] Based on the coupling relationship between the activity state and the environmental state, a two-way information transmission channel is established at three scales: micro, meso, and macro. Emergent aggregation channels reflect the bottom-up emergent pattern of urban activities, that is, the statistical superposition of micro-level individual behaviors spontaneously forms a meso-level grid-level activity distribution, which in turn forms a macro-level urban activity pattern. The process of establishing the emergent aggregation channel is as follows: Define the aggregation function from micro to meso level, take each grid as the statistical unit, take the micro-individual behavior decisions of all anonymous users falling within the grid range as input in each time step, and count the number of users entering, leaving and staying in the grid respectively; Dividing the difference between the number of users entering and the number of users leaving by the grid area yields the change in grid-level activity density for that grid. Dividing the number of users entering by the time step yields the inflow rate, and dividing the number of users leaving by the time step yields the outflow rate, thus forming the updated value of the meso-level social domain state. Define aggregation functions from meso to macro levels, and perform the following three calculations on the meso-level social domain state update values ​​of all grids: using the proportion of each grid area to the total urban built-up area as the area weight, perform area-weighted summation on the activity density update values ​​of all grids to obtain the total urban activity; The population proportion of each functional area of ​​the city is calculated by using the ratio of the sum of the activity density update values ​​of each functional label grid to the total activity of the city. The absolute value of the difference between the inflow rate and outflow rate of all grids is used as input, and a weighted average is performed using the area weight of each grid as a weighting coefficient to obtain the overall urban activity intensity index. The three indicators together constitute the updated value of the macro-level social domain state. For example, if a grid area is 0.25 square kilometers and the total built-up area of ​​the city is 200 square kilometers, then the area weight of the grid is 0.25 / 200 = 0.00125. The area weight is directly calculated from the grid division result. In each time step, the meso-level social domain state aggregation value obtained by the micro-to-meso-level aggregation function of the emergence aggregation channel, and the macro-level social domain state aggregation value obtained by the meso-to-macro-level aggregation function, will be saved and recorded along with the meso-scale and macro-scale activity state evolution values ​​directly output by the activity state time evolution equation at the corresponding time step, for use in subsequent scale consistency loss calculations.

[0058] The feedback modulation channel embodies the top-down constraint mechanism of macro-urban status on micro-individual decision-making, that is, the process by which macro-urban activity patterns guide micro-individual destination selection through grid-level modulation factors. The process of establishing the feedback modulation channel is as follows: Define the decomposition function from macro to meso, and calculate the normalized weight based on spatial distance and the normalized weight based on functional importance for each grid. The normalized weights based on spatial distance are obtained by normalizing all grids after taking the negative exponent of the Euclidean distance between the center point of each grid and the center point of the city. For example, the distance decay coefficient is taken as the average of the spatial interaction decay parameters of each grid. The normalized weights based on functional importance are obtained by normalizing all grids after using the weighted sum of the POI category entropy value and the historical mean of activity density for each grid as the functional importance score. For example, the weight coefficient of POI category entropy value is 0.4 and the weight coefficient of historical mean of activity density is 0.6. The values ​​of the two weight coefficients are derived from the fitting results of the regression relationship between historical activity observations and functional importance. The average of the distance normalization weight and the functional importance normalization weight of each grid is taken to obtain the comprehensive weight of each grid. The total urban activity in the macro-social domain state update value is multiplied by the comprehensive weight of each grid to obtain the grid-level modulation factor assigned to each grid. The grid with the higher the comprehensive weight, the larger the grid-level modulation factor it receives. Define gating functions from meso to micro levels. With the grid-level modulation factor of each grid as input, calculate the basic utility value of each candidate destination for each anonymous user falling within the grid range, based on the spatial distance between the user's current location and each candidate destination grid and the activity density of the candidate destination grid. The basic utility value is positively correlated with the activity density of the candidate destination grid and negatively correlated with the spatial distance. The grid-level modulation factor of the grid is added to the basic utility value of each candidate destination to obtain the modulated destination selection utility function value, so that changes in the macro state affect the destination selection behavior of micro individuals through the grid-level modulation factor.

[0059] The time evolution equations of the activity state, the time evolution equation of the environment state, the physical guiding terms of the time evolution equations of the activity state, the physical guiding terms of the time evolution equations of the environment state, the emergent aggregation channel, and the feedback modulation channel are integrated into a unified solvable framework. The joint state representation of each grid is used as the initial input, and the set time interval and time step are used as the solution parameters to form an initial urban simulation model of cross-scale co-evolution of the activity state and the environment state.

[0060] By integrating the aforementioned physical guiding terms with the driving vector field, and the emergent aggregation channel with the feedback modulation channel, the initial urban simulation model, based on the data-driven nonlinear evolution capability, also possesses the physical constraints of activity conservation, spatial carrying capacity limitation, and distance decay law, and realizes the two-way dynamic correlation between micro-level individual behavior, meso-level grid activity distribution, and macro-level urban activity pattern.

[0061] Methods for solving urban state trajectories include: Based on the initial urban simulation model, and taking the activity state vector and environmental state vector in the joint state representation of each grid as the starting point for the solution, the co-evolution process of the activity state and environmental state is continuously solved step by step within the set time interval and time step. Within each time step, using the activity state vector and environment state vector of the current time step as input, the micro-to-meta-level aggregation function and the meta-to-macro-level aggregation function of the emergent aggregation channel are executed sequentially to update the meta-level social domain state and the macro-level social domain state. Then, the macro-to-meta-level decomposition function and meta-to-micro-level gating function of the feedback modulation channel are executed to update the grid-level modulation factor of each grid and the destination selection utility function value of each anonymous user; Input the activity state vector and environment state vector of the current time step into the activity state time evolution equation and the environment state time evolution equation, respectively, to calculate the activity state change rate and the environment state change rate. Add the activity state change rate to the activity state vector of the current time step, and add the environment state change rate to the environment state vector of the current time step to obtain the activity state vector and the environment state vector of the next time step, respectively. Repeat the above process until the entire set time interval is covered. For example, if the set time interval is 72 hours and the time step is 30 minutes, the solution process will execute 144 time steps.

[0062] The activity state vectors and environmental state vectors of all time steps within a set time interval are arranged in chronological order to form the activity state and environmental state evolution trajectory of each grid, thus obtaining the urban state trajectory. For example, if the study area is divided into 800 grids, the time interval is 72 hours, and the time step is 30 minutes, the urban state trajectory contains a complete evolution record of activity density, inflow rate, outflow rate, surface temperature, air quality index, and green coverage of 800 grids within 144 time steps.

[0063] Methods for calculating loss based on urban state trajectories include: The simulation output activity state is calculated using the activity density, inflow rate, and outflow rate of each grid in the urban state trajectory at each time step. The actual activity observation values ​​of each grid obtained from mobile phone signaling data, GPS trajectory data, and public transportation travel data at the corresponding time step are used as the real observation data. The mean square error between the simulation output activity state and the real observation data is calculated. The simulation output environmental state is calculated using the surface temperature, air quality index, and green coverage rate of each grid in the urban state trajectory at each time step. The hourly observation records obtained from meteorological stations and environmental monitoring stations at the corresponding time step are used as the real observation data. The mean square error between the simulation output environmental state and the real observation data is calculated. The mean square error of the activity state and the mean square error of the environmental state are multiplied by their respective weighting coefficients and then summed to obtain the data fitting loss. The weighting coefficients are derived from the prior judgment of the importance of the prediction accuracy of the activity state and the environmental state. For example, the weighting coefficient of the activity state is 0.6 and the weighting coefficient of the environmental state is 0.4.

[0064] Using the activity state vector at each time step in the city state trajectory as input, the mean of all time steps is calculated according to the quantitative index calculation method of the activity conservation constraint to obtain the activity conservation deviation; the mean of the environmental capacity deviation is calculated for each grid at each time step according to the quantitative index calculation method of the environmental capacity constraint; the mean of the distance decay deviation is calculated for each grid pair at each time step according to the quantitative index calculation method of the distance decay constraint; the physical deviation loss is obtained by multiplying the activity conservation deviation, environmental capacity deviation, and distance decay deviation by their respective weight coefficients and then summing them. For example, the weight coefficient of the activity conservation deviation is 0.4, the weight coefficient of the environmental capacity deviation is 0.4, and the weight coefficient of the distance decay deviation is 0.2.

[0065] Using the activity state vector at each time step in the urban state trajectory as input, the meso-scale social domain state aggregation value obtained by statistically aggregating the micro-to-meso-scale aggregation function of the emergence aggregation channel on the micro-scale social domain state is compared with the meso-scale activity state evolution value directly output by the activity state time evolution equation at the corresponding time step on a grid-by-grid and time-by-time basis. The mean square error between the two is calculated to obtain the meso-scale consistency deviation. The total urban activity, population proportion of each functional area of ​​the city, and overall urban activity intensity index are obtained by weighting and aggregating the meso-to-macro aggregation function of the emerging aggregation channel with the meso-social domain state update value. This is compared with the macro-scale activity state inference value obtained by adding the activity state change rate of each grid output by the activity state time evolution equation at the current time step to the macro-social domain state of the previous time step. The mean square error between the two is calculated to obtain the macro-scale consistency deviation. The scale consistency loss is obtained by multiplying the mesoscale consistency deviation and the macroscale consistency deviation by their respective weighting coefficients and then summing them. The weighting coefficients reflect the relative importance of the consistency of the mesoscale and macroscale scales in the overall calibration objective. The values ​​are derived from the prior judgment of the simulation accuracy requirements at different scales. For example, the weighting coefficient for the mesoscale consistency deviation is 0.6 and the weighting coefficient for the macroscale consistency deviation is 0.4. The reason why the weighting coefficient for the mesoscale scale is higher than that for the macroscale scale is that the accuracy of the activity distribution at the mesoscale grid level has a more direct impact on the urban simulation results.

[0066] Methods for outputting calibrated urban simulation models include: The total loss is obtained by multiplying the data fitting loss, physical deviation loss, and scale consistency loss by their respective total loss weighting coefficients and then summing them. The total loss weighting coefficients reflect the relative importance of the three types of losses in the overall calibration objective. The values ​​are derived from a comprehensive trade-off between the requirements of urban simulation accuracy and physical consistency. For example, the weighting coefficient for data fitting loss is 0.5, the weighting coefficient for physical deviation loss is 0.3, and the weighting coefficient for scale consistency loss is 0.2.

[0067] Based on the total loss, the gradients of the total loss with respect to the learnable parameters of the activity state driving vector field, the learnable parameters of the environment state driving vector field, and the activity recovery rate parameter, spatial carrying capacity parameter, and spatial interaction decay parameter in the set of physical behavior parameters of each grid space activity are calculated. All the above parameters are updated according to the gradient direction, so that the total loss decreases along the gradient descent direction. The iterative process of forward solving the city state trajectory, calculating the total loss, calculating the gradient, and updating the parameters is repeated until the total loss no longer decreases significantly, at which point the total loss is considered converged. For example, the initial learning rate is 0.001, and after every 50 iterations, the learning rate is multiplied by 0.9 to decay, ensuring stable convergence of the parameter update process.

[0068] The learnable parameters of the activity state driving vector field corresponding to the convergence of the total loss, the learnable parameters of the environment state driving vector field, and the activity recovery rate parameters, spatial carrying capacity parameters, and spatial interaction decay parameters of each grid after iterative updates, together with the structural definitions of the activity state time evolution equation, the environment state time evolution equation, the emergent aggregation channel, and the feedback modulation channel, are encapsulated into a complete callable deduction framework, and the calibrated urban simulation model is output.

[0069] This embodiment also provides a computer device applicable to the intelligent construction method of urban simulation model based on activity flow and complex built environment, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the intelligent construction method of urban simulation model based on activity flow and complex built environment as proposed in the above embodiment.

[0070] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0071] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the intelligent construction method for urban simulation models based on activity flow and complex built environments as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0072] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the present invention.

Claims

1. A method for intelligently constructing urban simulation models based on activity flow and complex built environments, characterized in that, include: S1: Acquire urban activity flow data and built environment data, align and map them to the grid according to a unified spatiotemporal benchmark, and obtain the activity observation values ​​and environmental attribute values ​​of each grid. S2, construct the social domain state in layers based on activity observation values, construct the physical domain state in layers based on environmental attribute values, and concatenate the social domain state and physical domain state into a joint state representation; S3 estimates the spatial activity physical behavior parameters of each grid based on the social domain state and the physical domain state, and constructs physical conservation and spatial constraints accordingly. S4, taking the joint state representation as the initial condition, constructs an initial urban simulation model of the co-evolution of activity state and environmental state by superimposing the driving vector field and the physical guidance term of the physical behavior parameters of spatial activities, and solves the urban state trajectory. S5 calculates the loss based on the urban state trajectory, calibrates the driving vector field parameters and spatial activity physical behavior parameters, and outputs the calibrated urban simulation model.

2. The intelligent construction method for urban simulation models based on activity flow and complex built environments as described in claim 1, characterized in that, The methods for obtaining the activity observations and environmental attribute values ​​include: Mobile phone signaling, GPS trajectory, public transportation travel, and POI visit popularity data are collected as activity flow data; road network topology, building footprint, land cover, land use, infrastructure distribution, surface temperature, and air quality data are collected as built environment data. The activity flow data and built environment data were uniformly converted to a geographic coordinate system, standard time, and fixed time step, and divided into grids; the mobile phone signaling was clustered according to trajectory density and dwell time to extract dwell points and filter drift noise; Activity flow data is statistically summarized by grid to obtain activity observation values ​​for each grid; built environment data is spatially merged by grid to calculate road network density, building coverage, number of POIs and category entropy values ​​for each grid to obtain environmental attribute values ​​for each grid.

3. The intelligent construction method for urban simulation models based on activity flow and complex built environments as described in claim 2, characterized in that, The method for constructing the social domain state in a hierarchical manner includes: Individual-level analysis of activity observations is performed using anonymous users as units to extract the spatiotemporal behavioral characteristics of each user and obtain the micro-social domain state; By statistically aggregating the micro-social domain state using grids as the basic unit, the activity distribution and flow characteristics of each grid are obtained, thus yielding the meso-social domain state. By taking cities as units, a global weighted aggregation of the state of the meso-level social domain is obtained to obtain the overall activity pattern characteristics of cities and the state of the macro-level social domain.

4. The intelligent construction method for urban simulation models based on activity flow and complex built environments as described in claim 3, characterized in that, The method for constructing the physical domain state in a hierarchical manner includes: By taking a single building or street segment as a unit, the environmental attribute values ​​are analyzed at the element level to extract the spatial physical characteristics of each building and street, and the micro-physical domain state is obtained. By spatially aggregating the microscopic physical domain state using grids as the basic unit, the comprehensive characteristics of the built environment of each grid are obtained, thus yielding the mesoscopic physical domain state. By taking cities as units, the state of the meso-physical domain is globally aggregated to obtain the overall built environment characteristics of cities and the state of the macro-physical domain. The meso-level social domain state and the meso-level physical domain state of each grid are concatenated to form a joint state representation.

5. The intelligent construction method for urban simulation models based on activity flow and complex built environments as described in claim 4, characterized in that, The methods for obtaining the physical behavior parameters of the space activities include: Based on the activity flow data, the recovery process of the activity in each grid over time is fitted into an exponential decay form to obtain the activity recovery rate parameter of each grid. Based on the building area, road network density, and infrastructure distribution area in the mesophysical domain of each grid, calculate the spatial carrying capacity parameters of each grid. Based on public transportation travel data and GPS trajectory data, the relationship between inter-grid travel volume and spatial distance is fitted into a gravity attenuation form to obtain the spatial interaction attenuation parameters between grids. The activity recovery rate parameter, spatial carrying capacity parameter, and spatial interaction attenuation parameter are summarized into spatial activity physical behavior parameters for each grid.

6. The intelligent construction method for urban simulation models based on activity flow and complex built environments as described in claim 5, characterized in that, The methods for constructing physical conservation and spatial constraints include: Based on the continuous nature of activity over time, an activity conservation constraint is constructed by relating the net inflow rate at the boundary to the changes in activity at each grid, serving as a physical conservation constraint. Environmental capacity constraints are constructed based on the comparison between activity density and spatial carrying capacity parameters of each grid, and distance attenuation constraints are constructed based on the deviation between actual travel volume and spatial distance gravity attenuation relationship between grids, serving as spatial constraints.

7. The intelligent construction method for urban simulation models based on activity flow and complex built environments as described in claim 6, characterized in that, The method for constructing the initial urban simulation model includes: Using the joint state representation as the initial condition, time evolution equations for the activity state and the environment state are constructed respectively, and the vector field is constructed through the driving vector field parameterized by the multi-layer fully connected structure. Based on the physical behavior parameters of the space activity, the physical guiding terms of the activity state equation and the physical guiding terms of the environment state equation are constructed. The driving vector field is added to the corresponding physical guiding terms to establish the coupling relationship between the activity state and the environment state. Based on the coupling relationship, a two-way information transmission channel is established at three scales: micro, meso, and macro. This includes an emergent aggregation channel that transmits micro-level individual behavioral decisions to meso and macro levels through statistical aggregation, and a feedback modulation channel that transmits macro-level states down to modulate micro-level individual decisions after decomposing them according to grid weights. By integrating coupling relationships with two-way information transmission channels, an initial urban simulation model is constructed to demonstrate the cross-scale co-evolution of activity states and environmental states.

8. The intelligent construction method for urban simulation models based on activity flow and complex built environments as described in claim 7, characterized in that, The method for solving the urban state trajectory includes: Starting with the initial activity state and the initial environmental state, the initial city simulation model is continuously solved within a set time interval, and the evolution trajectory of the activity state and environmental state at any time point is output to obtain the city state trajectory.

9. The intelligent construction method for urban simulation models based on activity flow and complex built environments as described in claim 8, characterized in that, The method for calculating loss based on urban state trajectory includes: The activity status and environmental status in the urban state trajectory are compared with the actual observation data to obtain the data fitting loss; Based on the degree to which the urban state trajectory adheres to physical conservation and spatial constraints, the activity conservation deviation, environmental capacity deviation, and distance decay deviation are obtained, and the physical deviation loss is obtained by summarizing them. The scale consistency loss is obtained by considering the deviation between the aggregated states at each scale obtained from the emergent aggregation channel and the direct evolution states at the corresponding scale.

10. The intelligent construction method for urban simulation models based on activity flow and complex built environments as described in claim 9, characterized in that, The method for outputting the calibrated urban simulation model includes: The total loss is obtained by weighted summing of the data fitting loss, physical bias loss, and scale consistency loss. Based on the gradient of the total loss with respect to the driving vector field parameters and the physical behavior parameters of the space activity, the driving vector field parameters and the physical behavior parameters of the space activity are iteratively updated until the total loss converges. The converged driving vector field parameters and spatial activity physical behavior parameters are encapsulated to output a calibrated urban simulation model.

Images