Public service facility site selection method and system

Abstract

The invention relates to the technical field of geographic information systems and intelligent decisions, and discloses a public service facility site selection method and system, and the method comprises the steps: constructing a multi-modal urban population dynamic data collection and fusion module, a space-time dynamic population distribution and prediction module, a heterogeneous demand recognition and quantification module, and a multi-target intelligent site selection optimization module. According to the method and the system disclosed by the invention, through integration of multi-source heterogeneous city big data, construction of a refined space-time dynamic population distribution model, deep mining of heterogeneous service requirements of different crowds, and application of an advanced multi-target intelligent optimization algorithm, normal form transformation of public service facility site selection from static extensive to dynamic accurate is realized. According to the method and the system provided by the invention, the inherent contradiction of a traditional method in the aspects of dealing with population dynamics, demand differentiation and multi-dimensional complexity can be effectively solved, so that the scientificity, rationality, response capability and resource utilization efficiency of urban public service facility planning are remarkably improved.

Description

Technical Field

[0001] This invention relates to the fields of geographic information systems and intelligent decision-making technology, and discloses a method and system for selecting sites for public service facilities. Background Technology

[0002] For a considerable period, the site selection of public service facilities primarily relied on macro-statistical information such as static census data, administrative divisions, land use planning, and the distribution of fixed infrastructure. These methods typically included: dividing service radii based on regional population density, allocating quotas according to administrative divisions, and using geographic information systems for spatial overlay analysis to assess the accessibility and surrounding service capacity of specific plots. The core principle was to treat the urban population as a relatively stable and evenly distributed group, and through the analysis of historical data and planning documents, to estimate the long-term average demand of a specific area, thereby determining the number and approximate location of facilities. When planning schools, hospitals, or community service centers, the size of the resident population in each block, the proportion of school-age children, or the number of elderly people were usually considered, combined with the existing transportation network and current land use, striving to cover as many target groups as possible within the service radius. These methods demonstrated their rationality and effectiveness during periods of relatively stable urban structure and low population mobility, and in scenarios where requirements for facility utilization were not particularly stringent, laying the foundation for meeting basic public service needs. These features enable planners to make initial plans for urban functional zoning, ensuring a balanced distribution of infrastructure and avoiding large service gaps.

[0003] However, with the continuous evolution of information technology and the socio-economic environment, especially the accelerated pace of urban expansion and the increasing complexity of population structure in recent years, some inherent characteristics of the traditional site selection methods based on static data have gradually revealed deep-seated limitations in addressing the new challenges of current urban development. Fundamentally, the core contradiction of traditional methods lies in their assumption of the relative staticity and predictability of urban population distribution and mobility patterns, which clashes sharply with the reality of highly dynamic, unevenly distributed, and increasingly complex population behavior patterns in modern cities.

[0004] Traditional methods rely on census data and statistical yearbooks, which are often outdated and have coarse-grained timeframes, making it difficult to reflect instantaneous changes in population distribution. During weekdays, the population density in a city's central business district may far exceed its nighttime resident population; conversely, during holidays or large events, population concentration in specific areas can surge. Traditional methods fail to capture these intraday, intraweekly, and even seasonal tidal population flows, leading to facility site selection that is out of sync with actual needs. This results in severe resource shortages at certain times and significant underutilization at others, leading to resource waste. This inherent limitation of static-to-dynamic matching makes them inadequate for scenarios with high real-time requirements, such as public transportation hubs, emergency medical service points, and even commercial facilities.

[0005] Traditional methods focus on macro-level uniform coverage, making it difficult to finely identify the diverse needs of different population groups within a specific area. Within the same administrative region, there may be aging communities, clusters of young families, or areas with concentrated migrant worker populations, whose needs for public services (such as healthcare, education, and cultural entertainment) vary significantly in type and intensity. Traditional site selection methods often fail to penetrate this macro-level data to deeply uncover the micro-level characteristics of needs and service preferences, leading to a mismatch between the services provided and actual needs, thus reducing service efficiency and resident satisfaction. This contradiction between coarse-grained universality and fine-grained differentiated needs significantly diminishes the targetedness and effectiveness of public services.

[0006] Traditional site selection methods often fail to comprehensively consider the dynamic impacts of complex, multi-dimensional factors such as urban traffic congestion, environmental carrying capacity, and emergency response during the optimization decision-making process. Even if a location seems ideal based on population data, its actual effectiveness as a public service facility will be significantly reduced if it suffers from severe traffic congestion or is located in an environmentally sensitive area. Traditional methods lack the ability to perceive and iteratively optimize these dynamic and interconnected factors in real time, making site selection decisions rigid and inflexible when facing complex urban systems. This contradiction between linear static optimization and nonlinear dynamic coupling presents a significant challenge to urban planning in practical operation.

[0007] The aforementioned deep-seated technical contradictions indicate that existing methods for selecting public service facility locations are no longer adequate to accurately meet the highly dynamic and complex development needs of modern cities. The core issue lies in their inability to effectively integrate and utilize the massive, real-time, and multi-dimensional population flow and distribution data within cities, leading to planning delays, resource misallocation, and low service efficiency. Therefore, developing a method and system capable of real-time sensing and accurate prediction of urban population dynamics and flow trends, and based on this, constructing a more scientific, efficient, and intelligent method for selecting public service facility locations to improve the accuracy of urban resource allocation and the responsiveness of the service system, has become a key challenge and an urgent technical problem for those skilled in the art. Summary of the Invention

[0008] To achieve the above-mentioned objectives, this invention provides a method and system for selecting public service facilities, aiming to overcome the inherent limitations of existing methods for selecting public service facilities in dealing with the dynamic nature of urban population, heterogeneous demand, and the coupling of complex multi-dimensional factors. It provides a scientific method and integrated system that can perceive, accurately predict, and optimize the layout of public service facilities in real time.

[0009] The method for selecting the site for public service facilities disclosed in this invention includes the following steps: A multimodal urban population dynamic data acquisition and fusion module is constructed to collect and preprocess multi-source heterogeneous urban operation data in real time. The data covers population location, flow trajectory, activity pattern and environmental parameter information. A spatiotemporal dynamic population distribution and prediction module is constructed to establish a refined spatiotemporal distribution model of the urban population based on the fused urban operation data, and to predict the population distribution in future periods. A heterogeneous demand identification and quantification module is constructed to identify and quantify the differentiated demands of different regions and different groups of people for various public service facilities based on the activity patterns of the population and the predicted population distribution characteristics. A multi-objective intelligent site selection optimization module is constructed to comprehensively consider the spatiotemporal dynamic population distribution, heterogeneous demand quantification results, urban geospatial constraints, and multi-dimensional performance indicators to generate a set of Pareto optimal candidate site selection schemes for public service facilities.

[0010] Preferably, the multimodal urban population dynamic data acquisition and fusion module includes: The data interface layer is configured with multiple data interface units, including a mobile communication base station signaling data interface, a public transportation card swiping data interface, an urban Internet of Things sensor network data interface, a geofence event data interface, and an environmental monitoring data interface. The mobile communication base station signaling data interface is responsible for obtaining anonymized and aggregated signaling data such as user location updates, call records, and SMS sending from the signaling gateway of the mobile communication operator; The public transportation card swiping data interface is responsible for obtaining the card swiping records of bus cards and subway cards from the data center of the urban public transportation system. The records include the passenger's anonymous ID, swiping time, swiping station information, and route information. The urban IoT sensor network data interface is responsible for acquiring data from various smart sensors deployed in the city, such as road traffic flow detectors, pedestrian counters, environmental noise sensors, air quality sensors, and public area Wi-Fi probes. The geofence event data interface is responsible for obtaining anonymized records of personnel entry and exit events within a specific geographic area from the location-based service platform; The environmental monitoring data interface is responsible for obtaining real-time environmental parameters such as temperature, humidity, air pressure, and light intensity from urban environmental monitoring stations.

[0011] Preferably, the multimodal urban population dynamic data acquisition and fusion module further includes: The data preprocessing and cleaning unit preprocesses the raw data obtained from the data interface layer using data processing algorithms. The preprocessing includes data format standardization, missing value imputation, outlier detection and correction, and noise filtering. For signaling data, a density-based clustering method is used to identify and filter out abnormal base station location drift data; For card swipe data, a time series prediction-based method is used to fill in the missing card swipe records caused by equipment failure; For sensor data, a Kalman filter-based method is used to smooth high-frequency noise.

[0012] Preferably, the multimodal urban population dynamic data acquisition and fusion module further includes: The spatiotemporal alignment and fusion unit employs a unified geospatial reference system and timestamp synchronization mechanism to spatiotemporally align data from different data sources. The fusion process uses a weighted fusion algorithm, assigning different weights to each data source based on its reliability, coverage, and temporal resolution, to generate a unified urban population dynamic database. This database includes spatial distribution density, instantaneous flow direction and speed, and residence time characteristics within specific areas. The database is stored in a distributed time-series database and a spatial index is established to support efficient querying.

[0013] Preferably, the spatiotemporal dynamic population distribution and prediction module includes: The spatial gridding unit divides the entire urban planning area into multi-scale, non-overlapping spatial grids. The gridding adopts a hierarchical hexagonal grid system, and by adjusting the grid resolution parameters, it can achieve flexible division from macro to micro areas to adapt to population distribution modeling with different accuracy requirements. The feature engineering unit extracts and constructs multidimensional spatiotemporal features from the fused urban population dynamics database. The features include temporal features, spatial features, and contextual features, and the feature data is organized in tensor form.

[0014] Preferably, the spatiotemporal dynamic population distribution and prediction module further includes: A population distribution modeling unit, based on the feature data, constructs a deep learning model to simulate and learn the distribution patterns of urban population in different spatiotemporal dimensions; the model adopts a spatiotemporal graph convolutional network architecture, wherein the graph convolutional layer is used to capture the spatial dependencies between grid cells, and the temporal convolutional layer or long short-term memory network layer is used to capture the time series dynamic patterns of population distribution. The population prediction unit utilizes the trained population distribution modeling unit to perform rolling predictions of urban population distribution for a specific future time period. The prediction takes into account historical population distribution trends, real-time traffic conditions, weather forecasts, and pre-set holiday or large-scale event schedules. The prediction results are output in the form of a spatiotemporal population density map with confidence intervals, wherein each hexagonal grid cell contains a predicted population density value and its corresponding standard deviation at different future time steps.

[0015] Preferably, the heterogeneous demand identification and quantification module includes: The population profiling unit constructs refined profiles of different population categories based on behavioral patterns reflected in dynamic population data and auxiliary demographic data. The profiles are represented by vectors and capture the characteristics of each population category in terms of socioeconomic attributes, activity patterns, and service preferences. The dynamic population is then classified unsupervised using a clustering algorithm. The service demand pattern database construction unit analyzes historical public service facility usage records, user feedback data, relevant discussions on social media, and industry standards to establish demand pattern databases for different types of public service facilities. The databases include typical user groups, service duration distribution, peak period characteristics, and user preferences for service accessibility, quality, and environmental comfort for each type of service facility. The demand quantification and mapping unit combines the population profile with the service demand pattern library and the population density information output by the spatiotemporal dynamic population distribution and prediction module to quantify the immediate and future predicted demand for various public service facilities within a specific spatiotemporal grid unit. The quantification adopts a method based on weighted linear combination or nonlinear regression model to output the demand intensity index of specific public services for each grid unit at different time steps.

[0016] Preferably, the multi-target intelligent location optimization module includes: The objective function definition unit defines a series of mutually independent or partially coupled optimization objective functions, including maximizing accessibility, maximizing service coverage, equalizing facility utilization, optimizing social equity, optimizing environmental and ecological benefits, minimizing economic costs, and maximizing risk resistance and resilience. The constraint setting unit defines the hard constraints in the site selection process, including land use restrictions, minimum / maximum facility capacity, budget restrictions, service radius restrictions, and compatibility requirements with existing infrastructure.

[0017] Preferably, the multi-target intelligent location optimization module further includes: The multi-objective optimization algorithm unit employs a non-dominated sorting genetic algorithm II or multi-objective particle swarm optimization to handle multiple conflicting optimization objectives. The algorithm takes population distribution prediction, heterogeneous demand quantification results, and urban geographic information data as inputs, and uses the coordinates, facility type, and service capacity of each candidate site selection scheme as decision variables. The algorithm iteratively generates and evaluates a set of non-dominated solutions. The candidate scheme generation and screening unit generates a set of refined public service facility candidate site selection schemes from the Pareto optimal solution set output by the multi-objective optimization algorithm, based on preset decision preferences or through interactive screening.

[0018] This invention also discloses a public service facility site selection system, comprising: A data acquisition and fusion server cluster is used to deploy the multimodal urban population dynamic data acquisition and fusion module. The server cluster is configured with a high-throughput message queue and a big data storage system to support real-time access and distributed storage of massive amounts of data. A data processing and analysis server cluster is used to deploy the spatiotemporal dynamic population distribution and prediction module, the heterogeneous demand identification and quantification module, and the multi-objective intelligent location optimization module; the server cluster is configured with high-performance computing units to support parallel computing of deep learning model training and complex optimization algorithms; the cluster adopts containerized deployment and microservice architecture. The user interaction and visualization server is used to deploy the user interface portion of the site selection scheme evaluation and visualization module and the decision support and feedback iteration module; the server is configured with web server software and application server framework to provide stable web services and low-latency user response. A geospatial database is used to store basic urban geographic information, existing public service facility data, land use planning data, and historical population statistics; the database uses a PostgreSQL database extended with PostGIS. Distributed file systems and object storage are used to store model weights, training data snapshots, and historical addressing records.

[0019] Compared with the prior art, the beneficial effects of the present invention are: The methods and systems disclosed in this invention integrate multi-source, heterogeneous urban big data to construct a refined spatiotemporal dynamic population distribution model, deeply explore the heterogeneous service needs of different population groups, and utilize advanced multi-objective intelligent optimization algorithms to achieve a paradigm shift in public service facility site selection from static and extensive to dynamic and precise. The methods and systems provided effectively resolve the inherent contradictions of traditional methods in dealing with population dynamics, demand differentiation, and multi-dimensional complexity, thereby significantly improving the scientific rigor, rationality, responsiveness, and resource utilization efficiency of urban public service facility planning. Attached Figure Description

[0020] Figure 1 A schematic diagram of the overall architecture of the public service facility site selection system of this invention; Figure 2 A flowchart illustrating the public service facility site selection method of the present invention; Figure 3 A schematic diagram of the structure of the multimodal urban population dynamic data acquisition and fusion module in this invention; Figure 4 A schematic diagram of the spatiotemporal dynamic population distribution and prediction module in this invention; Figure 5 A schematic diagram of the heterogeneous demand identification and quantification module in this invention; Figure 6 A schematic diagram of the structure of the multi-target intelligent addressing optimization module in this invention. Detailed Implementation

[0021] This invention provides a method and system for selecting public service facility sites. It aims to integrate multi-source, heterogeneous urban big data to construct a refined spatiotemporal dynamic population distribution model, deeply explore the heterogeneous service needs of different population groups, and utilize advanced multi-objective intelligent optimization algorithms to achieve a paradigm shift in public service facility site selection from traditional static planning to dynamic and precise optimization. The method and system of this invention can effectively address the challenges posed by the dynamic nature of urban populations, the heterogeneity of demands, and the coupling of complex multi-dimensional factors, significantly improving the scientific rigor, rationality, responsiveness, and resource utilization efficiency of urban public service facility planning.

[0022] Please see Figure 1-6 The public service facility site selection method disclosed in this invention includes the following core steps: First, constructing a multimodal urban population dynamic data collection and fusion module; Second, constructing a spatiotemporal dynamic population distribution and prediction module; Third, constructing a heterogeneous demand identification and quantification module; Fourth, constructing a multi-objective intelligent site selection optimization module.

[0023] Example 1: The structure of the multimodal urban population dynamic data acquisition and fusion module is as follows: Figure 3 As shown, it specifically includes a data interface layer, a data preprocessing and cleaning unit, and a spatiotemporal alignment and fusion unit. The data interface layer is configured with multiple data interface units, which are used to obtain raw data from different urban operation data sources. The data interface units cover mobile communication base station signaling data interfaces, public transportation card swiping data interfaces, urban IoT sensor network data interfaces, geofence event data interfaces, and environmental monitoring data interfaces.

[0024] The mobile communication base station signaling data interface is responsible for establishing a TLS encrypted connection with the mobile communication operator's security gateway to obtain anonymized and aggregated signaling data such as user location update records, call duration distribution, and SMS sending frequency in real time. The data is processed in batches at an hourly granularity and transmitted in JSON or Protobuf format, with an average daily data processing volume of up to 50GB.

[0025] The public transportation card swiping data interface periodically retrieves bus and subway card swiping records from the city's public transportation system data center via a secure file transfer protocol. These records include encrypted anonymous user IDs, timestamps accurate to the second, swiping station information, and the route ID. The daily average number of swiping records exceeds 5 million. The city's IoT sensor network data interface acquires real-time data streams from various smart sensors widely deployed throughout the city via the MQTT protocol. These sensors include, but are not limited to, road traffic flow detectors (providing vehicle flow and speed data every 5 minutes), pedestrian counters (providing regional pedestrian flow data every 15 minutes), environmental noise sensors, air quality sensors (PM2.5, O3, NO2 concentration data, etc.), and public area Wi-Fi probes (providing Wi-Fi device density and connection duration). This data is sent in structured text or binary stream format to ensure low latency.

[0026] The geofence event data interface retrieves anonymized records of people entering and leaving specific geographic areas (e.g., commercial areas, parks, residential areas) from a location-based service platform. These records typically include event type (entry / departure), anonymous user ID, timestamp, and geofence ID, used to identify people's stay and activity patterns in specific areas. The environmental monitoring data interface is responsible for acquiring environmental parameters such as temperature, humidity, air pressure, and light intensity in real time from urban environmental monitoring stations. This data is updated hourly and provides environmental context for subsequent population behavior pattern analysis.

[0027] The data preprocessing and cleaning unit performs rigorous preprocessing on the raw data obtained from the data interface layer using a series of efficient data processing algorithms. The preprocessing includes data format standardization, missing value imputation, outlier detection and correction, and noise filtering. To address potential base station location drift in mobile communication signaling data, a density-based clustering method is employed for identification and filtering.

[0028] For missing card-swipe records in public transportation data due to equipment malfunction or network interruption, a time-series prediction-based method is used for filling in the gaps. Taking the card-swipe data of a specific bus route as an example, by analyzing the historical hourly card-swipe volume time series, the missing card-swipe volume for the time period is predicted and filled in, while also considering external factors such as days of the week and holidays. To address the high-frequency noise commonly found in urban IoT sensor data, Kalman filtering is used for smoothing. Through iterative updates of prediction and measurement states, measurement errors are effectively reduced. Applying Kalman filtering to road traffic flow sensor data allows the smoothed data to better reflect the actual traffic flow trend, reducing prediction errors by 15% to 20% compared to the original data.

[0029] The spatiotemporal alignment and fusion unit employs a unified geospatial reference system and a precise timestamp synchronization mechanism to accurately align preprocessed data from different data sources in spatiotemporal terms. All data timestamps are uniformly converted to UTC time and accurate to milliseconds. The fusion process uses a weighted fusion algorithm, assigning different weights to each data source based on its reliability, coverage, and temporal resolution. Mobile signaling data, due to its wide coverage and high update frequency, is given a relatively high weight (e.g., 0.4); public transportation card swipe data, reflecting real travel behavior and being accurate, can be weighted at 0.3; IoT sensor data, providing detailed information for specific areas, can be weighted at 0.2; and geofencing and environmental data, serving as auxiliary information, are weighted at 0.1. These weights can be dynamically adjusted based on data source quality assessments and actual application scenarios.

[0030] The merged data generates a unified urban population dynamics database, which includes population distribution density in urban space, instantaneous flow direction and speed (inferred from the trajectories of anonymous users at different time points), and dwell time characteristics within specific areas (obtained through geofencing events and Wi-Fi probe data analysis). This database is stored in the distributed time-series database InfluxDB, whose high write throughput and query performance effectively support massive amounts of time-series data. Furthermore, to support efficient spatial queries, an R-tree-based spatial index is built in InfluxDB, enabling complex spatial queries, such as querying all grid cells within a certain area where population density exceeds a threshold, to have response times in the hundreds of milliseconds.

[0031] Example 2: The spatiotemporal dynamic population distribution and prediction module has the following structure: Figure 4 As shown, it specifically includes spatial gridding units, feature engineering units, population distribution modeling units, and population prediction units. The spatial gridding unit is responsible for dividing the entire urban planning area into multi-scale, non-overlapping hexagonal spatial grids.

[0032] The gridding employs Uber's H3 geospatial indexing system, which allows for flexible division from macro to micro regions by adjusting grid resolution parameters. When the resolution is set to level 9, each hexagonal grid cell has an area of ​​approximately 0.25 square kilometers, suitable for macro-level population distribution analysis at the city area level. When the resolution is increased to level 12, the area of ​​each grid cell shrinks to approximately 0.0004 square kilometers, corresponding to a side length of approximately 7 meters, enabling fine-grained population distribution modeling at the street or even building scale. Each hexagonal grid cell is assigned a globally unique spatial index identifier, ensuring efficient data aggregation and querying at different granularities.

[0033] The feature engineering unit extracts and constructs multidimensional spatiotemporal features from the fused urban population dynamics database, providing rich input for subsequent population distribution prediction models. These features include: temporal features, such as the current hour, day of the week, holiday markers, seasons, and special event markers; spatial features, such as the area of ​​each grid cell, the density of points of interest within or near that grid cell, and road network accessibility indices, including centrality (measuring the number of shortest paths through the grid) and betweenness (measuring the reciprocal of the average shortest distance from that grid to all other grids); and contextual features, such as real-time weather conditions and ongoing public activities within a specific area. The feature data is organized in tensor form to accommodate the input requirements of deep learning models.

[0034] A population distribution modeling unit, based on the aforementioned feature data, constructs a deep learning model to simulate and learn the distribution patterns of urban population across different spatiotemporal dimensions. The model employs a spatiotemporal graph convolutional network architecture. In this network, graph convolutional layers capture the spatial dependencies between grid cells, i.e., population flows and mutual influences between adjacent areas or areas connected by road networks. Spatial relationships are represented by an adjacency matrix, where matrix elements represent the connectivity or inverse distance between grid cells. Graph convolution operations aggregate feature information from neighboring grids into the current grid. Temporal convolutional layers or long short-term memory (LSTM) layers capture the temporal dynamic patterns of population distribution, such as daily tidal phenomena (population concentrating in work areas during morning and evening rush hours and spreading to residential areas at night) or weekly cyclical patterns (differences in population activity between weekdays and weekends).

[0035] The model takes as input urban population density grid data at multiple historical time steps (e.g., hourly data from the past 24 hours) and corresponding spatiotemporal features, and outputs a predicted urban population density map for the current or future moment (e.g., the next hour). The model is trained end-to-end, using backpropagation and the Adam optimizer for parameter updates. Mean squared error is used as the loss function to minimize the difference between predicted and actual values. In training a specific urban area, the model uses 24 spatiotemporal graph convolutional network layers, each containing one graph convolutional layer and one temporal convolutional layer, with a hidden dimension of 64, a learning rate of 0.001, a batch size of 32, and a training period of 100 epochs.

[0036] The population prediction unit utilizes the trained population distribution modeling unit to perform rolling predictions of urban population distribution for specific future time periods. The prediction process comprehensively considers historical population distribution trends, real-time traffic conditions, future weather forecasts, and pre-set holiday or large-scale event schedules. The prediction results are output as a spatiotemporal population density map with confidence intervals. Each hexagonal grid cell contains a predicted population density value and its corresponding standard deviation at different future time steps to quantify the uncertainty of the prediction. For example, a commercial area grid cell might predict a population density of 300 people / km² for the next two hours, with a confidence interval of [280, 320] people / km² and a standard deviation of 10. The spatiotemporal resolution of the prediction matches the spatial gridding unit division, providing hourly temporal granularity and hundred-meter-level spatial granularity, ensuring the practicality and accuracy of the prediction results.

[0037] Example 3: The heterogeneous demand identification and quantification module has the following structure. Figure 5 As shown, the system specifically includes a user profile construction unit, a service demand pattern library construction unit, and a demand quantification and mapping unit. The user profile construction unit builds refined profiles for different user groups based on behavioral patterns reflected in dynamic population data and auxiliary demographic data. Behavioral patterns are obtained by analyzing stop patterns in mobile signaling data, travel chains in public transportation card swipe data, consumption habits, and social interactions.

[0038] Supplementary demographic data includes age groups inferred from anonymized signaling data (through call duration and app usage characteristics) and occupational group distribution (through weekday daytime residence locations). The profiles are represented using high-dimensional vectors to capture the characteristics of each population group in terms of socioeconomic attributes, activity patterns, and service preferences. Unsupervised classification of the dynamic population is performed using clustering algorithms, such as hierarchical clustering or Gaussian mixture models, to identify typical population categories such as morning rush hour commuters, students, elderly residents, and tourists.

[0039] The service demand pattern database construction unit establishes a demand pattern database for different types of public service facilities through in-depth analysis of historical public service facility usage records, user feedback data, relevant discussions on social media, and industry standards. The database includes typical user groups, service duration distribution, peak period characteristics, and user preferences for service accessibility, quality, and environmental comfort for each type of service facility. The database uses an ontology-based structure for storage, supporting complex semantic queries and reasoning, and can query community elderly care service points that are within walking distance and have barrier-free facilities. For example, for a community health service center, its service pattern might include a coverage radius of 2 kilometers, primarily serving the elderly and children, an average daily patient volume of 150, and a 20% increase in emergency demand on weekends.

[0040] The demand quantification and mapping unit combines the aforementioned population profile with the service demand pattern library, and incorporates population density information output from the spatiotemporal dynamic population distribution and prediction module, to quantify the immediate and future projected demand for various public service facilities within a specific spatiotemporal grid unit. The quantification employs a method based on a weighted linear combination or nonlinear regression model. This model takes as input the weighted quantity of different population groups within a specific grid unit (the number of elderly people multiplied by their preference weight for medical services), the corresponding service preference intensity, and the contextual characteristics of the area (e.g., the supply level of surrounding existing facilities, traffic congestion level), and outputs a demand intensity index for specific public services (outpatient services, compulsory education places, sports field usage) for each grid unit at different time steps.

[0041] The demand intensity index is expressed as a normalized value (e.g., between 0 and 1) and reflects the scarcity of service resources. A value of 1 indicates an extremely high and urgent demand for the service in the area, while 0 indicates saturated or nonexistent demand. For example, a grid cell with a demand index of 0.9 for children's education services at 2 PM on Wednesday indicates a very high demand for children's education resources during that time period.

[0042] Example 4: The structure of the multi-target intelligent location optimization module is as follows: Figure 6 As shown, it specifically includes an objective function definition unit, a constraint setting unit, a multi-objective optimization algorithm unit, and a candidate solution generation and screening unit. The objective function definition unit defines a series of independent or partially coupled optimization objective functions based on the complex objectives of public service facility site selection. These objective functions include, but are not limited to: (1) Maximizing accessibility: This objective aims to minimize the average travel time or distance from service facilities to the population within the service area, or to maximize the total population that can be served within a specific time period (e.g., within a 15-minute commute circle). The travel time is calculated using a dynamic traffic model that takes into account real-time traffic conditions and public transport network data, such as using network analysis based on the Dijkstra algorithm, and adjusting segment travel times in conjunction with real-time traffic conditions (via traffic sensor data) to reflect the impact of traffic congestion.

[0043] (2) Maximize service coverage: This objective aims to maximize the geographical coverage of the population in need by service facilities, ensuring that all needs within the grid unit are effectively responded to, or minimizing the areas of need that are not effectively covered. Coverage can be defined as the proportion of the population that can be covered by the facility within a given service radius or service time.

[0044] (3) Facility utilization leveling: This objective aims to minimize the difference in utilization rates between different facilities or the same facility at different times, and to avoid resource idleness or overload. The utilization rate is measured by the ratio of predicted demand to facility capacity. A facility is considered to be overloaded if its utilization rate exceeds 85%, and considered to be idle if it is below 30%.

[0045] (4) Social equity optimization: This goal aims to ensure the fair distribution of service resources among different social groups and regions with different income levels. Equity can be reflected by calculating the Gini coefficient to measure the balance of resource allocation and minimizing it; or by maximizing service coverage in areas where poor or vulnerable groups are concentrated.

[0046] (5) Environmental and ecological benefits optimization: This objective aims to minimize the impact of facility construction and operation on the surrounding environment, minimize new carbon emissions and noise pollution, or maximize the proportion of green space around new facilities. Carbon emissions can be estimated through factors such as construction materials, energy consumption, and traffic flow.

[0047] (6) Minimize economic costs: This objective aims to minimize the total cost of land acquisition, construction, operation and maintenance. This includes a comprehensive consideration of land market prices, construction costs, human resource costs and long-term maintenance budgets.

[0048] (7) Maximizing Risk Resistance and Resilience: This objective aims to consider emergency response capabilities under extreme events (such as natural disasters and public health emergencies). It is quantified by assessing indicators such as redundancy between facilities (whether other facilities can quickly take over the service capacity when one facility fails) and accessibility of backup routes (whether there are alternative transportation routes when roads are interrupted).

[0049] The constraint setting unit defines the hard constraints in the site selection process to ensure the feasibility of the site selection plan. These constraints include, but are not limited to: land use restrictions (achieved through comparison with the urban planning department's land use database); minimum / maximum facility capacity restrictions (depending on facility type and service standards); budget restrictions (set according to the city's fiscal budget); service radius restrictions; and compatibility requirements with existing infrastructure.

[0050] The multi-objective optimization algorithm unit employs an intelligent optimization algorithm capable of handling multiple conflicting optimization objectives, such as non-dominated sorting genetic algorithm II or multi-objective particle swarm optimization. The algorithm takes population distribution prediction, heterogeneous demand quantification results, and urban geographic information data as inputs. The coordinates, facility type, and service capacity of each candidate site selection scheme are used as decision variables.

[0051] The NSGA-II algorithm iteratively generates and evaluates a set of non-dominated solutions, meaning that any solution in this set cannot improve the performance of other objective functions without sacrificing the performance of at least one objective function. The algorithm's parameters include, but are not limited to, population size, crossover probability, mutation probability, and number of iterations. Through multiple iterations, the algorithm can find a diverse and high-performance set of non-dominated solutions in a vast solution space.

[0052] The candidate solution generation and screening unit generates a set of refined candidate site selection schemes for public service facilities from the Pareto optimal solution set output by the multi-objective optimization algorithm, based on preset decision preferences or through interactive screening. If planners prioritize maximizing accessibility and minimizing economic costs, the system will select schemes that perform well in both objectives from the Pareto front. Each scheme is accompanied by specific performance indicators for each optimization objective, facilitating trade-offs and comparisons by decision-makers.

[0053] Example 5: The site selection scheme evaluation and visualization module includes a multi-dimensional performance index calculation unit and an interactive visualization interface unit. For each candidate site selection scheme, the multi-dimensional performance index calculation unit calculates specific numerical indicators in terms of accessibility, coverage, utilization balance, social equity, environmental benefits, and economic costs, based on the spatiotemporal dynamic population distribution and prediction results and the quantification results of heterogeneous demand. The calculation employs advanced spatial analysis tools, using graph theory algorithms from the NetworkX library to calculate network accessibility, and using the shortest path algorithm to calculate the average travel time from densely populated areas to facilities on the urban road network map. Simultaneously, the calculation results are verified through simulation. This model can simulate the behavior of millions of individuals, including their travel choices and facility usage patterns, to more realistically assess the actual utilization rate of facilities and user satisfaction.

[0054] The interactive visualization interface unit provides a WebGIS-based interactive graphical user interface. This interface can intuitively display the evaluation results of various candidate site selection schemes in multiple formats, including map overlays, heat maps, flow maps, and charts. Map overlays can display spatial grid population density, service demand heat maps, existing facility distribution, transportation networks, and land parcel attribute information at different resolutions. Users can flexibly switch map layers to simultaneously display population density, school distribution, and traffic congestion status in a specific area. Heat maps use color intensity to represent demand intensity or population density within a region. Flow maps use arrows to indicate the direction and scale of population flow.

[0055] The interface supports user-defined views (switching between population distribution forecasts by timeline to observe demand changes over different periods), filtering conditions (filtering by facility type, filtering by target performance indicators, such as only displaying options with accessibility above 90%), and comparative analysis between different options, visually presenting the advantages and disadvantages of each option through side-by-side or overlay displays. The charts can dynamically display the specific values ​​and rankings of different site selection options on various evaluation indicators; for example, bar charts compare the construction costs of different options, and radar charts show the balance of multi-objective performance. The visualization interface interacts with the backend data service and optimization module via a RESTful API interface, ensuring real-time data updates and low-latency operation response.

[0056] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for selecting the site of public service facilities, characterized in that, Includes the following steps: A multimodal urban population dynamic data acquisition and fusion module is constructed to collect and preprocess multi-source heterogeneous urban operation data in real time. The data covers population location, flow trajectory, activity pattern and environmental parameter information. A spatiotemporal dynamic population distribution and prediction module is constructed to establish a refined spatiotemporal distribution model of the urban population based on the fused urban operation data, and to predict the population distribution in future periods. A heterogeneous demand identification and quantification module is constructed to identify and quantify the differentiated demands of different regions and different groups of people for various public service facilities based on the activity patterns of the population and the predicted population distribution characteristics. A multi-objective intelligent site selection optimization module is constructed to comprehensively consider the spatiotemporal dynamic population distribution, heterogeneous demand quantification results, urban geospatial constraints, and multi-dimensional performance indicators to generate a set of Pareto optimal candidate site selection schemes for public service facilities.

2. The method for selecting the site for public service facilities according to claim 1, characterized in that, The multimodal urban population dynamic data acquisition and fusion module includes: The data interface layer is configured with multiple data interface units, including a mobile communication base station signaling data interface, a public transportation card swiping data interface, an urban Internet of Things sensor network data interface, a geofence event data interface, and an environmental monitoring data interface. The mobile communication base station signaling data interface is responsible for obtaining anonymized and aggregated signaling data such as user location updates, call records, and SMS sending from the signaling gateway of the mobile communication operator; The public transportation card swiping data interface is responsible for obtaining the card swiping records of bus cards and subway cards from the data center of the urban public transportation system. The records include the passenger's anonymous ID, swiping time, swiping station information, and route information. The urban IoT sensor network data interface is responsible for acquiring data from various smart sensors deployed in the city, such as road traffic flow detectors, pedestrian counters, environmental noise sensors, air quality sensors, and public area Wi-Fi probes. The geofence event data interface is responsible for obtaining anonymized records of personnel entry and exit events within a specific geographic area from the location-based service platform; The environmental monitoring data interface is responsible for obtaining real-time environmental parameters such as temperature, humidity, air pressure, and light intensity from urban environmental monitoring stations.

3. The method for selecting the site for public service facilities according to claim 2, characterized in that, The multimodal urban population dynamic data acquisition and fusion module also includes: The data preprocessing and cleaning unit preprocesses the raw data obtained from the data interface layer using data processing algorithms. The preprocessing includes data format standardization, missing value imputation, outlier detection and correction, and noise filtering. For signaling data, a density-based clustering method is used to identify and filter out abnormal base station location drift data; For card swipe data, a time series prediction-based method is used to fill in the missing card swipe records caused by equipment failure; For sensor data, a Kalman filter-based method is used to smooth high-frequency noise.

4. The method for selecting the site for public service facilities according to claim 3, characterized in that, The multimodal urban population dynamic data acquisition and fusion module also includes: The spatiotemporal alignment and fusion unit employs a unified geospatial reference system and timestamp synchronization mechanism to spatiotemporally align data from different data sources. The fusion process uses a weighted fusion algorithm, assigning different weights to each data source based on its reliability, coverage, and temporal resolution, to generate a unified urban population dynamic database. This database includes spatial distribution density, instantaneous flow direction and speed, and residence time characteristics within specific areas. The database is stored in a distributed time-series database and a spatial index is established to support efficient querying.

5. The method for selecting the site for public service facilities according to claim 1, characterized in that, The spatiotemporal dynamic population distribution and prediction module includes: The spatial gridding unit divides the entire urban planning area into multi-scale, non-overlapping spatial grids. The gridding adopts a hierarchical hexagonal grid system, and by adjusting the grid resolution parameters, it can achieve flexible division from macro to micro areas to adapt to population distribution modeling with different accuracy requirements. The feature engineering unit extracts and constructs multidimensional spatiotemporal features from the fused urban population dynamics database. The features include temporal features, spatial features, and contextual features, and the feature data is organized in tensor form.

6. The method for selecting the site for public service facilities according to claim 5, characterized in that, The spatiotemporal dynamic population distribution and prediction module also includes: A population distribution modeling unit, based on the feature data, constructs a deep learning model to simulate and learn the distribution patterns of urban population in different spatiotemporal dimensions; the model adopts a spatiotemporal graph convolutional network architecture, wherein the graph convolutional layer is used to capture the spatial dependencies between grid cells, and the temporal convolutional layer or long short-term memory network layer is used to capture the time series dynamic patterns of population distribution. The population prediction unit utilizes the trained population distribution modeling unit to perform rolling predictions of urban population distribution for a specific future time period. The prediction takes into account historical population distribution trends, real-time traffic conditions, weather forecasts, and pre-set holiday or large-scale event schedules. The prediction results are output in the form of a spatiotemporal population density map with confidence intervals, wherein each hexagonal grid cell contains a predicted population density value and its corresponding standard deviation at different future time steps.

7. The method for selecting the site for public service facilities according to claim 1, characterized in that, The heterogeneous demand identification and quantification module includes: The population profiling unit constructs refined profiles of different population categories based on behavioral patterns reflected in dynamic population data and auxiliary demographic data. The profiles are represented by vectors and capture the characteristics of each population category in terms of socioeconomic attributes, activity patterns, and service preferences. The dynamic population is then classified unsupervised using a clustering algorithm. The service demand pattern database construction unit analyzes historical public service facility usage records, user feedback data, relevant discussions on social media, and industry standards to establish demand pattern databases for different types of public service facilities. The databases include typical user groups, service duration distribution, peak period characteristics, and user preferences for service accessibility, quality, and environmental comfort for each type of service facility. The demand quantification and mapping unit combines the population profile with the service demand pattern library and the population density information output by the spatiotemporal dynamic population distribution and prediction module to quantify the immediate and future predicted demand for various public service facilities within a specific spatiotemporal grid unit. The quantification adopts a method based on weighted linear combination or nonlinear regression model to output the demand intensity index of specific public services for each grid unit at different time steps.

8. The method for selecting the site for public service facilities according to claim 1, characterized in that, The multi-objective intelligent location optimization module includes: The objective function definition unit defines a series of mutually independent or partially coupled optimization objective functions, including maximizing accessibility, maximizing service coverage, equalizing facility utilization, optimizing social equity, optimizing environmental and ecological benefits, minimizing economic costs, and maximizing risk resistance and resilience. The constraint setting unit defines the hard constraints in the site selection process, including land use restrictions, minimum / maximum facility capacity, budget restrictions, service radius restrictions, and compatibility requirements with existing infrastructure.

9. The method for selecting the site for public service facilities according to claim 8, characterized in that, The multi-objective intelligent location optimization module also includes: The multi-objective optimization algorithm unit employs a non-dominated sorting genetic algorithm II or multi-objective particle swarm optimization to handle multiple conflicting optimization objectives. The algorithm takes population distribution prediction, heterogeneous demand quantification results, and urban geographic information data as inputs, and uses the coordinates, facility type, and service capacity of each candidate site selection scheme as decision variables. The algorithm iteratively generates and evaluates a set of non-dominated solutions. The candidate scheme generation and screening unit generates a set of refined public service facility candidate site selection schemes from the Pareto optimal solution set output by the multi-objective optimization algorithm, based on preset decision preferences or through interactive screening.

10. A public service facility site selection system, characterized in that, include: A data acquisition and fusion server cluster is used to deploy the multimodal urban population dynamic data acquisition and fusion module. The server cluster is configured with a high-throughput message queue and a big data storage system to support real-time access and distributed storage of massive amounts of data. A data processing and analysis server cluster is used to deploy the spatiotemporal dynamic population distribution and prediction module, the heterogeneous demand identification and quantification module, and the multi-objective intelligent site selection optimization module. The server cluster is configured with high-performance computing units to support parallel computing for deep learning model training and complex optimization algorithms. The cluster adopts containerized deployment and a microservice architecture; The user interaction and visualization server is used to deploy the user interface portion of the site selection scheme evaluation and visualization module and the decision support and feedback iteration module; the server is configured with web server software and application server framework to provide stable web services and low-latency user response. A geospatial database is used to store basic urban geographic information, existing public service facility data, land use planning data, and historical population statistics; the database uses a PostgreSQL database extended with PostGIS. Distributed file systems and object storage are used to store model weights, training data snapshots, and historical addressing records.

Images