Intelligent site selection system for urban emergency material storage points

Abstract

The invention relates to the technical field of urban emergency management, and discloses an intelligent site selection system for an urban emergency material storage point, and the system comprises a data fusion and preprocessing module which is used for obtaining and carrying out the multi-source fusion of mobile terminal data, underground pipe network three-dimensional data, a tree distribution and pruning period, regional environment elements, and historical material putting and distribution data, and carrying out the selection of the urban emergency material storage point; performing time and space alignment, abnormal value detection and missing value filling on the data, and outputting a standardized space-time feature set; and the space-time demand modeling module is used for dividing the daytime rapid demand hot area and the night conventional demand hot area of the target area based on the space-time feature set. The intelligent site selection system for the urban emergency material storage point aims to solve the problems that a real-time optimal site selection decision under deep coupling of multiple constraints such as dynamic space-time requirements, underground pipe networks and environmental factors is difficult to realize, and end-to-end closed-loop capability is lacked to support continuously evolved emergency deployment requirements.

Description

Technical Field

[0001] This invention relates to the field of urban emergency management technology, specifically to an intelligent site selection system for urban emergency material reserve points. Background Technology

[0002] With the advancement of smart city construction and the increasing demands for emergency management, the site selection and layout of urban material reserve points are becoming increasingly important for achieving efficient disaster emergency response. Existing methods for selecting emergency material reserve point sites are mostly based on static planning using historical data, GIS analysis, and baseline transportation cost assessments. These methods typically deploy resources on a fixed grid or regional basis, focusing on preliminary assessments of coverage, setting transportation radii, and meeting infrastructure requirements. Such solutions often rely on manually set constraints and weights, making it difficult to dynamically reflect time-varying factors such as sudden events, changes in population distribution, and fluctuations in material demand.

[0003] Existing methods typically rely on historical data and static GIS information, lacking real-time and efficient integration of multi-source data such as mobile data, 3D information of underground pipe networks, tree growth and pruning cycles, and seasonal environmental factors. This results in site selection decisions failing to reflect the latest scene changes in a timely manner. Summary of the Invention

[0004] The purpose of this invention is to solve the problem of making real-time optimal site selection decisions under the deep coupling of multiple constraints such as dynamic spatiotemporal needs, underground pipe networks and environmental factors, and the lack of end-to-end closed-loop capability to support continuously evolving emergency deployment needs. Therefore, an intelligent site selection system for urban emergency material reserve points is proposed.

[0005] The technical solution of the present invention to solve the above-mentioned technical problems is as follows:

[0006] A smart site selection system for urban emergency supplies reserve points includes:

[0007] The data fusion and preprocessing module is used to acquire and fuse mobile terminal data, underground pipeline network three-dimensional data, tree distribution and pruning cycle, regional environmental elements and historical material delivery and distribution data from multiple sources, perform time and space alignment, outlier detection and missing value filling on the data, and output a standardized spatiotemporal feature set.

[0008] The spatiotemporal demand modeling module is used to divide the daytime rapid demand hot zone and the nighttime regular demand hot zone of the target area based on the spatiotemporal feature set, and output the hot zone weights and the distribution of demand time slices.

[0009] The safety constraint and region expansion module is used to establish a dynamic region division that couples underground pipeline distance constraints with tree growth and pruning cycles, and to generate dynamic safety margin zones and similar region partitions to ensure the initial feasibility of the candidate point set.

[0010] The optimal location and incremental learning module is used to provide a set of candidate deployment points and their ranking based on the above hot zone and regional constraints through multi-objective optimization and incremental learning updates, and to continuously update the model parameters online to adapt to data changes.

[0011] The device deployment and migration scheduling module is used to generate deployment schemes and migration instructions for edge devices based on the terminal coverage requirements of candidate deployment points, the distribution of edge computing resources, and communication latency constraints, and to ensure connectivity and coverage continuity during the migration process.

[0012] The visualization and review module is used to intuitively display the final site selection plan, regional zoning, underground pipeline distance, hot zone coverage, etc., and supports manual intervention.

[0013] Based on the above technical solution, the present invention can be further improved as follows.

[0014] Furthermore, the data fusion and preprocessing module also includes: a multi-source data alignment unit based on a spatiotemporal grid, a data quality assessment unit, and a data caching and version control unit. The alignment unit unifies data at different time granularities to a unified grid scale and aligns them to the same coordinate system. The quality assessment unit generates data quality scores through outlier detection, missing value imputation, and consistency verification. The caching and version control unit stores version information for each data integration to support reproducibility analysis.

[0015] Furthermore, the spatiotemporal demand modeling module employs a hybrid method based on DTW and kernel density estimation to identify daytime rapid demand hotspots and nighttime regular demand hotspots, outputs a set of hotspots and their time windows, and introduces seasonal weights to adaptively adjust the boundaries of the hotspots.

[0016] Furthermore, when calculating dynamic regions, the security constraint and region expansion module adopts a coupled model of the three-dimensional distance field of underground pipe network and the tree pruning cycle field. Through multi-objective constraints, it minimizes the risk of underground pipe network conflict, maximizes the coverage hot zone, and maintains the redundancy between regions.

[0017] Furthermore, the optimal location and incremental learning module includes: initial screening based on a multi-objective evolutionary algorithm, local fine-tuning based on reinforcement learning, and online learning rules based on incremental data. The incremental learning collects new data through a sliding window and updates the model parameters in a rolling manner to achieve rapid adaptation to the distribution of new demands and changes in the environment.

[0018] Furthermore, the device deployment and migration scheduling module also includes: a low-latency communication protocol between edge nodes, coverage consistency constraints, and security and compliance checks for cross-regional migration, ensuring that new deployment points maintain continuous coverage and do not create dead zones during the migration process.

[0019] Furthermore, the visualization and review module provides interactive displays of 3D GIS views, spatiotemporal heat map overlays, underground pipe network 3D models, and the security level and coverage of candidate points, supporting users to make rapid decisions based on scenarios.

[0020] Furthermore, the objective function of the optimal location and incremental learning module includes the following combination relationships: minimizing underground pipeline distance, maximizing hot zone coverage area, minimizing total transportation cost, and minimizing unit material delivery time, and their weight coefficients are expressed in formula form as follows:

[0021]

[0022] Where (w1, w2, w3, w4) are weight vectors adaptively updated through online learning, and the constraints are as follows:

[0023]

[0024] To achieve a balance between multiple objectives and obtain the final optimal set of deployment points.

[0025] Furthermore, the system employs the following three formulas at the data and model levels to describe the relationship between key constraints and objectives:

[0026] Formula 1: Safety constraint distance function for underground pipe networks

[0027] D safe (p)=m g∈G {dist(p,g)-r g}

[0028] Where p is a candidate point, G is the set of underground pipe networks, and r g This indicates the safety distance threshold for the pipeline network.

[0029] Formula 2: Hot Zone Coverage Utility Function

[0030] U cover (p)=∫ A 1 {d(p,x)≤R} dx

[0031] Where A is the area of ​​the region, R is the coverage radius, and 1 is the indicator function.

[0032] Formula 3: Multi-objective integrated objective function

[0033]

[0034] Where α, β, and γ are adaptively adjustable weights, C(p) is a comprehensive measure of transportation and maintenance costs, and D... max The maximum distance scale within the region is used to achieve comprehensive optimization.

[0035] Furthermore, the system also includes an underground pipeline safety verification engine, which is used to perform three-dimensional safety verification of the minimum distance and accessibility of candidate points to the underground pipeline network, and output safety level labels for subsequent decision-making.

[0036] The system also includes a self-testing and fault-tolerant subsystem, which automatically activates redundant data sources, maintains coverage stability, and outputs emergency alternatives in the event of sensor failure, data delay, or network interruption.

[0037] The system is implemented in a cloud-edge collaborative architecture with a programmable interface, where computationally intensive models are trained and inferred in the cloud, while the edge is responsible for real-time data acquisition, local inference, and device scheduling to achieve low-latency on-site response.

[0038] Compared with the prior art, the technical solution of this application has the following beneficial technical effects:

[0039] The data fusion and preprocessing module of this invention breaks through the limitations of single data sources, integrating and preprocessing multi-source data such as mobile terminals, 3D underground pipe networks, tree cycles, and environmental elements, solving the problem of inefficient fusion of multi-source real-time data and providing high-quality support for dynamic decision-making. The spatiotemporal demand modeling module divides day and night demand hot zones and outputs weights and time slice distributions, replacing static coverage assessment, accurately matching demand characteristics at different times, and improving the fit between site selection and actual needs. The safety constraint and area expansion module combines underground pipe network distance and tree cycles to construct dynamic area division, filling the gaps in dynamic safety factors and ensuring the feasibility of candidate points throughout the entire cycle. The optimal site selection and incremental learning module generates candidate points through multi-objective optimization and uses incremental learning to achieve online model updates, eliminating reliance on manual constraints and adapting to time-varying factors such as emergencies and personnel movement. The equipment deployment and migration scheduling module considers terminal coverage, edge computing, and communication latency to ensure the usability of site selection and dynamic adjustment. The visualization and review module intuitively displays the scheme and supports manual intervention, balancing intelligence and experience. All modules form a closed loop through data interfaces to achieve dynamic deployment, significantly improving the flexibility, safety, and emergency response efficiency of site selection. Attached Figure Description

[0040] Figure 1 This is a diagram showing the overall module structure and data flow of the system of the present invention;

[0041] Figure 2 This is a flowchart of the data processing and modeling optimization process of the present invention;

[0042] Figure 3 This is a diagram illustrating the cloud-edge collaborative architecture and deployment of this invention. Detailed Implementation

[0043] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0044] The present invention provides an intelligent site selection system for urban emergency material reserve points, comprising:

[0045] The data fusion and preprocessing module is used to acquire and fuse mobile terminal data, underground pipeline network three-dimensional data, tree distribution and pruning cycle, regional environmental elements (temperature, wind speed, humidity, diurnal temperature difference, etc.) and historical material delivery and distribution data from multiple sources. It performs time and space alignment, outlier detection and missing value filling on the data and outputs a standardized spatiotemporal feature set.

[0046] The spatiotemporal demand modeling module is used to divide the daytime rapid demand hot zone and the nighttime regular demand hot zone of the target area based on the spatiotemporal feature set, and output the hot zone weights and the distribution of demand time slices.

[0047] The safety constraint and region expansion module is used to establish a dynamic region division that couples underground pipeline distance constraints with tree growth and pruning cycles, and to generate dynamic safety margin zones and similar region partitions to ensure the initial feasibility of the candidate point set.

[0048] The optimal location and incremental learning module is used to provide a set of candidate deployment points and their ranking based on the above hot zone and regional constraints through multi-objective optimization and incremental learning updates, and to continuously update the model parameters online to adapt to data changes.

[0049] The device deployment and migration scheduling module is used to generate deployment schemes and migration instructions for edge devices based on the terminal coverage requirements of candidate deployment points, the distribution of edge computing resources, and communication latency constraints, and to ensure connectivity and coverage continuity during the migration process.

[0050] The visualization and review module is used to visually display the final site selection plan, regional zoning, underground pipeline distance, hot zone coverage, etc., and supports manual intervention;

[0051] The modules mentioned above transmit information through data interfaces to achieve end-to-end closed-loop location decision-making and dynamic deployment.

[0052] The intelligent site selection system for urban emergency material reserve points proposed in this invention is designed and implemented based on an end-to-end business process. It operates within a closed-loop framework encompassing "multi-source data fusion—spatiotemporal demand identification—safe and feasible domain construction—multi-objective optimal site selection and online evolution—deployment and migration scheduling—visualization and manual review." The system runs on a cloud-edge collaborative architecture, connecting various modules into a continuous process through a unified data interface.

[0053] The data fusion and preprocessing module also includes: a multi-source data alignment unit based on spatiotemporal grids, a data quality assessment unit, and a data caching and version control unit. The alignment unit unifies data at different time granularities to a unified grid scale and aligns them to the same coordinate system. The quality assessment unit generates data quality scores through outlier detection, missing value imputation, and consistency verification. The caching and version control unit stores version information for each data integration to support reproducibility analysis. The data fusion and preprocessing module is responsible for aggregating multi-source inputs such as mobile terminal data, underground pipeline network 3D data, tree distribution and pruning cycles, regional environmental elements (temperature, wind speed, humidity, diurnal temperature range, etc.), and historical material delivery / distribution data, and outputting a standardized spatiotemporal feature set. To ensure the reliability and reproducibility of the data foundation, the module further incorporates a multi-source data alignment unit based on a spatiotemporal grid, a data quality assessment unit, and a data caching and version control unit. The alignment unit unifies heterogeneous data to the same coordinate system and grid scale, and resamples it at a uniform time granularity. The quality assessment unit generates data quality scores through outlier detection, missing value imputation, and cross-source consistency verification. The caching and version control unit records the version fingerprint of each integration, providing a basis for reproducibility analysis and auditing. This process brings data cleaning, alignment, and quality quantification to the forefront, ensuring that subsequent modeling and optimization are based on reliable data versions.

[0054] The spatiotemporal demand modeling module employs a hybrid method based on DTW (Dynamic Time Warping) and kernel density estimation (KDE) to identify daytime rapid demand hotspots and nighttime regular demand hotspots. It outputs a set of hotspots and their time windows, and introduces seasonal weights to adaptively adjust hotspot boundaries. Supported by a high-quality feature set, the module divides the target area into daytime rapid demand hotspots and nighttime regular demand hotspots. The module uses a hybrid modeling method based on DTW and KDE: first, DTW measures the similarity of demand time series across different grids, clustering to obtain demand patterns with significant day-night differences; then, KDE spatially estimates hotspot intensity, extracting hotspot boundaries and time windows. To adapt to seasonal changes, seasonal weights are introduced to incorporate environmental factors (such as high temperature, extreme wind speed, or humidity) into boundary adaptation and weight redistribution, allowing hotspots to dynamically fine-tune with the seasons, outputting a structured result containing a set of hotspots, time windows, and weights.

[0055] The safety constraint and region expansion module employs a coupled model of the underground pipeline network's three-dimensional distance field and the tree pruning cycle field when calculating dynamic regions. Through multi-objective constraints, it minimizes the risk of underground pipeline network conflicts and maximizes coverage hotspots while maintaining redundancy between regions. The system then enters the safety constraint and region expansion phase, establishing dynamic constraints on the underground pipeline network and the greening pruning cycle, generating dynamic safety margin zones and similar region partitions to ensure the initial feasibility of candidate points. To this end, a three-dimensional distance field of the underground pipeline network is constructed to calculate the margin of minimum safe distance from candidate points to different types of pipelines (such as gas, electricity, and communication), and coupled with the tree pruning cycle field to form a time-varying feasible region. This coupled model balances the optimization objectives of "minimizing the risk of underground pipeline network conflicts and maximizing coverage hotspots" and enhances robustness by maintaining redundancy between regions, ensuring that candidate points remain feasible throughout the entire cycle under different time periods and maintenance windows.

[0056] The optimal location and incremental learning module includes: initial screening based on a multi-objective evolutionary algorithm, local fine-tuning based on reinforcement learning, and online learning rules based on incremental data. The incremental learning collects new data through a sliding window and updates model parameters on a rolling basis to achieve rapid adaptation to new demand distributions and environmental changes. Within the candidate feasible domain, the optimal location and incremental learning module performs multi-stage collaborative optimization. The system first uses a multi-objective evolutionary algorithm for initial screening, constructing a family of objectives based on maximizing coverage utility, minimizing underground pipeline safety risks, minimizing transportation costs, and minimizing unit material delivery time, generating a Pareto front solution set. Then, in the simulation environment, local fine-tuning is performed using reinforcement learning, improving practical performance through micro-displacement, capacity allocation, and connectivity adjustments. Finally, online updates are performed using a sliding window based on new data, with rolling calibration of hot zone weights, seasonal weights, and cost parameters to quickly adapt to new demand distributions and environmental changes.

[0057] The device deployment and migration scheduling module also includes: low-latency communication protocols between edge nodes, coverage consistency constraints, and security and compliance checks for cross-regional migration. This ensures that new deployment points maintain continuous coverage without dead zones during migration. After optimizing and sorting the candidate deployment point set, the module generates deployment plans and migration instructions based on terminal coverage requirements, edge computing resource distribution, and communication latency constraints. During the migration period, a "shadow node" strategy is used to ensure connectivity and coverage continuity. The module enables low-latency communication protocols between edge nodes to enhance reliable instruction delivery and redundant transmission. Simultaneously, it performs security and compliance checks on cross-regional migration, including temporary road occupancy, vehicle routes, electromagnetic compatibility, and noise limitations, transforming the plan from theoretical feasibility into practical engineering feasibility.

[0058] The visualization and review module provides interactive displays of 3D GIS views, spatiotemporal heat map overlays, underground pipe network 3D models, and the safety levels and coverage of candidate points. This supports users in making rapid, scenario-based decisions. Before and after scheme release and execution, the visualization and review module provides 3D GIS views, interactively displaying the underground pipe network 3D model, spatiotemporal heat map overlays, and the safety levels and coverage of candidate points, supplemented by a timeline control demonstrating coverage changes during day / night and seasonal transitions. Decision-makers can perform parameterized interventions within the interface, such as adjusting heat map thresholds, safety distances, and seasonal weights, triggering incremental evaluation and rapid recalculation in the backend. This human-machine collaborative closed loop enhances the transparency and interpretability of the scheme while also allowing professional experience to be applied to intelligent decision-making.

[0059] The objective function of the optimal location and incremental learning module includes the following combination relationships: minimizing underground pipeline distance, maximizing hot zone coverage area, minimizing total transportation cost, and minimizing unit material delivery time, and their weight coefficients are expressed in formula form as follows:

[0060] W = [w1w2w3w4]T

[0061] Where (w1, w2, w3, w4) are weight vectors adaptively updated through online learning, and the constraints are as follows:

[0062]

[0063] To achieve a multi-objective trade-off and obtain the final optimal deployment point set, this invention uses multi-objective optimization and online weight updates to achieve real-time trade-offs between different objectives, centered around the objective function. The system uses "minimizing underground pipeline distance (equivalent to maximizing safety margin), maximizing hot zone coverage area, minimizing total transportation cost, and minimizing unit material delivery time" as its objective family, represented by the vector W = [w1w2w3w4]. T Express weights, satisfy During the online operation period, the weights are adaptively updated based on external performance feedback (such as coverage, 95th percentile delivery latency, response compliance rate, etc.), enabling the system to automatically adjust and optimize preferences in different stages and scenarios.

[0064] The system employs the following three formulas at the data and model levels to describe the relationship between key constraints and objectives:

[0065] Formula 1: Safety constraint distance function for underground pipe networks

[0066] D safe (p)=m g∈G {dist(p,g)-r g}

[0067] Where p is a candidate point, G is the set of underground pipe networks, and r g This indicates the safety distance threshold for the pipeline network.

[0068] Formula 2: Hot Zone Coverage Utility Function

[0069] U cover (p)=∫ A 1 {d(p,x)≤R} dx

[0070] Where A is the area of ​​the region, R is the coverage radius, and 1 is the indicator function.

[0071] Formula 3: Multi-objective integrated objective function

[0072]

[0073] Where α, β, and γ are adaptively adjustable weights, C(p) is a comprehensive measure of transportation and maintenance costs, and D... max To achieve comprehensive optimization, the system uses the maximum distance scale within the region. To transform constraints and objectives into a calculable and comparable indicator system, the system implements three core formulas at the data and model levels, which are integrated throughout the optimization and evaluation process. Firstly, the underground pipeline network safety constraint distance function D... safe (p)=m g∈G {dist(p,g)-r g} is used to determine whether candidate points meet the minimum safe distance and to quantify the safety margin; secondly, the hot zone coverage utility function U cover (p)=∫ A 1 {d(p,x)≤R} dx is used to evaluate the utility of hot zones within the coverage area, and can be weighted by combining time period and seasonal weights; thirdly, a multi-objective comprehensive objective function. The coverage, standardized safety margin, and overall cost C(p) are unified within an optimizable framework. The above three equations correspond one-to-one with claim 9 and serve as the basic operators for constraint checking and target scoring in the system, supporting batch and incremental calculations.

[0074] The system also includes an underground pipeline safety verification engine, which is used to perform three-dimensional safety verification of the minimum distance and accessibility of candidate points to underground pipelines, and output safety level labels for subsequent decision-making.

[0075] The system also includes a self-testing and fault-tolerant subsystem, which automatically activates redundant data sources, maintains coverage stability, and outputs emergency alternatives in the event of sensor failure, data delay, or network interruption.

[0076] The system is implemented using a cloud-edge collaborative architecture with a programmable interface. Computationally intensive models are trained and inferred in the cloud, while the edge handles real-time data acquisition, local inference, and equipment scheduling to achieve low-latency on-site response. For safety and reliability, the system is equipped with an underground pipeline safety verification engine, a self-checking and fault-tolerant subsystem, and enhances its practical resilience through a cloud-edge collaborative implementation. The safety verification engine receives candidate point coordinates, the underground pipeline model, and pipeline type thresholds, performs 3D nearest neighbor queries and reachability checks, and outputs a safety level label including minimum boundary distance, risk annotations, and rectification suggestions to assist in final decision-making. The self-checking and fault-tolerant subsystem automatically activates redundant data sources, maintains stable coverage, and outputs emergency alternatives in the event of sensor failure, data latency, or network interruption. The edge has offline inference and local caching capabilities, and performs difference merging and consistency verification after network recovery. In the cloud-edge collaborative architecture, the cloud handles training and global recalculation, while the edge handles real-time data acquisition, local inference, and scheduling execution. Both communicate via a programmable interface for parameter distribution and log feedback.

[0077] During the system's daily operation, the cloud performs rolling reassessments of seasonal weights and hotspot boundaries in the early morning hours. During the day, the edge side updates demand slices hourly and triggers local recalculation when distribution shifts occur. When a sudden increase in population or an event in a specific area causes a DTW (Demand-Driven Warp) shift alarm, the system increases the daytime rapid demand weight for that area and initiates incremental optimization. The deployment module generates a temporary migration plan, which is then deployed after being prompted by the visualization module and confirmed manually. Simultaneously, the security verification engine performs 3D verification on newly added candidate points. If the security level is below the threshold, it automatically performs a second evaluation according to the recommended micro-displacement direction and amplitude until the deployment standard is met. This operational example demonstrates the sequential connection and mutual constraints of each module in the closed loop, reflecting the overall design philosophy of this invention: "data-driven, constraint-first, goal-coordinated, online evolution, and engineering implementation."

[0078] Regarding parameters and implementation points, the spatial grid resolution can be selected between 500m and 1km based on the city density, and the time granularity is recommended to be 5-15 minutes to balance timeliness and throughput; the coverage radius R is set according to the material type and service timeliness; the regional redundancy is not less than 2, and can be set to 3 in the core urban area to enhance resilience; in online learning, the total variation constraint and cold start protection are set for the update amplitude of weights and thresholds to avoid frequent oscillations.

[0080] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0081] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An intelligent site selection system for urban emergency material reserve points, characterized in that, include: The data fusion and preprocessing module is used to acquire and fuse mobile terminal data, underground pipeline network three-dimensional data, tree distribution and pruning cycle, regional environmental elements and historical material delivery and distribution data from multiple sources, perform time and space alignment, outlier detection and missing value filling on the data, and output a standardized spatiotemporal feature set. The spatiotemporal demand modeling module is used to divide the daytime rapid demand hot zone and the nighttime regular demand hot zone of the target area based on the spatiotemporal feature set, and output the hot zone weights and the distribution of demand time slices. The safety constraint and region expansion module is used to establish a dynamic region division that couples underground pipeline distance constraints with tree growth and pruning cycles, and to generate dynamic safety margin zones and similar region partitions to ensure the initial feasibility of the candidate point set. The optimal location and incremental learning module is used to provide a set of candidate deployment points and their ranking based on the above hot zone and regional constraints through multi-objective optimization and incremental learning updates, and to continuously update the model parameters online to adapt to data changes. The device deployment and migration scheduling module is used to generate deployment schemes and migration instructions for edge devices based on the terminal coverage requirements of candidate deployment points, the distribution of edge computing resources, and communication latency constraints, and to ensure connectivity and coverage continuity during the migration process. The visualization and review module is used to intuitively display the final site selection plan, regional zoning, underground pipeline distance, hot zone coverage, etc., and supports manual intervention.

2. The intelligent site selection system for urban emergency material reserve points according to claim 1, characterized in that, The data fusion and preprocessing module also includes: a multi-source data alignment unit based on spatiotemporal grids, a data quality assessment unit, and a data caching and version control unit. The alignment unit unifies data at different time granularities to a unified grid scale and aligns them to the same coordinate system. The quality assessment unit generates data quality scores through outlier detection, missing value imputation, and consistency verification. The caching and version control unit stores version information for each data integration to support reproducibility analysis.

3. The intelligent site selection system for urban emergency material reserve points according to claim 1, characterized in that, The spatiotemporal demand modeling module uses a hybrid method based on DTW and kernel density estimation to identify daytime rapid demand hotspots and nighttime regular demand hotspots, outputs a set of hotspots and their time windows, and introduces seasonal weights to adaptively adjust the boundaries of the hotspots.

4. The intelligent site selection system for urban emergency material reserve points according to claim 1, characterized in that, When calculating dynamic regions, the security constraint and region expansion module adopts a coupled model of the three-dimensional distance field of underground pipe network and the tree pruning cycle field. Through multi-objective constraints, it minimizes the risk of underground pipe network conflict, maximizes the coverage of hot areas, and maintains the redundancy between regions.

5. The intelligent site selection system for urban emergency material reserve points according to claim 1, characterized in that, The optimal location and incremental learning module includes: initial screening based on a multi-objective evolutionary algorithm, local fine-tuning based on reinforcement learning, and online learning rules based on incremental data. The incremental learning collects new data through a sliding window and updates the model parameters in a rolling manner to achieve rapid adaptation to the distribution of new demands and changes in the environment.

6. The intelligent site selection system for urban emergency material reserve points according to claim 1, characterized in that, The device deployment and migration scheduling module also includes: a low-latency communication protocol between edge nodes, coverage consistency constraints, and security and compliance checks for cross-regional migration, ensuring that new deployment points maintain continuous coverage and do not create dead zones during the migration process.

7. The intelligent site selection system for urban emergency material reserve points according to claim 1, characterized in that, The visualization and review module provides interactive displays of 3D GIS views, spatiotemporal hotspot overlays, underground pipe network 3D models, and the security level and coverage of candidate points, supporting users to make rapid decisions based on scenarios.

8. The intelligent site selection system for urban emergency material reserve points according to claim 5, characterized in that, The objective function of the optimal location and incremental learning module includes the following combination relationships: minimizing underground pipeline distance, maximizing hot zone coverage area, minimizing total transportation cost, and minimizing unit material delivery time, and their weight coefficients are expressed in formula form as follows: Where (w1, w2, w3, w4) are weight vectors adaptively updated through online learning, and the constraints are as follows: To achieve a balance between multiple objectives and obtain the final optimal set of deployment points.

9. A smart site selection system for urban emergency material reserve points according to any one of claims 1 to 8, characterized in that, The system employs the following three formulas at the data and model levels to describe the relationship between key constraints and objectives: Formula 1: Safety constraint distance function for underground pipe networks D safe (p)=m g∈G {dist(p,g)-r g } Where p is a candidate point, G is the set of underground pipe networks, and r g This indicates the safety distance threshold for the pipeline network. Formula 2: Hot Zone Coverage Utility Function U cover (p)=∫ A 1 {d(p,x)≤R} dx Where A is the area of ​​the region, R is the coverage radius, and 1 is the indicator function. Formula 3: Multi-objective integrated objective function Where α, β, and γ are adaptively adjustable weights, C(p) is a comprehensive measure of transportation and maintenance costs, and D... max The maximum distance scale within the region is used to achieve comprehensive optimization.

10. The intelligent site selection system for urban emergency material reserve points according to claim 9, characterized in that, The system also includes an underground pipeline safety verification engine, which is used to perform three-dimensional safety verification of the minimum distance and accessibility of candidate points to underground pipelines, and output safety level labels for subsequent decision-making. The system also includes a self-testing and fault-tolerant subsystem, which automatically activates redundant data sources, maintains coverage stability, and outputs emergency alternatives in the event of sensor failure, data delay, or network interruption. The system is implemented in a cloud-edge collaborative architecture with a programmable interface, where computationally intensive models are trained and inferred in the cloud, while the edge is responsible for real-time data acquisition, local inference, and device scheduling to achieve low-latency on-site response.

Images