Smart city emergency response method and system combined with artificial intelligence

Abstract

The application provides a smart city emergency response method and system combined with artificial intelligence, and relates to the technical field of artificial intelligence. After receiving a city emergency event trigger signal containing geographic coordinate coding and time coding of an event occurrence, a city space grid identification sequence is generated through a spatial positioning branch module of an emergency response decision network. A related emergency resource unit descriptor set is extracted using a resource association analysis module. A propagation path probability distribution map is generated with the help of an event evolution prediction module. Finally, an emergency response scheduling instruction set containing a resource scheduling time window and a spatial path point column is generated. The application can quickly and accurately locate the event position, associate emergency resources, predict the event propagation path, realize accurate dynamic scheduling of emergency resources, and improve the city emergency response capability.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and more specifically, to a smart city emergency response method and system that incorporates artificial intelligence. Background Technology

[0002] In the process of building and developing smart cities, urban emergency response capabilities are a key element in ensuring the safe and stable operation of cities. With the acceleration of urbanization, the scale of cities continues to expand, and the population becomes increasingly dense. The frequency of various emergencies such as natural disasters, public health incidents, and safety accidents has increased, and their scope of impact and degree of harm have become increasingly complex and diverse.

[0003] Traditional urban emergency response methods primarily rely on manual decision-making and pre-established emergency plans. When faced with emergencies, decision-makers must manually locate and assess the status of emergency resources based on limited information and experience, and formulate corresponding deployment plans. However, these methods have several limitations. On the one hand, the manual decision-making process is time-consuming, making it difficult to react quickly in emergencies and potentially missing the optimal time for emergency response. On the other hand, pre-established emergency plans often fail to cover all possible contingencies, lacking flexibility and adaptability, and unable to dynamically adjust emergency strategies according to the actual development of the event. Furthermore, traditional methods lack effective integration and analysis tools when processing information in spatial and temporal dimensions, making it difficult to accurately predict the propagation path and scope of impact of events, resulting in inaccurate deployment of emergency resources and causing resource waste or inadequate emergency response. Summary of the Invention

[0004] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a smart city emergency response method incorporating artificial intelligence, the method comprising: Receive urban emergency event trigger signal, wherein the urban emergency event trigger signal includes the geographic coordinate code of the event occurrence and the event occurrence time code; The spatial positioning branch module of the emergency response decision network is activated based on the geographic coordinate code of the event occurrence, and spatial grid mapping processing is performed on the geographic coordinate code of the event occurrence to generate a city spatial grid identifier sequence corresponding to the geographic coordinate code of the event occurrence. The urban spatial grid identifier sequence is input into the resource association analysis module of the emergency response decision network. Based on the urban spatial grid identifier sequence, the emergency resource deployment database is traversed to extract the set of emergency resource unit descriptors that are directly related to the urban spatial grid identifier sequence. Activate the event evolution prediction module of the emergency response decision network, and perform temporal dependency modeling on the emergency resource unit descriptor set based on the emergency resource unit descriptor set and in conjunction with the event occurrence time code, to generate a probability distribution map of the propagation path of emergency events in the urban spatial grid identification sequence; A smart city emergency response dispatch instruction set is generated based on the propagation path probability distribution map. The smart city emergency response dispatch instruction set includes a resource dispatch time window and a resource dispatch spatial path point list for each emergency resource unit in the emergency resource unit descriptor set.

[0005] Furthermore, embodiments of the present invention also provide a smart city emergency response system incorporating artificial intelligence, characterized in that it includes: A processor; a machine-readable storage medium for storing machine-executable instructions of the processor; wherein the processor is configured to execute the aforementioned smart city emergency response method incorporating artificial intelligence by executing the machine-executable instructions.

[0006] Based on the above, by receiving urban emergency event trigger signals containing geographic coordinate codes and time codes of the event occurrence, and utilizing the spatial positioning branch module of the emergency response decision network for spatial grid mapping processing, the specific location of the event in the urban space can be quickly and accurately determined, generating a corresponding urban spatial grid identifier sequence. This sequence is then input into the resource association analysis module, which traverses the emergency resource deployment database to extract relevant emergency resource unit descriptor sets, achieving a precise association between emergency resources and the event's location, ensuring the targeted and effective scheduling of emergency resources. Activating the event evolution prediction module, based on the emergency resource unit descriptor set and the event occurrence time code, models the temporal dependency relationship, generating a probability distribution map of the propagation path of the emergency event in the urban spatial grid identifier sequence. This allows for the prediction of the event's development trend and impact range in advance. Finally, based on the propagation path probability distribution map, a smart city emergency response scheduling instruction set containing resource scheduling time windows and spatial path point sequences is generated, achieving precise scheduling and dynamic adjustment of emergency resources. This significantly improves the efficiency and accuracy of urban emergency response and effectively reduces the losses caused by emergencies to the city. Attached Figure Description

[0007] Figure 1 This is a schematic diagram of the execution flow of the smart city emergency response method combining artificial intelligence provided in the embodiments of the present invention.

[0008] Figure 2 This is a schematic diagram of exemplary hardware and software components of a smart city emergency response system that incorporates artificial intelligence, provided in an embodiment of the present invention. Detailed Implementation

[0009] Figure 1 This is a flowchart illustrating a smart city emergency response method incorporating artificial intelligence, provided in one embodiment of the present invention. A detailed description follows.

[0010] Step S110: Receive an urban emergency event trigger signal, wherein the urban emergency event trigger signal includes the geographic coordinate code of the event occurrence and the event occurrence time code.

[0011] In this embodiment, the urban emergency event trigger signal is collected in real time through the distributed sensor network of the urban emergency monitoring system. The signal is encapsulated in a structured binary format, containing a 128-byte fixed header and a variable-length data body. The geographic coordinates of the event occurrence are encoded using planar coordinates in the Gauss-Kruger projection coordinate system, consisting of eastward and northward coordinate components. Each component occupies 8 bytes of double-precision floating-point data, in meters, accurate to three decimal places. The event occurrence time is encoded using Coordinated Universal Time (UTC), in the format of year-month-day T hour:minute:second.millisecond, stored as a 64-bit timestamp integer, accurate to the millisecond level. The signal header contains metadata such as an event type identifier (4 bytes), a signal source device number (8 bytes), and a data checksum (16 bytes). When an emergency event (such as a building fire) occurs in a certain area of ​​the city, temperature sensors, smoke detectors, and video surveillance equipment deployed in that area will simultaneously trigger alarm signals. After being aggregated and processed by edge computing nodes, a standard format urban emergency event trigger signal is generated and transmitted to the emergency response decision system via a 5G private network.

[0012] Step S120: Activate the spatial positioning branch module of the emergency response decision network according to the geographic coordinate code of the event occurrence, perform spatial grid mapping processing on the geographic coordinate code of the event occurrence, and generate a city spatial grid identifier sequence corresponding to the geographic coordinate code of the event occurrence.

[0013] The spatial positioning branch module of the emergency response decision-making network adopts a modular design, including a coordinate resolution unit, a grid index conversion unit, a neighborhood retrieval unit, and a hierarchical coding unit. Upon receiving the geographic coordinate code of an event, the system first activates the coordinate resolution unit to perform range verification on the eastward and northward coordinate components, ensuring they fall within the boundaries of the city's administrative area. The valid coordinate data is then passed to the grid index conversion unit, which loads the metadata of the city's spatial grid division reference system (including grid origin coordinates, grid size, hierarchical division rules, etc.) and converts the geographic coordinates into grid index values ​​through linear transformation. The neighborhood retrieval unit uses the central grid as a reference and searches for surrounding grid units according to an eight-direction neighborhood expansion algorithm, generating a grid set containing multiple levels of neighborhoods. The hierarchical coding unit then performs hierarchical coding on the aforementioned grid units, forming an identifier sequence with spatial inclusion relationships. This sequence serves as the core input for the subsequent resource association analysis module, and its data structure is stored in a linked list format. Each node contains a grid identifier, hierarchical depth, parent node pointer, and a list of child nodes.

[0014] Step S121: Analyze the geographic coordinate code of the event occurrence carried by the urban emergency event trigger signal, and extract the longitude coordinate component and latitude coordinate component from the geographic coordinate code of the event occurrence.

[0015] The input for this step is the data body of the city emergency event trigger signal, where the geographic coordinates are stored as a continuous byte stream. The parsing process uses a bit-by-bit reading method. First, the starting offset of the coordinate data is located (20th byte after the header), and 16 bytes of data are read continuously. The first 8 bytes are the eastward coordinate component (longitude), and the last 8 bytes are the northward coordinate component (latitude). Each coordinate component is represented using IEEE 754 standard double-precision floating-point numbers, and byte order conversion (from network byte order to host byte order) is required during parsing. After parsing, the coordinate values ​​are validated, specifically: checking whether the coordinate values ​​are within the preset city boundary coordinate range, and calculating whether the checksum of the coordinate values ​​matches the checksum in the signal header. If the validation fails, a signal retransmission mechanism is triggered; if the validation passes, the two coordinate components are stored as a 32-bit floating-point array as input for the next step of grid index conversion.

[0016] Step S122: Input the longitude and latitude coordinate components into the spatial positioning branch module, perform grid row index value conversion on the longitude coordinate components according to the urban spatial grid division reference system, and perform grid column index value conversion on the latitude coordinate components according to the urban spatial grid division reference system, and output the grid row index value corresponding to the longitude coordinate components and the grid column index value corresponding to the latitude coordinate components.

[0017] The urban spatial grid division reference system adopts a multi-level quadtree structure, with different grid sizes at each level. In this embodiment, the third-level grid (grid size A meters × A meters) is used for initial mapping. During the conversion process, the grid division reference system parameters are first loaded from the system configuration file, including the longitude coordinates (Lon0), latitude coordinates (Lat0), and grid size (D) of the grid origin. For the longitude coordinate component (Lon), the grid row index value is calculated as (Lon - Lon0) divided by D and then rounded down. For the latitude coordinate component (Lat), the grid column index value is calculated as (Lat - Lat0) divided by D and then rounded down. To avoid floating-point operation errors, 64-bit fixed-point arithmetic is used during the calculation process. The coordinate difference is first multiplied by 10000 and rounded down before the division operation. After the conversion is completed, two 32-bit integer grid row index values ​​and grid column index values ​​are output, for example, the row index value is R and the column index value is C, which together locate the initial grid cell where the event occurred.

[0018] Step S123: Using the grid cell located by both the grid row index value and the grid column index value as the center, search for adjacent grid cells layer by layer outward according to the preset number of neighborhood expansion layers, and generate a set of adjacent grid identifiers that includes the identifier of the grid cell itself and the identifiers of all adjacent grid cells.

[0019] The preset neighborhood expansion layer is 3 layers, and a breadth-first search algorithm is used for layer-by-layer retrieval. The identifier of the central grid cell is (R, C). The first layer of neighborhood includes grid cells in four directions: up (R-1, C), down (R+1, C), left (R, C-1), and right (R, C+1). The second layer of neighborhood includes grid cells in four diagonal directions: upper left (R-1, C-1), upper right (R-1, C+1), lower left (R+1, C-1), and lower right (R+1, C+1). The third layer of neighborhood extends outward by one grid unit from the second layer. During the retrieval process, it is necessary to check whether the index value of each grid cell exceeds the city grid range. For boundary grid cells, only adjacent grids within the valid range are retained. The set of adjacent grid identifiers is stored using a hash table structure, with the grid identifier (R, C) as the key and the distance weight between the grid and the central grid as the value (the closer the distance, the higher the weight). The final set of adjacent grid identifiers contains 1+4+4+8=17 grid cell identifiers, including the central grid (assuming the central grid is not a boundary grid).

[0020] Step S124: Combine and encode the grid row index value, the grid column index value, and the set of adjacent grid identifiers to form an initial grid identifier corresponding to the geographic coordinate code of the event occurrence. The initial grid identifier records the spatial inclusion relationship of the grid row index value, the grid column index value, and each adjacent grid unit identifier in the set of adjacent grid identifiers in a hierarchical encoding structure.

[0021] The combined encoding process employs a space-filling encoding method based on Z-order curves, converting the two-dimensional grid index (R, C) into a one-dimensional encoded value. Specifically, the binary representations (32 bits each) of the row index value R and the column index value C are bit-crossed to obtain a 64-bit Z-order code as the unique identifier for the initial grid. For the set of adjacent grid identifiers, a tree structure is used for encoding, with the central grid as the root node, the first-level neighboring grids as first-level child nodes, the second-level neighboring grids as second-level child nodes, and so on. Each node stores the number of child nodes, adjacent grid identifiers, and relative direction vectors. The final data structure of the initial grid identifier is a composite structure containing the Z-order code (8 bytes), the level depth (1 byte), the number of child nodes (1 byte), and the child node list (variable length), with a total length not exceeding 256 bytes.

[0022] Step S125: Traverse the urban spatial grid hierarchy tree, extract the set of upper-level grid identifiers that have a direct parent-level relationship with the initial grid identifier from the urban spatial grid hierarchy tree, and extract the set of lower-level grid identifiers that have a direct child-level relationship with the initial grid identifier from the urban spatial grid hierarchy tree.

[0023] The urban spatial grid hierarchy tree uses a B+ tree storage structure. Each node contains a grid identifier, a level number, a parent node pointer, child node pointers, and spatial extent information. The traversal process starts from the node corresponding to the initial grid identifier and traces upwards to the root node, extracting the grid identifier of its parent node (upper-level grid identifier). This upper-level grid typically corresponds to a higher-level (larger-size) grid cell. Moving downwards, it extracts all child node identifiers of the current node (lower-level grid identifiers), which correspond to lower-level (smaller-size) grid cells. Both the upper-level and lower-level grid identifier sets are stored using dynamic arrays. The size of the upper-level grid identifier set depends on the level depth of the initial grid (e.g., a third-level grid has 3 upper-level grids), while the size of the lower-level grid identifier set is fixed at 4 (each node in a quadtree structure has 4 child nodes).

[0024] Step S126: Perform hierarchical aggregation, semantic enhancement, sequence sorting, compression encoding and time binding processing based on the initial grid identifier, and output a city spatial grid identifier sequence with time validity mark.

[0025] Hierarchical aggregation merges the upper-level grid identifier set, the initial grid identifier set, and the lower-level grid identifier set into a complete grid hierarchy list, arranged in descending order of hierarchy (i.e., grid size from largest to smallest). Semantic enhancement adds attribute labels to each grid identifier by calling an urban functional area identification model. This model uses a convolutional neural network architecture, taking satellite image features of the grid as input and outputting functional area type (e.g., residential, commercial, industrial, etc.) and infrastructure density level (levels 1-5). Sequence sorting is performed according to grid hierarchy and spatial location, forming an ordered sequence from macro to micro. Compression encoding uses a run-length encoding algorithm to remove duplicate attribute information between adjacent grids. Time binding processing encodes the event occurrence time into a timestamp, appended to the sequence header as a validity marker. The final generated urban spatial grid identifier sequence is a JSON format string containing three fields: sequence length, timestamp, and grid identifier array.

[0026] Step S1261: Based on spatial inclusion relationships, logically aggregate the upper-level grid identifier set, the initial grid identifier, and the lower-level grid identifier set to generate a complete spatial grid hierarchical identifier list corresponding to the geographic coordinate code of the event occurrence. The identifiers in the complete spatial grid hierarchical identifier list are organized and arranged in order from macro to micro according to the spatial scale of their corresponding grids.

[0027] Spatial containment relationships are determined through the parent-child node relationships in the grid hierarchy tree. Each grid in the upper-level grid identifier set contains the initial grid, and the initial grid contains each grid in the lower-level grid identifier set. The logical aggregation process first creates an empty list, then sequentially adds the highest-level upper-level grid identifier, the second-highest-level upper-level grid identifier, the initial grid identifier, the first-level lower-level grid identifier, and the second-level lower-level grid identifier. For example, if the initial grid is at the third level, the order of the complete spatial grid hierarchy identifier list is: first-level grid (city level), second-level grid (region level), third-level grid (initial grid), fourth-level grid (sub-grids 1-4), and fifth-level grid (sub-sub-grids 1-16). Each grid identifier in the list includes its hierarchy number, Z-order code, and spatial extent coordinates.

[0028] Step S1262: Perform a semantic label mapping operation on each grid identifier in the complete spatial grid hierarchical identifier list. Add a corresponding urban functional area type label to each grid identifier through a pre-trained urban functional area identification model. At the same time, add an infrastructure density level descriptor to each grid identifier according to the infrastructure distribution density of the area corresponding to the grid identifier, so as to obtain a spatial grid identifier unit with enhanced semantics.

[0029] The urban functional zone identification model adopts the ResNet-50 architecture. The input is a 256×256 pixel grid area satellite image, and the output is an 8-dimensional functional zone type probability vector. The final type label is determined by the argmax function. This model was trained on an urban functional zone dataset containing 100,000 labeled images, using SGD as the optimizer, an initial learning rate of 0.01, 200 training epochs, and achieving a Top-1 accuracy of 89%. The infrastructure density level descriptor is calculated by counting the number of infrastructure POIs (Points of Interest) within the grid area. Specifically, the density level equals the number of POIs divided by the grid area (unit: points / km²), then mapped to an integer level of 1-5. The semantically enhanced spatial grid identifier unit extends the original grid identifier with two fields: a functional zone type label (string, such as "residential") and an infrastructure density level (integer, 1-5).

[0030] Step S1263: Sort all the spatial grid identification units with enhanced semantics according to the hierarchical order of spatial inclusion relationship to form a city spatial grid identification sequence with a progressive arrangement structure from the upper macro grid to the lower micro grid.

[0031] The sorting process employs a stable sorting algorithm, using grid level number as the primary sorting key (ascending order) and Z-order encoding as the secondary sorting key (ascending order). For grid cells at the same level, they are sorted according to their spatial position within the parent grid (top left, top right, bottom left, bottom right). The sorted urban spatial grid identifier sequence is an array, with a length equal to the number of grids in the complete spatial grid level identifier list. Each element is a structure containing a grid identifier, functional zone type label, and infrastructure density level. The first element of the sequence represents the highest-level (macro) grid, the last element represents the lowest-level (micro) grid, and the intermediate elements are arranged sequentially according to level.

[0032] Step S1264: Perform redundancy compression encoding on the urban spatial grid identifier sequence, identify and remove duplicate urban functional area type label information and duplicate infrastructure density level descriptor information between adjacent enhanced semantic spatial grid identifier units in the urban spatial grid identifier sequence, and generate a compressed urban spatial grid identifier sequence.

[0033] Redundancy compression coding employs a differential coding strategy. For urban functional area type labels, only label values ​​different from the previous grid cell are stored; identical labels are represented by a special marker (0x00). For infrastructure density level descriptors, the difference between the current value and the previous value is stored (represented by a 1-byte signed integer). During compression, the urban spatial grid identifier sequence is first traversed, recording the label and level value of each element. Then, it is compared with the previous element to generate a differential coded stream. The compressed sequence contains the original grid identifier array and two differential coded arrays (label differential array and level differential array), reducing the total data volume by approximately 40% compared to the original sequence.

[0034] Step S1265: Perform a timestamp binding operation between the compressed urban spatial grid identifier sequence and the event occurrence time code in the urban emergency event triggering signal, add a time validity mark to the compressed urban spatial grid identifier sequence, and output the urban spatial grid identifier sequence with the time validity mark.

[0035] The timestamp binding operation encodes the event occurrence time into a UNIX timestamp (the number of seconds since January 1, 1970), stored as a 64-bit integer. This timestamp is added to the header metadata of the compressed urban spatial grid identifier sequence as a time validity marker. Simultaneously, the system calculates the sequence's validity period (defaulting to 24 hours after the event) and stores it as the end value of the timestamp. The final output urban spatial grid identifier sequence with time validity markers is in binary format, containing: a header (16 bytes, including the timestamp start value, end value, and sequence length), a grid identifier array (variable length), a label difference array (variable length), and a rank difference array (variable length).

[0036] Step S130: Input the urban spatial grid identifier sequence into the resource association analysis module of the emergency response decision network, traverse the emergency resource deployment database according to the urban spatial grid identifier sequence, and extract the set of emergency resource unit descriptors that are directly related to the urban spatial grid identifier sequence.

[0037] The resource correlation analysis module of the emergency response decision-making network consists of a feature extraction unit, a query construction unit, a database interaction unit, and a result processing unit. The urban spatial grid identifier sequence is first decompressed and semantically parsed by the feature extraction unit to recover the functional area type label and infrastructure density level of each grid unit. The query construction unit generates structured query conditions based on these features, including spatial extent conditions (based on grid coordinates), resource type conditions (based on functional area labels), and quantity conditions (based on density levels). The database interaction unit connects to the emergency resource deployment database (a PostgreSQL database containing resource, location, and status tables) using the JDBC protocol, executes query commands, and returns a result set. The result processing unit filters, sorts, and encapsulates the returned resource records to generate a set of emergency resource unit descriptors. This set is stored in a linked list structure, with each descriptor containing fields such as resource ID, type code, geographic coordinates, current status, and available quantity.

[0038] Step S131: Sequentially extract the urban functional area type label, infrastructure density level descriptor, grid row index value, and grid column index value contained in each enhanced semantic spatial grid identifier unit from the urban spatial grid identifier sequence.

[0039] The extraction process begins at the head of the urban spatial grid identifier sequence. First, the time validity marker is read and the current time is verified to be within the validity period. Then, the grid identifier array is parsed, and the row and column indices of each grid are obtained through Z-order decoding. Next, the label difference array and the level difference array are decoded to recover the urban functional area type label and infrastructure density level descriptor for each grid. Specifically, for the label difference array, the current label value is updated and recorded when a non-0x00 marker is encountered; for the level difference array, each difference is added to the previous level value to obtain the current level value. The extracted results are stored in a two-dimensional array, with each row corresponding to an enhanced semantic spatial grid identifier unit, containing four elements: row index, column index, functional area label, and density level.

[0040] Step S132: Input the urban functional area type label and infrastructure density level descriptor of each enhanced semantic spatial grid identification unit into the resource association analysis module. Match the corresponding emergency resource category code set for the urban functional area type label according to the preset urban functional area emergency resource mapping table. At the same time, match the corresponding resource deployment density threshold range for the infrastructure density level descriptor according to the infrastructure density level descriptor.

[0041] The urban functional area emergency resource mapping table is a JSON-formatted configuration file stored in the local cache of the resource association analysis module. The table's structure is as follows: the key is the urban functional area type label (e.g., "residential"), and the value is an array containing commonly used emergency resource category codes for that functional area (e.g., "fire_extinguisher", "ambulance", etc.). The correspondence between infrastructure density level descriptors and resource deployment density threshold ranges is defined through a lookup table. For example, density level 1 corresponds to a threshold range of 0-5 resources / km², level 2 corresponds to 6-15 resources / km², and so on. The matching process uses a hash lookup algorithm. For each spatial grid identifier unit with enhanced semantics, it first looks up the resource category code set corresponding to its functional area label, then determines the threshold range for the resource quantity based on the density level, and finally generates the resource query conditions (category code + quantity range) for each grid unit.

[0042] Step S133: Convert the grid row index value and grid column index value into a spatial query window coordinate range. The spatial query window coordinate range covers the grid cells located by the grid row index value and grid column index value, as well as the adjacent grid cells located by all adjacent grid cell identifiers in the adjacent grid identifier set.

[0043] The calculation of the coordinate range of the spatial query window is based on the parameters of the urban spatial grid division reference system. For a given grid row index value R and column index value C, the coordinates of the lower left corner of the corresponding grid cell are: The coordinates of the upper right corner are To cover adjacent grid cells, the query window needs to expand outward by one grid layer. Therefore, the final coordinates of the lower left corner of the spatial query window are: During coordinate transformation, attention must be paid to handling boundary conditions. When the grid is located at the city edge, the query window should not exceed the city boundary coordinate range. The coordinate range of the transformed spatial query window is represented by four floating-point numbers (minLon, minLat, maxLon, maxLat), which serve as the conditions for spatial queries in the database.

[0044] Step S134: Using the coordinate range of the spatial query window as the spatial constraint and each emergency resource category code in the emergency resource category code set as the resource category filtering condition, a structured query instruction is jointly constructed and sent to the emergency resource deployment database.

[0045] Structured query commands are written in SQL and consist of two parts: spatial extent query and attribute filtering. Spatial extent query uses the PostGIS extended ST_Intersects function to determine if a resource location is within the query window; attribute filtering uses the IN clause to match resource category codes. The basic format of the query command is: The system executes the following query: `_resourcesWHEREST_Intersects(geometry, ST_MakeEnvelope(minLon, minLat, maxLon, maxLat, 4326)) AND resource_typeIN('type1', 'type2', ...) AND available_quantity>0.` To improve query efficiency, the system optimizes query conditions, including creating spatial indexes and using parameterized queries to avoid SQL injection. The completed query command is sent to the emergency resource deployment database via a JDBC connection, with a timeout of 5 seconds.

[0046] Step S135: Receive the preliminary emergency resource unit record set returned by the emergency resource deployment database, which matches both the coordinate range of the spatial query window and the emergency resource category code set. Each preliminary emergency resource unit record in the preliminary emergency resource unit record set contains the resource unit geographic coordinates, resource unit category code, resource unit current status identifier, and resource unit available quantity fields.

[0047] The query result returned by the emergency resource deployment database is a ResultSet object containing all emergency resource unit records that meet the conditions. The system iterates through this result set using an iterator, converting each record into a preliminary emergency resource unit record structure. The resource unit's geographical coordinates are stored as latitude and longitude (double-precision floating-point numbers) in the WGS84 coordinate system, the resource unit category is encoded as a string (e.g., "fire_extinguisher"), the resource unit's current status identifier is an enumeration type (0=idle, 1=busy, 2=under maintenance), and the number of available resource units is an integer. The preliminary emergency resource unit record set is stored using an ArrayList data structure. Each record also includes a unique resource unit identifier (UUID) and a last update timestamp field for subsequent data association and status tracking.

[0048] Step S136: Based on the preliminary emergency resource unit record set, perform filtering, sorting, weighting, descriptor encapsulation and tag attachment processing, and output an emergency resource unit descriptor set with spatial traceability tags.

[0049] The filtering and sorting process first filters out records with a non-zero status identifier (non-idle). Then, it calculates the spatial straight-line distance between the geographic coordinates of the resource unit and the geographic coordinates of the event occurrence for each record, using this distance as a distance-related weighting factor. The distance calculation uses the Haversine formula to convert latitude and longitude coordinates into Earth surface distances (in meters). Sorting is done in ascending order of the distance-related weighting factor, with closer resource units having higher priority. Weighting normalizes the distance-related weighting factor to the range of 0-1 (higher values ​​for closer units) and appends it to each record. Descriptor encapsulation converts the filtered and sorted records into a standard emergency resource unit descriptor format, including resource ID, type code, coordinates, available quantity, and weighting factor. Tag appending adds a spatial tracing tag to each descriptor, recording the grid identifier and hierarchical information to which the resource unit belongs. The final generated set of emergency resource unit descriptors is in JSON array format for easy network transmission and subsequent processing.

[0050] Step S1361: Filter out all preliminary emergency resource unit records whose current status identifier indicates a non-idle state from the preliminary emergency resource unit record set, and generate a candidate emergency resource unit record set that is in a schedulable state.

[0051] The filtering process iterates through each record in the initial emergency resource unit record set, checking the resource unit's current status identifier field. If the value of this field is 0 (idle), the record is retained; if it is 1 (busy) or 2 (under maintenance), it is removed from the set. To ensure data accuracy, the system also checks the resource unit availability field; if the availability is 0, it will be filtered even if the status is idle. After filtering, the generated candidate emergency resource unit record set typically has 30%-50% fewer records than the initial set (depending on the actual resource usage). This set is still stored using an ArrayList, retaining all fields from the original records.

[0052] Step S1362: Extract the geographical coordinates of each candidate emergency resource unit record in the candidate emergency resource unit record set, calculate the spatial straight-line distance between the geographical coordinates of the resource unit and the geographical coordinate code of the event occurrence, and add the spatial straight-line distance as a distance association weight factor to the candidate emergency resource unit record.

[0053] The geographic coordinates of resource units are WGS84 latitude and longitude (Lon_r, Lat_r), while the geographic coordinates of the event occurrence are encoded in Gauss-Kruger projection coordinates (X_e, Y_e). The former needs to be converted to Gauss-Kruger coordinates first. The conversion uses a coordinate projection transformation algorithm, utilizing the parameters of the Gauss-Kruger projection zone where the city is located (central meridian, projection coefficient, etc.). The spatial straight-line distance is calculated using the Euclidean distance formula: distance equals the square root of ((X_r - X_e)² plus (Y_r - Y_e)²), where (X_r, Y_r) are the Gauss-Kruger coordinates of the resource unit. The calculation result is in meters, accurate to the nearest integer. An additional distance-related weighting factor is added to the candidate emergency resource unit record, storing this distance value for subsequent sorting and route planning.

[0054] Step S1363: Sort all candidate emergency resource unit records in the candidate emergency resource unit record set in ascending order according to the distance association weight factor, and generate a resource unit priority list arranged according to spatial proximity.

[0055] The sorting uses the quicksort algorithm, with distance as the sorting key, and records are arranged in ascending order. For records with the same distance (extremely low probability), they are sorted in descending order of the number of available resource units, with priority given to resource units with more available units. After sorting, the first element of the resource unit priority list is the resource unit closest to the event location, and the last element is the resource unit furthest away. The system records the order of records before and after sorting for subsequent traceability and auditing.

[0056] Step S1364: Select candidate emergency resource unit records whose sorting position is within the preset proportion range of the preceding order from the resource unit priority list, and encapsulate the resource unit geographical coordinates, resource unit category code, resource unit current status identifier, resource unit available quantity field and distance association weight factor in the selected candidate emergency resource unit records into an emergency resource unit descriptor, forming a set of emergency resource unit descriptors that are directly related to the urban spatial grid identification sequence.

[0057] The preset ratio range is set through system configuration parameters, defaulting to the top 30%. The selection process calculates the selection quantity based on the total length N of the resource unit priority list, multiplying N by 30% and rounding up. For example, if the list contains 20 records, the first 6 are selected. The encapsulation process converts the selected records into an emergency resource unit descriptor structure, which includes: a unique resource unit identifier (UUID), a resource unit category code (string), resource unit geographic coordinates (Gauss-Kruger coordinates, double-precision array), a resource unit current status identifier (integer), the number of available resource units (integer), and a distance-related weighting factor (floating-point number). The emergency resource unit descriptor set is stored using a linked list structure, with each node containing a descriptor and a pointer to the next node.

[0058] Step S1365: Associate and store the set of emergency resource unit descriptors with the event occurrence time code, attach the spatial tracing tag of the urban spatial grid identification sequence to the set of emergency resource unit descriptors, and output the set of emergency resource unit descriptors with the spatial tracing tag.

[0059] The associated storage stores the set of emergency resource unit descriptors and the event occurrence time code (UNIX timestamp) in the same data object, facilitating subsequent querying and analysis. The spatial tracing tag contains the grid level to which the resource unit belongs, the grid identifier, and its index position in the city's spatial grid identification sequence. This information allows tracing the spatial relationship between the resource unit and the event's location. Tag attachment adds a metadata field to the header of the emergency resource unit descriptor set, storing spatial tracing information. The final output set of emergency resource unit descriptors with attached spatial tracing tags is in binary format, containing a timestamp (8 bytes), tag length (4 bytes), tag data (variable length), and a descriptor array (variable length).

[0060] Step S140: Activate the event evolution prediction module of the emergency response decision network, and perform temporal dependency modeling on the emergency resource unit descriptor set based on the emergency resource unit descriptor set and in conjunction with the event occurrence time code, to generate a probability distribution map of the propagation path of the emergency event in the urban spatial grid identification sequence.

[0061] The event evolution prediction module of the emergency response decision network adopts a hybrid architecture of graph neural networks (GNNs) and recurrent neural networks (RNNs), including a graph structure building unit, a spatiotemporal feature fusion unit, a temporal dependency modeling unit, and a probability distribution generation unit. The activation process first loads pre-trained model parameters (including weight matrices, bias vectors, etc.), then preprocesses the input set of emergency resource unit descriptors to extract features such as resource location, type, and weight. The event occurrence time encoding is converted into a temporal feature vector (including periodic features such as hour, weekday / weekend, and season), and fused with spatial features. The graph structure building unit constructs a topology graph based on the spatial relationship between resources and grids. The spatiotemporal feature fusion unit aggregates spatial features through graph convolution operations. The temporal dependency modeling unit uses a gated recurrent unit (GRU) to capture dynamic temporal changes. The probability distribution generation unit outputs the event arrival probability for each grid unit. The propagation path probability distribution map is a two-dimensional array with the same dimension as the number of grids in the urban spatial grid identifier sequence. Each element represents the event propagation probability (a floating-point number between 0 and 1) for the corresponding grid unit.

[0062] Step S141: Activate the event evolution prediction module of the emergency response decision network, and extract the resource unit geographic coordinates, resource unit category code and distance association weight factor contained in each emergency resource unit descriptor from the emergency resource unit descriptor set.

[0063] The activation process is completed by calling the initialization function of the event evolution prediction module. This function loads the model structure configuration file and pre-trained weight parameters, and sets the hyperparameters of each layer of the network (such as learning rate, dropout probability, etc.). The extraction process traverses the set of emergency resource unit descriptors, reading the resource unit's geographical coordinates (Gauss-Kruger coordinates X, Y), resource unit category code (string), and distance correlation weight factor (floating-point number) from each descriptor. The resource unit category code is converted into a binary vector (dimension equal to the total number of resource types, such as 20 dimensions) through one-hot encoding, while the distance correlation weight factor is directly used as a numerical feature. The extracted features are stored in a feature matrix, with each row corresponding to an emergency resource unit, and the number of columns is 2 (coordinates) + 20 (category code) + 1 (weight) = 23 columns.

[0064] Step S142: Map the geographical coordinates of each emergency resource unit descriptor to the urban spatial grid division reference system, calculate the resource-residence grid row index value and resource-residence grid column index value corresponding to the geographical coordinates of the resource unit in the urban spatial grid division reference system, and generate a resource-residence grid identifier unit.

[0065] The mapping process is similar to step S122, using the parameters of the urban spatial grid division reference system (origin coordinates Lon0, Lat0, grid size D). For the geographic coordinates (X, Y) of a resource unit, the resource-residence grid row index is equal to (Y-Lat0) divided by D and rounded down, and the resource-residence grid column index is equal to (X-Lon0) divided by D and rounded down. The calculation result is two integers (R_r, C_r), which together constitute the resource-residence grid identifier unit. To ensure mapping accuracy, the system verifies the calculation results, checking whether the resource-residence grid is within the urban grid range. If it is not within the range, the resource unit will be marked as invalid and excluded from further processing.

[0066] Step S143: Perform spatial overlay analysis on the resource-residence grid identifier unit and the enhanced semantic spatial grid identifier unit in the urban spatial grid identifier sequence to identify the spatial inclusion relationship or spatial adjacency relationship between the resource-residence grid identifier unit and the enhanced semantic spatial grid identifier unit, and establish spatial association edges between the emergency resource unit and the urban spatial grid identifier sequence based on the spatial inclusion relationship or spatial adjacency relationship.

[0067] Spatial overlay analysis employs a grid index matching method, comparing the row and column indices of resource-resident grid identifiers (R_r, C_r) with the row and column indices of each semantically enhanced spatial grid identifier in the urban spatial grid identifier sequence. If they match perfectly, it's a spatial containment relationship; if the absolute values ​​of both row and column index differences are less than or equal to 1, it's a spatial adjacency relationship. For each relationship, a spatial association edge is established, with attributes including relationship type (containment / adjacency) and distance association weight factor. Spatial association edges are stored using an edge list data structure, with each edge containing four attributes: source node (resource ID), target node (grid identifier), relationship type, and weight.

[0068] Step S144: Using each enhanced semantic spatial grid identifier unit in the urban spatial grid identifier sequence as a node, and the spatial association edge between the emergency resource unit and the urban spatial grid identifier sequence as the initial connection edge, construct an emergency event propagation spatial topology map. The node attributes of the emergency event propagation spatial topology map include urban functional area type labels and infrastructure density level descriptors, and the edge attributes of the emergency event propagation spatial topology map include distance association weight factors.

[0069] The spatial topology graph for emergency event propagation is represented using an adjacency list. Nodes in the graph are spatial grid identifiers with enhanced semantics. Each node contains a unique identifier (Z-order encoding), an attribute dictionary (functional area type labels and infrastructure density level descriptors), and a list of adjacent edges. The initial connection edges are the spatial association edges established in step S143. In addition, the system automatically adds spatial adjacency edges between grid cells (such as adjacent grids vertically, horizontally, and vertically), with the edge weight being 1 divided by the distance between grids. The urban functional area type labels in the node attributes are converted into numerical vectors through one-hot encoding, while the infrastructure density level descriptors are directly used as numerical features. These attributes will serve as input features for the graph neural network.

[0070] Step S145: Input the emergency event propagation spatial topology map into the graph structure temporal convolution unit of the event evolution prediction module, perform spatial domain graph convolution operation on the emergency event propagation spatial topology map, aggregate the urban functional area type label information and infrastructure density level descriptor information of the neighboring areas of each node, and generate node spatial neighborhood aggregated feature vector.

[0071] The graph-structured temporal convolutional unit comprises two graph convolutional layers, each consisting of a linear transformation, neighborhood aggregation, and an activation function. The first graph convolutional layer takes node attribute features (functional area type vector + density level) as input, performs a linear transformation using a trainable weight matrix, and then multiplies it with the adjacency matrix to achieve neighborhood feature aggregation. The activation function used is ReLU. The second graph convolutional layer takes the output features of the first layer as input and performs a similar operation, but with a different weight matrix size. The specific process of the spatial domain graph convolution operation is as follows: for each node, its own features are summed with the features of all neighboring nodes according to the edge weights to obtain the aggregated feature vector. The node spatial neighborhood aggregated feature vector has a dimension of 128, containing the comprehensive spatial features of the node itself and its neighborhood.

[0072] Step S146: Based on the node spatial neighborhood aggregated feature vector, perform spatiotemporal feature fusion, temporal dependency modeling, probability weight allocation and distribution map generation processing to generate a probability distribution map of the propagation path of emergency events in the urban spatial grid identification sequence.

[0073] Spatiotemporal feature fusion concatenates the node spatial neighborhood aggregated feature vector (128-dimensional) with the event occurrence time encoded temporal feature vector (24-dimensional, including periodic features such as hours, days of the week, and months) along the feature dimensions to form a 152-dimensional spatiotemporal joint feature vector. Temporal dependency modeling employs a two-layer stacked gated recurrent unit (GRU). The first GRU takes the spatiotemporal joint feature vector as input and outputs a 64-dimensional hidden state; the second GRU takes the hidden state from the first layer as input and outputs a 32-dimensional hidden state feature sequence. Probability weight allocation maps the hidden state feature sequence to the grid number dimension through a fully connected layer, and then normalizes it using a Softmax function to obtain the event propagation probability weight for each grid cell. Distribution map generation maps the probability weights to the grid positions of the urban spatial grid identification sequence, forming a two-dimensional propagation path probability distribution map, where the probability value of each grid cell is between 0 and 1.

[0074] For example, step S1461: the node spatial neighborhood aggregation feature vector and the event occurrence time code are spliced ​​together on the time axis to form a spatiotemporal joint feature tensor with spatial feature dimension and temporal feature dimension.

[0075] The node spatial neighborhood aggregation feature vector is a 128-dimensional vector. The event occurrence time encoding is first converted into a time feature vector. The construction process of the time feature vector is as follows: the UNIX timestamp is converted into fields such as year, month, day, hour, minute, and second. Then, the year, month, and day are one-hot encoded (10-dimensional, 12-dimensional, and 31-dimensional respectively), and the hour, minute, and second are sine and cosine encoded (2-dimensional for each field). Finally, they are concatenated to obtain a 2+10+12+31+2+2+2=61-dimensional time feature vector. Time axis concatenation involves concatenating the node spatial neighborhood aggregation feature vector (128-dimensional) and the time feature vector (61-dimensional) along the feature dimensions (column direction) to form a 128+61=189-dimensional spatiotemporal joint feature vector. If the city spatial grid identifier sequence contains M grid cells, the shape of the spatiotemporal joint feature tensor is M×189.

[0076] Step S1462: Input the spatiotemporal joint feature tensor into the temporal dependency modeling subnetwork of the event evolution prediction module. The temporal dependency modeling subnetwork contains stacked gated recurrent unit layers. The spatiotemporal joint feature tensor is forward propagated recursively calculated along the time dimension through the gated recurrent unit layers to capture the spatial propagation dependency pattern of the emergency event at continuous time nodes and output the hidden state feature sequence.

[0077] The temporal dependency modeling subnetwork comprises two layers of Gated Recurrent Units (GRUs), each with a hidden layer dimension of 64. The first GRU layer takes a spatiotemporal joint feature tensor (M×189) as input, which, after computation via reset and update gates, outputs a 64-dimensional hidden state. The reset gate determines how new inputs are combined with previous memories, while the update gate determines how much of the previous memory is retained. The second GRU layer takes the hidden state from the first layer as input, also undergoes gating, and outputs a 64-dimensional sequence of hidden state features. Forward propagation recursively computes the feature vector of one grid cell at each time step, capturing temporal dependencies between grid cells through the internal state propagation of the GRU. The hidden state feature sequence has a shape of M×64, where M is the number of grid cells.

[0078] Step S1463: Input the hidden state feature sequence into the probability distribution generation layer of the event evolution prediction module, perform fully connected mapping processing on each hidden state feature vector in the hidden state feature sequence, and convert the mapped feature vector into the set of predicted grid row index values ​​and the set of predicted grid column index values ​​in the urban spatial grid division reference system through the spatial coordinate inverse calculation function.

[0079] The probability distribution generation layer consists of a fully connected layer and a spatial coordinate inverse calculation unit. The fully connected layer takes a sequence of hidden state features (M×64) as input and outputs a 2×M dimension (predicted row and column indices for each grid cell). The weight matrix of the fully connected layer is 64×2M in size, and the bias vector is 2M in size. The mapped feature vector is a 2×M matrix, where the two elements in the i-th column are the floating-point values ​​of the predicted row and column indices for the i-th grid cell, respectively. The spatial coordinate inverse calculation function rounds these floating-point values ​​to integers, obtaining a set of predicted grid row indices and a set of predicted grid column indices, each set containing M integers.

[0080] Step S1464: Assign an event propagation probability weight to each grid cell corresponding to the predicted grid row index value and the predicted grid column index value set. The event propagation probability weight is proportional to the activation response intensity of the hidden state feature vector in the fully connected mapping process.

[0081] The allocation of event propagation probability weights is calculated using the Softmax function, which transforms the unnormalized score output by the fully connected layer into a probability distribution. Specifically, for each grid cell i, its score value s_i output by the fully connected layer is calculated. Then, the ratio of this score value to the sum of the scores of all grid cells is calculated, yielding the event propagation probability weight p_i = exp(s_i) / sum(exp(s_j))forallj. The activation response strength is equivalent to the score value s_i; therefore, p_i is directly proportional to s_i. After the probability weights are allocated, each grid cell has a probability value between 0 and 1, representing the likelihood of an emergency event propagating to that grid cell.

[0082] Step S1465: The predicted grid row index value set, the predicted grid column index value set, and the corresponding event propagation probability weights are jointly rendered into a continuous probability distribution field on the urban spatial grid division reference system, generating a propagation path probability distribution map of the emergency event in the urban spatial grid identification sequence. The propagation path probability distribution map records the event arrival probability value with grid cells as the basic unit.

[0083] The rendering process transforms discrete grid cell probability values ​​into a continuous probability distribution field, employing a bilinear interpolation algorithm for smooth transitions between grid points. The specific steps are as follows: a two-dimensional array with the same size as the urban spatial grid division reference frame (resolution equal to the total number of urban grid rows × total number of grid columns) is created; the probability weight of each predicted grid cell is filled into the corresponding position in the array; and then interpolation calculations are performed on the positions in the array not covered by predictions. The data structure of the propagation path probability distribution map is a two-dimensional floating-point array, where the rows and columns correspond to the grid row and column indices, and the array element values ​​are the event arrival probability values ​​(0-1). In addition, the map also contains metadata such as generation timestamps and confidence indices.

[0084] Step S150: Generate a smart city emergency response dispatch instruction set based on the propagation path probability distribution map. The smart city emergency response dispatch instruction set includes a resource dispatch time window and a resource dispatch spatial path point list for each emergency resource unit in the emergency resource unit descriptor set.

[0085] The generation process is completed collaboratively by the path planning unit, the time window calculation unit, and the instruction encoding unit. Path planning unit The resource unit plans the optimal path from its current location to the target grid unit, considering grid traffic conditions and event propagation probability. The time window calculation unit calculates the time interval for the resource to reach the target grid based on path length and resource movement speed. The instruction encoding unit encapsulates the path points and time window into a standard scheduling instruction format, including information such as resource ID, target grid, path sequence, start time, and end time. The smart city emergency response scheduling instruction set is a JSON array, with each element corresponding to a scheduling instruction, which can be directly sent to the emergency resource scheduling execution terminal.

[0086] Step S151: Analyze the propagation path probability distribution map, extract all grid cells whose event propagation probability weight exceeds a preset activation threshold from the propagation path probability distribution map, and mark the grid cells as the target grid cell set.

[0087] The preset activation threshold is set via system configuration parameters, with a default value of 0.6. The parsing process iterates through the two-dimensional array of the propagation path probability distribution map, checking the probability weight value of each grid cell. If the value is greater than or equal to 0.6, the row and column indices of that grid cell are added to the target grid cell set. The target grid cell set is stored using a hash set to avoid duplicate elements. Simultaneously, the system records the probability weight value of each target grid cell for subsequent priority sorting.

[0088] Step S152: Sort each target grid cell in the target grid cell set in descending order according to the event propagation probability weight, and generate a priority sequence of threatened grid cells arranged according to the urgency of the event impact.

[0089] The sorting algorithm uses heap sort, with the event propagation probability weight as the key, arranged from largest to smallest. After sorting, the first element of the threatened grid priority sequence is the target grid cell with the highest probability weight (most urgent), and the last element is the target grid cell with the lowest probability weight. If two grid cells have the same probability weight, they are sorted in ascending order according to their distance from the event location, with closer grid cells given priority.

[0090] Step S153: Read the resource unit geographic coordinates, resource unit category code, available quantity of resource units, and distance association weight factor contained in each emergency resource unit descriptor from the set of emergency resource unit descriptors in sequence.

[0091] The reading process follows the storage order of the emergency resource unit descriptor set, accessing each descriptor node sequentially. For each descriptor, the resource unit's geographic coordinates (Gauss-Kruger coordinates X, Y), resource unit category code (e.g., "ambulance"), available resource unit quantity (integer), and distance-related weighting factor (floating-point number) are extracted. This information will be used for resource scheduling priority assessment and path planning.

[0092] Step S154: For the currently processed emergency resource unit descriptor, match its resource unit category code with the preset emergency resource scheduling priority mapping table to obtain the resource scheduling priority level corresponding to the emergency resource unit descriptor.

[0093] The emergency resource scheduling priority mapping table is an XML configuration file that defines the scheduling priorities for different resource types. For example, "ambulance" has a priority of 1 (highest), "fire_extinguisher" has a priority of 2, and "police_car" has a priority of 3, etc. The matching process is implemented through string lookup, finding the corresponding priority level (integer 1-5, with 1 being the highest) in the mapping table based on the resource unit category code. The resource scheduling priority level will affect the order in which scheduling instructions are generated; resources with higher priority are path planned first.

[0094] Step S155: Map the geographical coordinates of the resource unit of the current emergency resource unit descriptor to the urban spatial grid division reference system, generate the resource-residence grid row index value and the resource-residence grid column index value corresponding to the emergency resource unit descriptor, and form a resource-residence grid identifier.

[0095] The mapping process is the same as step S142, using the parameters (origin coordinates, grid size) of the urban spatial grid division reference system to convert the geographic coordinates of resource units into grid row and column indices. The resource-residence grid is identified as an integer pair (R_r, C_r), used for starting point location in path planning.

[0096] Step S156: Based on the resource-resident grid identifier, perform path planning, weight evaluation, scheduling parameter calculation and instruction generation processing, and output the smart city emergency response scheduling instruction set.

[0097] Starting from the resource-residence grid identifier and ending at the target grid cell in the priority sequence of threatened grids, the optimal path is searched. Weight evaluation comprehensively considers path length, traffic conditions, and event urgency to calculate a comprehensive score for the path. Dispatch parameters are calculated based on path length and resource movement speed (different resource types have different speeds, e.g., ambulances 80 km / h, fire trucks 60 km / h) to determine the arrival time window. Instruction generation encapsulates the path point sequence, time window, and resource information into dispatch instructions, which are then added to the smart city emergency response dispatch instruction set.

[0098] For example, step S1561: Perform spatial path planning calculation on the resource-residence grid identifier and each target grid cell in the threatened grid priority sequence, and use a breadth-first search algorithm based on grid adjacency to generate the shortest grid path sequence from the resource-residence grid identifier to each target grid cell.

[0099] The breadth-first search algorithm starts with the resource-residing grid identifier and expands layer by layer to adjacent grid cells until the target grid cell is reached. During the search, each grid cell records its parent node for backtracking. The shortest grid path sequence is an array of grid identifiers containing the row and column indices of all grid cells traversed from the starting point to the destination. For example, the path sequence from (R1, C1) to (R5, C5) could be [(R1, C1), (R1, C2), (R2, C2), (R3, C2), (R4, C2), (R5, C2), (R5, C3), (R5, C4), (R5, C5)].

[0100] Step S1562: Calculate the path evaluation weight for each shortest grid path sequence. The calculation process of the path evaluation weight includes: standardizing the number of grid cells traversed by the shortest grid path sequence to obtain a path length score; standardizing the event propagation probability weight of the set target grid cell to obtain a target urgency score; and then weighting the path length score and the target urgency score according to a preset weight coefficient to obtain the path evaluation weight.

[0101] The path length score is calculated as follows: the path length score equals 1 minus (the number of path grids divided by the maximum possible number of path grids), standardized to the range of 0-1. The target urgency score directly uses the event propagation probability weight of the target grid cell (already within the range of 0-1). The preset weight coefficients are 0.4 for the path length score and 0.6 for the target urgency score. The path evaluation weight equals the path length score multiplied by 0.4 plus the target urgency score multiplied by 0.6, with the result between 0 and 1. A larger value indicates a better path.

[0102] Step S1563: Based on the path evaluation weight, perform a comprehensive evaluation on all shortest grid path sequences, and select the target shortest grid path sequence with the optimal path evaluation weight as the resource scheduling space path point sequence of the emergency resource unit descriptor. The resource scheduling space path point sequence consists of a sequence of continuous grid row index values ​​and a sequence of continuous grid column index values ​​traversed from the resource-residence grid identifier to the set target grid unit.

[0103] A comprehensive evaluation of the path evaluation weights corresponding to all target grid cells is conducted, and the path with the highest weight is selected as the optimal path. The resource scheduling space path point sequence stores the grid row index and column index of the optimal path as two separate arrays. For example, the row index sequence is [R1, R1, R2, R3, R4, R5, R5, R5, R5], and the column index sequence is [C1, C2, C2, C2, C2, C2, C3, C4, C5]. This representation facilitates subsequent coordinate transformations and navigation.

[0104] Step S1564: Based on the total grid step length of the resource scheduling space path point column and the preset average travel time parameter of the grid cell, calculate the travel time span required for the emergency resource unit descriptor to move from the resource residence grid identifier to the set target grid cell.

[0105] The total grid step size equals the number of grid cells in the path point sequence minus 1 (steps = number of points - 1). The preset average travel time parameter per grid cell is determined based on the resource type and grid type. For example, the average travel time for an ambulance in a regular road grid is A minutes / grid, and in a highway grid it is B minutes / grid (A > B). The travel time span equals the total grid step size multiplied by the average travel time parameter, in minutes. If the path passes through different types of grids, the time for each segment is calculated separately and then summed.

[0106] Step S1565: Perform timeline superposition operation on the trip time span and the event occurrence time code, take the time indicated by the event occurrence time code as the starting time and the trip time span as the duration, and generate the resource scheduling time window of the emergency resource unit descriptor. The resource scheduling time window includes the absolute time of scheduling start and the absolute time of scheduling end.

[0107] The event occurrence time is encoded as a UNIX timestamp (in seconds). The absolute start time of scheduling equals the event occurrence time encoding plus the resource preparation time (default is 5 minutes, or 300 seconds). The absolute end time of scheduling equals the absolute start time of scheduling plus the travel time span (converted to seconds). The time window is represented by two UNIX timestamps, for example, the start time is T1 and the end time is T2. The resource must depart at time T1 and arrive at the target grid before time T2.

[0108] Step S1566: Jointly encode the resource scheduling priority level, the resource scheduling time window, and the resource scheduling spatial path point list to generate a single emergency resource scheduling instruction for the current emergency resource unit descriptor, and add the single emergency resource scheduling instruction to the smart city emergency response scheduling instruction set.

[0109] The joint encoding uses JSON format. A single emergency resource dispatch instruction contains the following fields: resource_id (unique identifier for the resource unit), priority (resource dispatch priority level), start_time (absolute start time of dispatch), end_time (absolute end time of dispatch), path_row (array of row index sequences), and path_col (array of column index sequences). The generated instructions are added to an array of smart city emergency response dispatch instructions, which is sorted in descending order of resource dispatch priority.

[0110] Step S1567: Repeat the above steps until all emergency resource unit descriptors in the set of emergency resource unit descriptors have been processed, and output the smart city emergency response dispatch instruction set. Each emergency resource dispatch instruction in the smart city emergency response dispatch instruction set contains a unique identifier for the resource unit, a resource dispatch time window, and a list of resource dispatch spatial path points.

[0111] The system iteratively processes each descriptor in the set of emergency resource unit descriptors, generating a scheduling instruction for each descriptor. After processing, the smart city emergency response scheduling instruction set contains the same number of scheduling instructions as the set of emergency resource unit descriptors. The system performs a final verification of the instruction set to ensure that the time windows of all instructions do not overlap (for the same resource) and that the path point columns are valid, before outputting the instruction set.

[0112] Step S210: Collect a set of historical emergency event records for the city. Each historical emergency event record in the set includes the geographic coordinate code of the historical event, the time code of the historical event, and the grid identifier sequence of the actual impact range of the historical event.

[0113] The city's historical emergency event record collection is sourced from the city's emergency management department's historical database, containing records of various emergency events (fires, earthquakes, floods, etc.) that occurred within the past five years. Each record's geographic coordinates are encoded using Gauss-Kruger projection coordinates, and the event time is encoded using a UNIX timestamp. The actual impact area grid identifier sequence is the set of affected grid cells determined through on-site investigation and remote sensing image analysis after the event, stored as an array of grid row and column indices. Data acquisition is conducted through a database query interface, employing an incremental acquisition strategy, collecting only newly added event records each time to avoid duplication. The collected records are stored in a Hadoop distributed file system, managed by event type and occurrence time partitions.

[0114] Step S220: Perform spatial gridding processing on each historical emergency event record in the city's historical emergency event record set, convert the geographic coordinates of the historical event occurrence into historical event grid row index values ​​and historical event grid column index values, and parse the grid identifier sequence of the actual impact range of the historical event into a set of historical impact grid row index values ​​and a set of historical impact grid column index values.

[0115] The spatial gridding process is the same as the grid index conversion method in step S122, using the same urban spatial grid division reference system parameters. For the geographic coordinate encoding of historical events (X_h, Y_h), the historical event grid row index value is equal to (Y_h-Lat0) divided by D and rounded down; the column index value is similar. The grid identifier sequence of the actual impact range of historical events is a JSON-formatted string. After parsing, it yields a set of historical impact grid row index values ​​and a set of historical impact grid column index values. Each set is an integer array of equal length. For example, the row index set is [R1, R2, R3, ...], the column index set is [C1, C2, C3, ...], and the corresponding grid cells are (R1, C1), (R2, C2), etc.

[0116] Step S230: Using the historical event grid row index value and the historical event grid column index value as the center, and the historical influence grid row index value set and the historical influence grid column index value set as the spatial expansion boundary, construct a historical event spatial propagation trajectory map. The historical event spatial propagation trajectory map uses grid cells as nodes and the event propagation direction as directed edges.

[0117] The construction of the historical event spatial propagation trajectory map first determines the central grid (historical event grid row index value, historical event grid column index value), and then adds grid cells from the historically affected grid identifier sequence as nodes to the graph. The direction of the directed edges indicates the direction of event propagation, from the first affected grid to the next affected grid. The weight of the edge is the propagation time difference (the time the next grid is affected minus the time the previous grid is affected). If there is no propagation time information in the historical data, it is assumed that the propagation is isotropic, and the edge weight is 1. The trajectory map is represented by an adjacency matrix, where the matrix element (i, j) represents the edge weight from node i to node j.

[0118] Step S240: Input the historical event spatial propagation trajectory map into the initialized graph convolutional neural network. The graph convolutional neural network includes an input layer, multiple stacked graph convolutional hidden layers, and an output layer. The input layer receives the spatial adjacency matrix and the initial feature vector of each grid node. The initial feature vector of the node is formed by concatenating the urban functional area type label encoding and the infrastructure density level descriptor encoding corresponding to the grid cell.

[0119] The initialized graph convolutional neural network consists of an input layer (dimension equal to the length of the node feature vector), two graph convolutional hidden layers (64-dimensional and 32-dimensional respectively), and an output layer (dimension equal to the number of grid cells). The spatial adjacency matrix is ​​the adjacency matrix of the historical event spatial propagation trajectory graph. The construction process of the initial node feature vector is as follows: the urban functional area type label is converted into a 10-dimensional vector through one-hot encoding, and the infrastructure density level descriptor is converted into a 1-dimensional numerical value (1-5). After concatenation, an 11-dimensional initial node feature vector is formed. The input layer passes the adjacency matrix and feature vector to the first graph convolutional hidden layer to begin the forward propagation calculation.

[0120] Step S250: The graph convolutional hidden layer performs Laplacian normalization on the input spatial adjacency matrix, performs linear transformation on the node feature vector through a trainable weight matrix, and aggregates the feature vectors of neighboring nodes to generate an updated node feature vector.

[0121] The Laplacian normalization process involves: calculating the degree matrix of the adjacency matrix (a diagonal matrix where the diagonal elements are the degrees of the nodes), and then calculating the normalized Laplacian matrix. D^(-1 / 2), where I is the identity matrix, D is the degree matrix, and A is the adjacency matrix. The linear transformation is achieved through a trainable weight matrix W (size equal to the input feature dimension × output feature dimension), and the transformed... Where X is the input feature matrix. Neighbor node features In practice, the final updated node feature vector is activated by the ReLU function. The result.

[0122] Step S260: Based on the updated node feature vectors, perform prediction mapping, accuracy calculation, weight optimization, network solidification and embedding processing to construct a spatial feature coding backbone network and complete deployment.

[0123] Prediction mapping is achieved through a fully connected layer in the output layer, mapping the output feature vector of the last graph convolutional hidden layer to the grid number dimension. The predicted probability for each grid is then obtained through a Softmax function. Accuracy is calculated by comparing the overlap between the predicted influence grid set and the historical influence grid set, using the F1 score as the evaluation metric. Weight optimization employs the Adam optimizer with a learning rate of 0.001, a batch size of 32, and 200 training epochs, updating all trainable weight matrices through backpropagation. Network solidification involves saving the trained weight parameters and network structure as a model file. Embedding processing integrates this model as the spatial feature encoding backbone network into the emergency response decision network for real-time spatial feature extraction.

[0124] For example, in step S261: the output layer performs a fully connected mapping on the node feature vectors output by the last graph convolutional hidden layer, and converts the mapped feature vectors into a set of predicted influence grid row index values ​​and a set of predicted influence grid column index values ​​through a spatial coordinate inverse calculation function.

[0125] The node feature vector output by the convolutional hidden layer in the last graph has a dimension of 32, the weight matrix of the fully connected layer has a size of 32×M (M is the total number of grid cells in the city), and the bias vector has a size of M. The mapped feature vector is M-dimensional, with each element representing the predicted score of the corresponding grid cell. The spatial coordinate inverse calculation function uses the indices of the top K grid cells with the highest scores (K equals the number of historically influential grid cells) as the set of row index values ​​and column index values ​​for the predicted influential grid cells.

[0126] Step S262: Calculate the number of spatially overlapping grids between the predicted influence grid row index value set and the predicted influence grid column index value set and the historical influence grid row index value set and the historical influence grid column index value set, and use the ratio of the number of spatially overlapping grids to the total number of historical influence grids as the spatial prediction accuracy evaluation index.

[0127] The number of grid cells in the spatially overlapping region is calculated using set intersection. The predicted impact grid set and the historical impact grid set are converted into hash sets, and the size of the intersection is calculated. The spatial prediction accuracy is equal to the intersection size divided by the total number of historical impact grid cells, with a value between 0 and 1. The closer to 1, the more accurate the prediction.

[0128] Step S263: Using the maximization of the spatial prediction accuracy evaluation index as the objective function, the backpropagation algorithm is used to iteratively optimize the trainable weight matrix in the graph convolutional neural network until the spatial prediction accuracy evaluation index converges to a preset stable range.

[0129] The backpropagation algorithm calculates the gradient of the prediction accuracy with respect to the weight matrix of each layer, and then uses the Adam optimizer to update the weights. During the iterative optimization process, the spatial prediction accuracy of the validation set is calculated every 10 training epochs. If the change in the validation accuracy is less than 0.001 for 5 consecutive epochs, it is considered to have converged to the preset stable range, and training is stopped.

[0130] Step S264: Save the trainable weight matrix of the converged graph convolutional neural network, and solidify the structural parameters and weight parameters of the graph convolutional neural network to form a spatial feature encoding backbone network.

[0131] The converged weight matrix and structural parameters (dimensions of each layer, activation function types, etc.) are saved as a model file in HDF5 format, containing configuration information for the input, hidden, and output layers. The solidification process converts the model file to an inference optimization format (such as ONNX format), removing training-related parameters and improving inference speed.

[0132] Step S265: Embed the spatial feature encoding backbone network into the emergency response decision network as a feature extraction pre-unit of the spatial positioning branch module and the resource association analysis module, and use it to perform spatial feature encoding processing on the geographic coordinate encoding of the event occurrence and the urban spatial grid identification sequence.

[0133] The embedding process is implemented through a model integration interface, connecting the input and output ports of the spatial feature coding backbone network with the corresponding modules of the emergency response decision network. In the spatial positioning branch module, the urban spatial grid identifier sequence is first input into the spatial feature coding backbone network to extract a 128-dimensional spatial feature vector; in the resource association analysis module, the geographic coordinates of resource units are also extracted through this network for resource matching and priority ranking.

[0134] For example, spatial feature encoding processing is performed on the geographic coordinate encoding of the event occurrence and the urban spatial grid identifier sequence, including: step S310: receiving the geographic coordinate encoding of the event occurrence in the urban emergency event trigger signal, and converting the geographic coordinate encoding of the event occurrence into the grid row index value and the grid column index value to be encoded.

[0135] The conversion process is the same as step S122. The reference system parameters of the urban spatial grid are used to convert the geographic coordinate encoding of the event occurrence (X_e, Y_e) into the row index value R_e and column index value C_e of the grid to be encoded, which are used for subsequent core event grid node marking.

[0136] Step S320: Receive the urban spatial grid identifier sequence with time validity mark output by the spatial positioning branch module, and extract the grid row index value, grid column index value, urban functional area type label and infrastructure density level descriptor contained in each spatial grid identifier unit with enhanced semantics from the urban spatial grid identifier sequence.

[0137] The extraction process parses the urban spatial grid identifier sequence with time validity markers. First, it verifies the time validity, then decodes the grid identifier array, label difference array, and level difference array to recover the complete information of each spatial grid identifier unit with enhanced semantics, and stores it in a list.

[0138] Step S330: Mark the grid cells located by the grid row index value and the grid column index value to be encoded as core event grid nodes, and mark the grid cells corresponding to each spatial grid identifier cell with enhanced semantics in the urban spatial grid identifier sequence as associated spatial grid nodes.

[0139] The core event grid node is the grid cell where the event occurred, identified by (R_e, C_e). Associated spatial grid nodes include all grid cells in the city's spatial grid identifier sequence; these grid cells have a hierarchical containment or adjacency relationship with the core event grid node. The labeling process adds a type label (core / associated) to each grid cell.

[0140] Step S340: Using the core event grid node and all associated spatial grid nodes as vertices and the grid adjacency relationships in the urban spatial grid partitioning reference system as edges, construct the spatial topology subgraph to be encoded.

[0141] The vertex set of the spatial topological subgraph to be encoded is the union of the core event grid nodes and the associated spatial grid nodes, and the edge set is the spatial adjacency relationship between vertices (up, down, left, right, and diagonally adjacent). The weight of each edge is 1 divided by the Euclidean distance between vertices. The topological subgraph is represented using an adjacency list, which is convenient for graph convolutional neural networks to process.

[0142] Step S350: Generate an initial feature encoding vector for each vertex in the spatial topology subgraph to be encoded. The initial feature encoding vector is composed of the one-hot encoding vector of the urban functional area type label corresponding to the vertex, the normalized value of the infrastructure density level descriptor, and the degree centrality index of the vertex in the spatial topology subgraph to be encoded.

[0143] The one-hot encoding vector for the urban functional area type label has a dimension of 10 (assuming there are 10 types of urban functional areas). The infrastructure density level descriptor is normalized to the range of 0-1 (original value 1-5 divided by 5). The degree centrality index is the degree of the vertex divided by (number of vertices - 1). The initial feature encoding vector after concatenating these three elements has a dimension of 10+1+1=12.

[0144] Step S360: Based on the topological subgraph of the space to be encoded and the initial feature encoding vector, perform graph convolution operation, feature compression, spatiotemporal fusion and feature sequence output processing to complete spatial feature encoding.

[0145] Graph convolution operations use a spatial feature encoding backbone network. The input is the adjacency matrix of the spatial topology subgraph to be encoded and an initial feature encoding vector. The output is a 128-dimensional node spatial feature vector. Feature compression reduces the 128-dimensional vector to 64 dimensions through a fully connected layer. Spatiotemporal fusion concatenates the compressed features of the core event grid nodes with the temporal feature vector encoded by the event occurrence time, forming a 76-dimensional joint feature. The feature sequence output arranges the compressed features of all associated spatial grid nodes in hierarchical order, forming a feature sequence that serves as the spatial feature encoding for the city's spatial grid identification sequence.

[0146] For example, step S361: input the adjacency matrix of the spatial topology subgraph to be encoded and the initial feature encoding vectors of all vertices into the spatial feature encoding backbone network.

[0147] The adjacency matrix of the spatial topological subgraph to be encoded is an N×N sparse matrix (N is the number of vertices), and the initial feature encoding vector is an N×12 matrix. The spatial feature encoding backbone network loads pre-trained weight parameters and performs forward propagation calculations on the input data.

[0148] Step S362: The spatial feature encoding backbone network performs Laplacian normalization on the adjacency matrix and performs graph convolution on the initial feature encoding vector through the converged trainable weight matrix to aggregate the feature information of each vertex's neighboring vertices and generate an updated vertex spatial feature vector.

[0149] The Laplacian normalization operation is the same as in step S250. The graph convolution operation uses the pre-trained weight matrix, so no further training is required. The aggregated vertex spatial feature vector has a dimension of 128, containing the spatial features of the vertex and its neighborhood.

[0150] Step S363: Input the updated vertex spatial feature vector into the fully connected mapping layer of the spatial feature encoding backbone network, compress the dimension of the vertex spatial feature vector to a preset encoding length, and output the compressed spatial feature encoding of the core event grid node and the compressed spatial feature encoding of each associated spatial grid node.

[0151] The weight matrix of the fully connected mapping layer has a size of 128×64, compressing the 128-dimensional vertex space feature vector to 64 dimensions. The compressed spatial feature encoding of the core event grid node is its corresponding 64-dimensional vector, and the compressed spatial feature encoding of the associated spatial grid node is similar.

[0152] Step S364: Concatenate the compressed spatial feature code of the core event grid node with the event occurrence time code to generate a joint feature representation that integrates spatiotemporal information.

[0153] The time feature vector encoded by the event occurrence time is 61-dimensional (as described in step S1461). After being concatenated with the 64-dimensional compressed spatial feature encoding of the core event grid node, a joint feature representation of 64+61=125-dimensional spatiotemporal information is formed, which is used as the input of the event evolution prediction module.

[0154] Step S365: Compressed spatial feature codes of all associated spatial grid nodes are arranged into a feature sequence according to the order in the urban spatial grid identifier sequence, and the feature sequence is output as the spatial feature code of the urban spatial grid identifier sequence.

[0155] The compressed spatial feature codes of the associated spatial grid nodes are arranged in hierarchical order in the urban spatial grid identification sequence to form an N×64 feature matrix (N is the number of associated nodes), which serves as the spatial feature code of the urban spatial grid identification sequence and is used for feature matching in the resource association analysis module.

[0156] Step S410: Receive the smart city emergency response dispatch instruction set, and extract the resource dispatch spatial path point list contained in each emergency resource dispatch instruction from the smart city emergency response dispatch instruction set. The resource dispatch spatial path point list is composed of a continuous grid row index value sequence and a continuous grid column index value sequence.

[0157] The input for this step is the smart city emergency response dispatch instruction set generated in step S150, which is in JSON array format. The extraction process is achieved by parsing the "path_row" and "path_col" fields of each instruction in the JSON array, obtaining a sequence of continuous grid row index values ​​(integer array) and a sequence of continuous grid column index values ​​(integer array), respectively. For example, if the path point column of an instruction contains 10 grid cells, then both the row index sequence and the column index sequence contain 10 integer elements, and both have the same length. The extraction results are stored as a two-dimensional list structure, with the outer list corresponding to different dispatch instructions and the inner list containing the row index sequence and column index sequence.

[0158] Step S420: Convert the grid cell corresponding to each grid row index value and grid column index value in the resource scheduling spatial path point column into a spatial path node, and convert the directed connection edge between adjacent spatial path nodes into a path movement segment.

[0159] The transformation of spatial path nodes is based on the urban spatial grid division reference system, converting the grid row index value R and column index value C into Gauss-Kruger coordinates. Where Lon0 and Lat0 are the coordinates of the grid origin, and D is the grid size. Each spatial path node contains two attributes: coordinates (X, Y) and grid index (R, C). A path movement segment is defined as a directed line segment from the i-th spatial path node to the (i+1)-th spatial path node, containing three attributes: start node ID, end node ID, and segment length (calculated by the Euclidean distance between the two points). The transformed set of path movement segments is stored using an edge list data structure.

[0160] Step S430: Obtain the traffic congestion status identifier of the grid cell covered by the resource scheduling spatial path point column at the current moment from the city real-time traffic flow monitoring platform. The traffic congestion status identifier includes three status categories: smooth flow, slow flow, and blocked flow.

[0161] The city's real-time traffic flow monitoring platform provides traffic data query services via a RESTful API. Request parameters include a set of grid row index values ​​and a set of column index values. The platform returns a JSON response containing a traffic congestion status identifier for each grid cell: 0 indicates smooth traffic (speed > 60 km / h), 1 indicates slow traffic (30 km / h ≤ speed ≤ 60 km / h), and 2 indicates congestion (speed < 30 km / h). Data acquisition uses a batch query method, requesting a maximum of 100 grid cells' traffic status at a time, with a timeout of 3 seconds. The results are stored in a dictionary structure, with the grid identifier (R, C) as the key and the status identifier as the value.

[0162] Step S440: Traverse each path movement segment in the resource scheduling spatial path point list, identify the grid row index value and grid column index value of the two spatial path nodes connected by the path movement segment, and query the traffic congestion status identifier of the corresponding grid cell based on the grid row index value and grid column index value.

[0163] The traversal process proceeds in the order of the path movement segments. For each path movement segment, the grid indices (R1, C1) and (R2, C2) of its starting and ending nodes are extracted. Since the path movement segments are connections between adjacent grids, (R2, C2) can only be (R1±1, C1), (R1, C1±1), or (R1±1, C1±1). During the query, it first checks whether (R1, C1) and (R2, C2) exist in the traffic status data dictionary. If they exist, the status identifier is read; otherwise, it is marked as unobstructed by default. The query results are appended to the attributes of the path movement segment.

[0164] Step S450: If the traffic congestion status identifier of any grid cell covered by the path movement segment is blocked, then mark the path movement segment as an impassable path segment, and record the starting spatial path node and ending spatial path node of the path movement segment as a breakpoint path node pair.

[0165] The judgment process checks the traffic congestion status identifiers corresponding to the start and end nodes of the path segment. If either one is 2 (blocked state), the path segment is marked as impassable. Breakpoint path node pairs are stored as tuple structures (start node ID, end node ID), and the position index of the node pair in the path point column is recorded for subsequent path replacement. The marking result of impassable path segments is stored in the "status" attribute of the path segment (0 = passable, 1 = impassable).

[0166] Based on breakpoint path nodes, the system performs alternative path calculations, path updates, scheduling window recalculations, and instruction issuance processing to complete dynamic obstacle avoidance and replanning of resource scheduling paths. This includes, for example: Step S460: For each pair of breakpoint path nodes, take the starting spatial path node of the breakpoint path node pair as the search starting point and the ending spatial path node of the breakpoint path node pair as the search target. In the urban spatial grid division reference system, a grid path search algorithm based on heuristic cost function is used to avoid all grid cells with traffic congestion status identifiers as blocked, and calculate the detour alternative path grid sequence.

[0167] The heuristic cost function is defined as f(n) = g(n) + h(n), where g(n) is the actual cost (path length) from the starting point to the current node, and h(n) is the estimated Manhattan distance from the current node to the target node. The search process uses a priority queue, selecting the node with the smallest f(n) value for expansion at each step. Obstacle avoidance logic is implemented by checking the traffic congestion status identifier of the expanded node; if it is 2, the node is skipped. The detour alternative path grid sequence contains the row and column indices of all intermediate grid cells from the starting point to the target node, and its length is typically 1-3 grid cells longer than the original path.

[0168] Step S470: Insert the detour alternative path grid sequence into the corresponding breakpoint path node pair position in the resource scheduling space path point list, replace the original impassable path segment, and generate the updated resource scheduling space path point list.

[0169] The insertion process first locates the position of the breakpoint path node pair in the original path point list. For example, if the original path point list is [N0, N1, N2, N3], where N1→N2 is an impassable path segment, then the detour replacement path grid sequence [N1, A, B, N2] is inserted to replace N1→N2, forming the new path point list [N0, N1, A, B, N2, N3]. The updated resource scheduling space path point list must maintain the same length for both the row index sequence and the column index sequence, and adjacent nodes must be in a grid adjacency relationship.

[0170] Step S480: Recalculate the total grid step length for the updated resource scheduling spatial path point list, and recalculate the resource scheduling time window corresponding to the emergency resource scheduling instruction based on the average travel time parameter of the grid unit and the travel speed reduction coefficient corresponding to the real-time traffic congestion status identifier of the grid unit passed through.

[0171] The total grid step size is equal to the number of nodes in the updated path point column minus 1. The speed reduction factor is determined based on the traffic congestion status identifier: the reduction factor is 1.0 for smooth flow (0), 0.6 for slow flow (1), and 0.2 for blocked flow (2) (this status will not occur after actual obstacle avoidance). The travel time of each grid cell is equal to the average travel time parameter of the grid cell multiplied by the reduction factor, and the total travel time is the sum of the travel times of all grid cells. The start time of the resource scheduling time window remains unchanged, and the end time is the start time plus the total travel time.

[0172] Step S490: Associate and replace the recalculated resource scheduling time window with the updated resource scheduling spatial path point list to generate a dynamic obstacle avoidance and replanning emergency resource scheduling instruction, which overwrites the original emergency resource scheduling instruction in the smart city emergency response scheduling instruction set.

[0173] The associated replacement process updates the "path_row", "path_col", "start_time", and "end_time" fields in the emergency resource scheduling instructions. "Start_time" remains unchanged, while "end_time" is recalculated based on the new total travel time. The updated instructions are matched with the original instructions using the resource unit's unique identifier, achieving in-situ updates of the instruction set. The system records the changes in the time window before and after path replanning as a basis for evaluating scheduling efficiency.

[0174] Step S4100: Send the smart city emergency response dispatch instruction set after dynamic obstacle avoidance and replanning to the emergency resource dispatch execution terminal to trigger the emergency resource unit to execute the emergency response task according to the updated resource dispatch spatial path point list and resource dispatch time window.

[0175] The sending process uses the MQTT protocol, encapsulating the updated instruction set into a JSON format message and publishing it to a specified topic. The emergency resource scheduling execution terminal subscribes to this topic, receives the message, parses the instructions, and executes the scheduling task. The system waits for an acknowledgment message from the terminal; if no acknowledgment is received within 5 seconds, it resends the message, with a maximum of 3 retries.

[0176] For example, the method may further include: step S510: receiving the propagation path probability distribution map of the emergency event in the urban spatial grid identification sequence generated by the event evolution prediction module, and extracting the event propagation probability weight of each grid unit from the propagation path probability distribution map.

[0177] The propagation path probability distribution map is a two-dimensional floating-point array, with the dimension being the number of grids in the city spatial grid identifier sequence. The extraction process iterates through this array, associating the probability weight value (a floating-point number between 0 and 1) of each grid cell with its corresponding grid row and column indices, forming a list of (R, C, p) triples, where p is the event propagation probability weight. Grid cells with p < 0.1 are filtered out during extraction to reduce subsequent computation.

[0178] Step S520: All grid cells in the propagation path probability distribution map are classified and statistically analyzed according to the event propagation probability weight. The number of grid cells whose event propagation probability weight exceeds the first preset activation threshold is counted as the high probability influence area grid count. The number of grid cells whose event propagation probability weight exceeds the second preset activation threshold is counted as the medium probability influence area grid count. The number of grid cells whose event propagation probability weight is lower than the third preset activation threshold is counted as the low probability influence area grid count.

[0179] The preset activation thresholds are set through the system configuration file: the first preset activation threshold is 0.7, the second is 0.3, and the third is 0.1. The hierarchical statistical process iterates through the list of (R, C, p) triples, judging each p value: p > 0.7 is counted in the high-probability affected area grid count; 0.3 ≤ p ≤ 0.7 is counted in the medium-probability affected area grid count; and p < 0.1 is counted in the low-probability affected area grid count. The statistical results are stored as three integer variables.

[0180] Step S530: Input the high-probability impact area grid count, the medium-probability impact area grid count, and the low-probability impact area grid count into the impact intensity mapping function of the emergency event situation assessment branch module. The impact intensity mapping function determines the emergency event spatial coverage breadth parameter based on the ratio of the high-probability impact area grid count to the total number of grids in the urban spatial grid identifier sequence, and simultaneously determines the emergency event spatial clustering degree parameter based on the ratio of the high-probability impact area grid count to the medium-probability impact area grid count.

[0181] The total number of grids in the urban spatial grid identification sequence is M (the sequence length in step S1263). The spatial coverage breadth parameter S = high probability influence area grid count / M, with a value range of 0-1. The spatial clustering parameter C = high probability influence area grid count / (medium probability influence area grid count + 1) (adding 1 to avoid division by zero), with a value range of 0-∞. The influence intensity mapping function is implemented using a piecewise linear function, mapping S and C to quantitative scores of influence intensity.

[0182] Step S540: Perform a weighted summation operation on the emergency event spatial coverage breadth parameter and the emergency event spatial aggregation degree parameter to obtain the quantitative score of the impact intensity of the emergency event at the current moment, and map the quantitative score of the impact intensity to the emergency event impact intensity level descriptor according to the preset impact intensity level division interval.

[0183] The weighted summation formula is: Influence intensity quantification (C values ​​exceeding 2 are calculated as 2), with a result range of 0-1.6. The intensity level classification intervals are: [0, 0.4) → "Slight", [0.4, 0.8) → "Moderate", [0.8, 1.2) → "Slightly Severe", [1.2, 1.6] → "Severe". The mapping process determines the corresponding emergency event intensity level descriptor (string type) based on the score interval.

[0184] Step S550: Extract the decay gradient sequence of the event propagation probability weight as the distance of the grid cell from the core event grid node from the propagation path probability distribution map, perform curve fitting on the decay gradient sequence, and generate the event propagation probability space decay function.

[0185] The core event grid node is (R_e, C_e) marked in step S330. Distance is calculated using the grid Manhattan distance d = |R - R_e| + |C - C_e|. The decay gradient sequence is a set of (d, p) data pairs, where d is the distance and p is the average probability weight of all grid cells at that distance. Curve fitting uses the least squares method to fit the data pairs to an exponential decay function. Where a, b, and c are the fitting parameters. The Levenberg-Marquardt algorithm is used to solve for the optimal parameters during the fitting process.

[0186] Step S560: Perform correlation analysis on the event propagation probability space decay function, the event occurrence time code, and the preset or historically obtained event space propagation rate parameter; calculate the time interval from the start time to the probability peak time of the emergency event based on the half-fading width parameter of the event propagation probability space decay function and the event space propagation rate parameter, and use the time interval as the first duration component required for the emergency event to develop into the mature stage; calculate the time interval from the probability peak time to the time when the probability value is lower than the fourth preset activation threshold based on the tail length parameter of the event propagation probability space decay function and the event space propagation rate parameter, and use the time interval as the second duration component required for the emergency event to decay from the mature stage to the end stage.

[0187] The half-life width parameter d_half is satisfied By solving the equations We obtain d_half = ln(2) / b. The event space propagation rate parameter v (unit: grids / minute) is obtained through historical event statistics, for example, for a fire event v = 0.5 grids / minute. The first duration component t1 = d_half / v. The tail length parameter d_tail is a distance value that satisfies p(d_tail) = 0.1, and the equation is solved. We obtain d_tail. The second duration component t2 = (d_tail - d_half) / v.

[0188] Step S570: Based on the start time indicated by the event occurrence time encoding, the first duration component is added to obtain the estimated event peak time, and the second duration component is added to obtain the estimated event end time, thereby generating an emergency event duration interval prediction descriptor. The emergency event duration interval prediction descriptor includes the estimated event start time, the estimated event peak time, and the estimated event end time.

[0189] The event occurrence time is encoded as a UNIX timestamp T0 (seconds). The estimated peak time of the event is T_p. Estimated end time of the event. The emergency event duration interval prediction descriptor is a dictionary structure containing three timestamps: {"start":T0,"peak":T_peak,"end":T_end}.

[0190] Step S580: Attach the emergency event impact intensity level descriptor and the emergency event duration interval prediction descriptor to the metadata field of the propagation path probability distribution map to generate a propagation path probability distribution map with a situation assessment label.

[0191] Two new key-value pairs have been added to the metadata fields of the propagation path probability distribution map: "severity_level": a descriptor for the severity level of the emergency event's impact, and "duration": a descriptor for the predicted duration interval of the emergency event. The metadata and probability distribution map data are stored together in HDF5 format for easy visualization and analysis later.

[0192] Step S590: Transmit the propagation path probability distribution map with attached situation assessment label to the city emergency command center visualization system to render color layers corresponding to different impact intensity levels on the digital twin city map, and overlay and display the emergency event duration interval prediction descriptor.

[0193] The transmission uses the WebSocket protocol to push data in real time. After receiving the data, the visualization system colors the grid cells according to probability weight values ​​(e.g., red-yellow-green represents high-medium-low probability) and adjusts the transparency according to the intensity of the impact. Time interval information is overlaid in the corner of the map as a text box, displaying the local time of the event's start, peak, and end (converting UNIX timestamps to the local time zone).

[0194] For example, the method may include: step S610: real-time collection of feedback data on the actual impact range of emergency events generated during the execution of smart city emergency response, wherein the feedback data on the actual impact range of emergency events includes the geographic coordinate code of the actual occurrence of the emergency event, the time code of the actual occurrence of the event, and the grid identification sequence of the actual impact range obtained by post-event mapping.

[0195] The actual impact range feedback data of emergency events is entered into the system through the post-event assessment report of the city's emergency command center. The grid identification sequence of the actual impact range is generated manually by staff using a digital twin map to select affected grid units. The data acquisition interface provides a CSV file upload function, containing three fields: R (row index), C (column index), and impact_time (impact timestamp). The system performs format validation on the uploaded data to ensure that the grid index is within a valid range.

[0196] Step S620: The actual geographic coordinates of the emergency event are encoded into actual event grid row index values ​​and actual event grid column index values, and the actual impact range grid identifier sequence is parsed into a set of actual impact grid row index values ​​and a set of actual impact grid column index values.

[0197] The conversion process is the same as step S220, using the urban spatial grid division reference system parameters to convert the real geographic coordinates into grid indices. The real impact range grid identifier sequence is parsed from the CSV file into two integer arrays: the set of real impact grid row index values ​​R_true and the set of real impact grid column index values ​​C_true, both of which are of equal length and correspond one-to-one.

[0198] Step S630: Construct a real event spatial propagation trajectory map with the real event grid row index value and the real event grid column index value as the center and the real influence grid row index value set and the real influence grid column index value set as the boundary.

[0199] The construction method is similar to step S230, with the central grid being (R_true0, C_true0), nodes being actual affected grid cells, and edge directions determined by the impact timestamp `impact_time` (from the first affected node to the next affected node). The edge weight is the time difference. If `impact_time` is missing, the default edge weight is 1.

[0200] Step S640: Input the actual geographical coordinates of the emergency event into the emergency response decision network, generate the corresponding propagation path probability distribution map in sequence, and extract the set of predicted impact grid row index values ​​and the set of predicted impact grid column index values ​​from the propagation path probability distribution map.

[0201] The input process simulates a real-time emergency response flow, using real geographic coordinates as new event trigger signals input into the emergency response decision network, and executing steps S110-S140 to generate a propagation path probability distribution map. The method for extracting the predicted impact grid row index value set R_pred and column index value set C_pred is the same as in step S1463, taking the top K grid cells with probability weights (K equals the number of actual impact grids).

[0202] Step S650: Calculate the number of spatially overlapping grids between the predicted impact grid row index value set and the predicted impact grid column index value set and the actual impact grid row index value set and the actual impact grid column index value set, and use the ratio of the number of spatially overlapping grids to the total number of actual impact grids as the prediction accuracy feedback value of the current emergency response event.

[0203] The number of grid cells in the spatially overlapping region is calculated using the intersection of sets: (R_pred[i], C_pred[i]) and (R_true[j], C_true[j]) are converted into a set of tuples, and the size of the intersection is calculated. The prediction accuracy feedback value acc = intersection size / len(R_true), with a value range of 0-1.

[0204] Step S660: Compare the prediction accuracy feedback value with a preset accuracy threshold. If the prediction accuracy feedback value is lower than the preset accuracy threshold, trigger the incremental learning update module.

[0205] The preset accuracy threshold is set through system configuration, with a default value of 0.7. During the comparison process, if acc < 0.7, the incremental learning trigger flag is set to True; otherwise, it is set to False. The trigger flag is stored in a system status variable and is periodically checked by the incremental learning update module.

[0206] Step S670: Sample a preset number of real propagation path nodes and their corresponding real propagation direction edges from the real event space propagation trajectory map to form an incremental training sample batch.

[0207] The preset number is set via hyperparameters, with a default value of 32. Sampling employs a random walk algorithm, randomly selecting a starting point from the real event space propagation trajectory graph and generating paths along directed edges. Each path contains 5 nodes and 4 edges. A total of 32 paths are sampled to form an incremental training sample batch, with each sample containing path node features and edge attributes.

[0208] Step S680: Input the incremental training samples into the event evolution prediction module in batches, freeze all network layer parameters in the event evolution prediction module except for the last gated recurrent unit layer, and perform gradient calculation and backpropagation update only on the trainable weight matrix of the last gated recurrent unit layer.

[0209] The freezing process is achieved by setting the "requires_grad" attribute of the network layers, with only the weight matrix and bias vector of the last GRU layer set to trainable. During forward propagation, incremental training sample batches are processed through graph convolutional layers and the first few GRU layers, with the last GRU layer outputting the prediction result.

[0210] Step S690: With the goal of minimizing the cross-entropy loss function between the predicted influence grid row index value set and the predicted influence grid column index value set and the actual influence grid row index value set and the actual influence grid column index value set, the weight parameters of the last gated recurrent unit level are adjusted using the mini-batch stochastic gradient descent algorithm.

[0211] The cross-entropy loss function calculates the loss between the predicted probability distribution and the true labels (true influence grid cells are 1, others are 0). The optimizer uses SGD with a learning rate of 0.0001 (much smaller than in the pre-training stage), a batch size of 8, and 10 iterations. In each iteration, the gradient is calculated and the weight parameters of the last GRU layer are updated.

[0212] Step S6100: After completing the weight parameter update, write the updated weight parameters back to the event evolution prediction module, and record the event trigger time encoding of this incremental learning and the change in the prediction accuracy feedback value before and after the update.

[0213] The write-back process saves the optimized weight parameters to the corresponding network layer of the event evolution prediction module, overwriting the original parameters. The change delta_acc = updated acc - original acc, and is recorded in the incremental learning log file along with the event trigger time code (current UNIX timestamp).

[0214] Step S6110: Write the change in the prediction accuracy feedback value before and after the update into the operation log of the emergency response decision network.

[0215] The operation logs are in CSV format and include fields such as timestamp, event ID, and delta_acc. The system archives the operation logs daily for analyzing incremental learning effectiveness and model performance trends. Log files are stored on a secure storage server and retained for one year.

[0216] In one exemplary embodiment, a smart city emergency response system incorporating artificial intelligence is provided. This system may include terminals, servers, etc., and its internal structure diagram may be as follows. Figure 2 As shown, this smart city emergency response system incorporating artificial intelligence includes a processor, memory, input / output interfaces, a communication interface, a display unit, and input devices. The processor, memory, and input / output interfaces are connected via a system bus, and the communication interface, display unit, and input devices are also connected to the system bus via the input / output interfaces. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The input / output interfaces are used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, near-field communication, or other technologies. When the computer program is executed by the processor, it implements a smart city emergency response method incorporating artificial intelligence. The display unit is used to form a visually visible image and can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device can be a touch layer covering the display screen, or a button, trackball, or touchpad set on the shell of a smart city emergency response system that integrates artificial intelligence, or an external keyboard, touchpad, or mouse, etc.

[0217] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A smart city emergency response method combining artificial intelligence, characterized in that, The method includes: Receive urban emergency event trigger signal, wherein the urban emergency event trigger signal includes the geographic coordinate code of the event occurrence and the event occurrence time code; The spatial positioning branch module of the emergency response decision network is activated based on the geographic coordinate code of the event occurrence, and spatial grid mapping processing is performed on the geographic coordinate code of the event occurrence to generate a city spatial grid identifier sequence corresponding to the geographic coordinate code of the event occurrence. The urban spatial grid identifier sequence is input into the resource association analysis module of the emergency response decision network. Based on the urban spatial grid identifier sequence, the emergency resource deployment database is traversed to extract the set of emergency resource unit descriptors that are directly related to the urban spatial grid identifier sequence. Activate the event evolution prediction module of the emergency response decision network, and perform temporal dependency modeling on the emergency resource unit descriptor set based on the emergency resource unit descriptor set and in conjunction with the event occurrence time code, to generate a probability distribution map of the propagation path of emergency events in the urban spatial grid identification sequence; A smart city emergency response dispatch instruction set is generated based on the propagation path probability distribution map. The smart city emergency response dispatch instruction set includes a resource dispatch time window and a resource dispatch spatial path point list for each emergency resource unit in the emergency resource unit descriptor set.

2. The smart city emergency response method combining artificial intelligence according to claim 1, characterized in that, The step of activating the spatial positioning branch module of the emergency response decision network based on the geographic coordinate code of the event occurrence, performing spatial grid mapping processing on the geographic coordinate code of the event occurrence, and generating a city spatial grid identifier sequence corresponding to the geographic coordinate code of the event occurrence, includes: The event occurrence geographic coordinate code carried by the urban emergency event trigger signal is analyzed, and the longitude coordinate component and latitude coordinate component are extracted from the event occurrence geographic coordinate code; The longitude and latitude coordinate components are input into the spatial positioning branch module. The longitude coordinate components are converted into grid row index values ​​according to the urban spatial grid division reference system. At the same time, the latitude coordinate components are converted into grid column index values ​​according to the urban spatial grid division reference system. The grid row index value corresponding to the longitude coordinate component and the grid column index value corresponding to the latitude coordinate component are output. Centered on the grid cell located by both the grid row index value and the grid column index value, neighboring grid cells are searched outward layer by layer according to a preset number of neighborhood expansion layers to generate a set of neighboring grid identifiers that includes the identifier of the grid cell itself and the identifiers of all neighboring grid cells. The grid row index value, the grid column index value, and the set of adjacent grid identifiers are combined and encoded to form an initial grid identifier corresponding to the geographic coordinate code of the event occurrence. The initial grid identifier records the spatial inclusion relationship of the grid row index value, the grid column index value, and the identifier of each adjacent grid unit in the set of adjacent grid identifiers in a hierarchical encoding structure. Traverse the urban spatial grid hierarchy tree, extract the set of upper-level grid identifiers that have a direct parent-level relationship with the initial grid identifier from the urban spatial grid hierarchy tree, and extract the set of lower-level grid identifiers that have a direct child-level relationship with the initial grid identifier from the urban spatial grid hierarchy tree. Based on the initial grid identifier, hierarchical aggregation, semantic enhancement, sequence sorting, compression encoding and time binding processing are performed to output a sequence of urban spatial grid identifiers with time validity markings.

3. The smart city emergency response method combining artificial intelligence according to claim 2, characterized in that, The process of performing hierarchical aggregation, semantic enhancement, sequence sorting, compression encoding, and time binding based on the initial grid identifier outputs a sequence of urban spatial grid identifiers with time validity markers, including: Based on spatial inclusion relationships, the upper-level grid identifier set, the initial grid identifier, and the lower-level grid identifier set are logically aggregated to generate a complete spatial grid hierarchical identifier list corresponding to the geographic coordinate code of the event occurrence. The identifiers in the complete spatial grid hierarchical identifier list are organized and arranged in order from macro to micro according to the spatial scale of their corresponding grids. A semantic label mapping operation is performed on each grid identifier in the complete spatial grid hierarchical identifier list. A corresponding urban functional area type label is attached to each grid identifier through a pre-trained urban functional area identification model. At the same time, an infrastructure density level descriptor is attached to each grid identifier according to the infrastructure distribution density of the area corresponding to the grid identifier, so as to obtain a spatial grid identifier unit with enhanced semantics. All semantically enhanced spatial grid identifier units are sorted according to the hierarchical order of spatial inclusion relationships to form a city spatial grid identifier sequence with a progressive arrangement structure from the upper macro grid to the lower micro grid; The urban spatial grid identifier sequence is subjected to redundancy compression encoding processing. The duplicate urban functional area type label information and duplicate infrastructure density level descriptor information between adjacent enhanced semantic spatial grid identifier units in the urban spatial grid identifier sequence are identified and removed to generate a compressed urban spatial grid identifier sequence. The compressed urban spatial grid identifier sequence is timestamped with the event occurrence time code in the urban emergency event triggering signal, and a time validity mark is added to the compressed urban spatial grid identifier sequence, and the urban spatial grid identifier sequence with the time validity mark is output.

4. The smart city emergency response method combining artificial intelligence according to claim 1, characterized in that, The resource association analysis module that inputs the urban spatial grid identifier sequence into the emergency response decision network traverses the emergency resource deployment database based on the urban spatial grid identifier sequence to extract a set of emergency resource unit descriptors that are directly related to the urban spatial grid identifier sequence, including: The urban functional area type label, infrastructure density level descriptor, grid row index value, and grid column index value contained in each spatial grid identifier unit with enhanced semantics are extracted sequentially from the urban spatial grid identifier sequence. The city functional area type label and infrastructure density level descriptor of each enhanced semantic spatial grid identification unit are input into the resource association analysis module. According to the preset city functional area emergency resource mapping table, the city functional area type label is matched with the corresponding emergency resource category code set. At the same time, the infrastructure density level descriptor is matched with the corresponding resource deployment density threshold range. The grid row index value and grid column index value are converted into a spatial query window coordinate range, which covers the grid cell located by the grid row index value and grid column index value, as well as the adjacent grid cells located by all adjacent grid cell identifiers in the adjacent grid identifier set. Using the coordinate range of the spatial query window as a spatial constraint and each emergency resource category code in the emergency resource category code set as a resource category filtering condition, a structured query command is jointly constructed and sent to the emergency resource deployment database; Receive a preliminary set of emergency resource unit records returned by the emergency resource deployment database, which matches both the coordinate range of the spatial query window and the set of emergency resource category codes. Each preliminary emergency resource unit record in the preliminary set of emergency resource unit records contains the resource unit geographic coordinates, resource unit category code, resource unit current status identifier, and resource unit available quantity fields. Based on the initial emergency resource unit record set, the system performs filtering, sorting, weighting, descriptor encapsulation, and tag attachment processing, outputting an emergency resource unit descriptor set with spatial traceability tags.

5. The smart city emergency response method combining artificial intelligence according to claim 4, characterized in that, The process of filtering, sorting, weighting, descriptor encapsulation, and tag attachment based on the preliminary emergency resource unit record set outputs a set of emergency resource unit descriptors with spatial tracing tags, including: Filter out all preliminary emergency resource unit records whose current status identifier indicates a non-idle state from the preliminary emergency resource unit record set, and generate a candidate emergency resource unit record set that is in a schedulable state; Extract the resource unit geographic coordinates of each candidate emergency resource unit record in the candidate emergency resource unit record set, calculate the spatial straight-line distance between the resource unit geographic coordinates and the event occurrence geographic coordinate code, and add the spatial straight-line distance as a distance association weight factor to the candidate emergency resource unit record; Based on the distance correlation weight factor, all candidate emergency resource unit records in the candidate emergency resource unit record set are sorted in ascending order to generate a resource unit priority list arranged according to spatial proximity. Select candidate emergency resource unit records whose sorting position is within the preset ratio range of the preceding order from the resource unit priority list. Encapsulate the resource unit geographical coordinates, resource unit category code, resource unit current status identifier, resource unit available quantity field, and distance association weight factor in the selected candidate emergency resource unit records into an emergency resource unit descriptor, forming a set of emergency resource unit descriptors that are directly related to the urban spatial grid identification sequence. The set of emergency resource unit descriptors is associated with and stored in relation to the event occurrence time code. A spatial tracing tag of the urban spatial grid identification sequence is attached to the set of emergency resource unit descriptors, and the set of emergency resource unit descriptors with the spatial tracing tag is output.

6. The smart city emergency response method combining artificial intelligence according to claim 1, characterized in that, The event evolution prediction module that activates the emergency response decision network, based on the set of emergency resource unit descriptors and in conjunction with the event occurrence time code, performs temporal dependency modeling on the set of emergency resource unit descriptors to generate a probability distribution map of the propagation path of emergency events in the urban spatial grid identifier sequence, including: Activate the event evolution prediction module of the emergency response decision network, and extract the resource unit geographic coordinates, resource unit category code and distance association weight factor contained in each emergency resource unit descriptor from the emergency resource unit descriptor set; Map the geographical coordinates of each emergency resource unit descriptor to the urban spatial grid division reference system, calculate the resource stationary grid row index value and resource stationary grid column index value corresponding to the geographical coordinates of the resource unit in the urban spatial grid division reference system, and generate a resource stationary grid identification unit. Spatial overlay analysis is performed on the resource-residence grid identifier unit and the enhanced semantic spatial grid identifier unit in the urban spatial grid identifier sequence to identify the spatial inclusion relationship or spatial adjacency relationship between the resource-residence grid identifier unit and the enhanced semantic spatial grid identifier unit, and spatial association edges between the emergency resource unit and the urban spatial grid identifier sequence are established based on the spatial inclusion relationship or spatial adjacency relationship. Using each enhanced semantic spatial grid identifier unit in the urban spatial grid identifier sequence as a node, and the spatial association edge between the emergency resource unit and the urban spatial grid identifier sequence as the initial connection edge, an emergency event propagation spatial topology map is constructed. The node attributes of the emergency event propagation spatial topology map include urban functional area type labels and infrastructure density level descriptors, and the edge attributes of the emergency event propagation spatial topology map include distance association weight factors. The emergency event propagation spatial topology map is input into the graph structure temporal convolution unit of the event evolution prediction module. The spatial domain graph convolution operation is performed on the emergency event propagation spatial topology map to aggregate the urban functional area type label information and infrastructure density level descriptor information of the neighboring areas of each node, and generate the node spatial neighborhood aggregation feature vector. Based on the feature vectors aggregated by the node spatial neighborhood, spatiotemporal feature fusion, temporal dependency modeling, probability weight allocation and distribution map generation are performed to generate a probability distribution map of the propagation path of emergency events in the urban spatial grid identification sequence.

7. The smart city emergency response method combining artificial intelligence according to claim 1, characterized in that, The generation of a smart city emergency response dispatch instruction set based on the propagation path probability distribution map includes: Analyze the propagation path probability distribution map, extract all grid cells whose event propagation probability weight exceeds a preset activation threshold from the propagation path probability distribution map, and mark the grid cells as the target grid cell set; Each target grid cell in the target grid cell set is sorted in descending order according to the event propagation probability weight to generate a priority sequence of threatened grid cells arranged according to the urgency of the event impact. Read sequentially from the set of emergency resource unit descriptors the geographical coordinates of each emergency resource unit, the resource unit category code, the available quantity of resource units, and the distance association weight factor contained in each emergency resource unit descriptor; For the currently processed emergency resource unit descriptor, match its resource unit category code with a preset emergency resource scheduling priority mapping table to obtain the resource scheduling priority level corresponding to the emergency resource unit descriptor; Map the geographical coordinates of the resource unit of the current emergency resource unit descriptor to the urban spatial grid division reference system, generate the resource-residence grid row index value and resource-residence grid column index value corresponding to the emergency resource unit descriptor, and form a resource-residence grid identifier; Based on the resource-resident grid identifier, the system performs path planning, weight evaluation, scheduling parameter calculation, and instruction generation and processing to output a smart city emergency response scheduling instruction set.

8. The smart city emergency response method combining artificial intelligence according to claim 1, characterized in that, The emergency response decision network includes a pre-trained spatial feature encoding backbone network, which constructs its network structure and connections through the following steps: A collection of historical emergency event records for the city is collected. Each historical emergency event record in the collection contains the geographic coordinate code of the historical event, the time code of the historical event, and the grid identifier sequence of the actual impact range of the historical event. For each historical emergency event record in the city's historical emergency event record set, spatial gridding processing is performed, and the geographic coordinates of the historical event occurrence are encoded and converted into historical event grid row index values ​​and historical event grid column index values. The actual impact range grid identifier sequence of the historical event is parsed into a set of historical impact grid row index values ​​and a set of historical impact grid column index values. Centered on the historical event grid row index value and the historical event grid column index value, and with the historical influence grid row index value set and the historical influence grid column index value set as the spatial expansion boundary, a historical event spatial propagation trajectory map is constructed. The historical event spatial propagation trajectory map has grid cells as nodes and the event propagation direction as directed edges. The historical event spatial propagation trajectory map is input into the initialized graph convolutional neural network. The graph convolutional neural network includes an input layer, multiple stacked graph convolutional hidden layers, and an output layer. The input layer receives the spatial adjacency matrix and the node initial feature vector of each grid node. The node initial feature vector is formed by concatenating the urban functional area type label encoding and the infrastructure density level descriptor encoding corresponding to the grid cell. The graph convolutional hidden layer performs Laplacian normalization on the input spatial adjacency matrix, and performs linear transformation on the node feature vector through a trainable weight matrix. At the same time, it aggregates the feature vectors of neighboring nodes to generate an updated node feature vector. Based on the updated node feature vectors, prediction mapping, accuracy calculation, weight optimization, network solidification and embedding processes are performed to construct a spatial feature coding backbone network and complete its deployment.

9. The smart city emergency response method combining artificial intelligence according to claim 1, characterized in that, The emergency response decision network also includes a resource scheduling path optimization branch module. This module performs dynamic obstacle avoidance and replanning on the resource scheduling spatial path point list in the smart city emergency response scheduling instruction set, including: Receive the smart city emergency response dispatch instruction set, and extract the resource dispatch spatial path point column contained in each emergency resource dispatch instruction from the smart city emergency response dispatch instruction set. The resource dispatch spatial path point column is composed of a continuous grid row index value sequence and a continuous grid column index value sequence. The grid cell corresponding to each grid row index value and grid column index value in the resource scheduling spatial path point column is converted into a spatial path node, and the directed connection edge between adjacent spatial path nodes is converted into a path movement segment. The traffic congestion status identifiers of the grid units covered by the resource scheduling spatial path point column at the current moment are obtained from the urban real-time traffic flow monitoring platform. The traffic congestion status identifiers include three status categories: smooth flow, slow flow, and blocked flow. Traverse each path movement segment in the resource scheduling spatial path point list, identify the grid row index value and grid column index value of the two spatial path nodes connected by the path movement segment, and query the traffic congestion status identifier of the corresponding grid cell based on the grid row index value and grid column index value; If the traffic congestion status identifier of any grid cell covered by the path movement segment is blocked, then the path movement segment is marked as an impassable path segment, and the starting spatial path node and the ending spatial path node of the path movement segment are recorded as a breakpoint path node pair. Based on the breakpoint path nodes, the system performs alternative path calculation, path update, scheduling window recalculation, and instruction issuance processing to complete the dynamic obstacle avoidance and replanning of resource scheduling paths.

10. A smart city emergency response system incorporating artificial intelligence, characterized in that: include: processor; A machine-readable storage medium for storing machine-executable instructions of the processor; The processor is configured to execute the smart city emergency response method incorporating artificial intelligence as described in any one of claims 1 to 9 by executing the machine-executable instructions.

Images