Artificial intelligence-based geospatial data management method and system

Abstract

The application provides a kind of geographic space data management method and system based on artificial intelligence, first acquire the geographic space data set of target area, include time series variation geographic entity observation record and corresponding spatial location identification, then the feature extraction is carried out to geographic space data set, generate time series correlation feature and spatial topology feature, then call pre-training geographic space analysis model to time series correlation feature and spatial topology feature are dynamically characteristic fusion processing, generate feature set after fusion, again based on feature set after fusion generates the geographic space optimization strategy containing geographic entity state prediction result and resource allocation priority configuration, finally according to the geographic space optimization strategy triggers the geographic information system update operation of target area, to realize the dynamic management and timely update of geographic space data, improve the accuracy, timeliness and effectiveness of geographic space data management.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, in particular to a geographic space data management method and system based on artificial intelligence. BACKGROUND

[0002] The management of geographic space data is crucial for many fields such as urban planning, resource allocation, environmental monitoring, and disaster warning. With the rapid development of information technology, the acquisition methods of geographic space data are increasingly diversified, and the data volume is also growing explosively. However, the existing methods of geographic space data management have many limitations. On the one hand, traditional data management methods often focus on the storage and query of static geographic space data, and are difficult to effectively process time-varying geographic entity observation records, unable to capture the evolution rules of geographic entities over time, resulting in a lag in the cognition of geographic space state. On the other hand, in the processing of geographic space data, the existing methods mostly treat time sequence features and spatial features in isolation, lack of in-depth mining and comprehensive analysis of the correlation between the two, and are difficult to generate comprehensive and accurate geographic space optimization strategies, unable to provide scientific and effective basis for resource allocation, geographic information system update and other operations, thereby affecting the application effect and decision quality of geographic space data management in various fields. SUMMARY

[0003] In view of the above-mentioned problems, in combination with the first aspect of the present application, the embodiments of the present application provide a geographic space data management method based on artificial intelligence, which comprises:

[0004] acquiring a geographic space data set of a target area, the geographic space data set comprising time-varying geographic entity observation records and corresponding spatial location identifiers;

[0005] performing feature extraction on the geographic space data set to generate time sequence correlation features and spatial topology features of the geographic entity observation records;

[0006] calling a pre-trained geographic space analysis model to perform dynamic feature fusion processing on the time sequence correlation features and the spatial topology features to generate a fused feature set;

[0007] generating a geographic space optimization strategy based on the fused feature set, the geographic space optimization strategy comprising a geographic entity state prediction result and a resource allocation priority configuration;

[0008] According to the geographic space optimization strategy, a geographic information system update operation of the target area is triggered. In still another aspect, the embodiment of the present application also provides a geographic space data management system based on artificial intelligence, which comprises a processor, a machine readable storage medium, the machine readable storage medium is connected with the processor, the machine readable storage medium is used for storing programs, instructions or codes, and the processor is used for running the programs, instructions or codes in the machine readable storage medium to realize the above-mentioned method.

[0009] Based on the above aspects, by acquiring a geographic space data set of a target area, covering time-varying geographic entity observation records and corresponding spatial position identifiers, feature extraction is performed on the geographic space data set to generate time-correlation features and spatial topology features, which accurately depict the characteristics of the geographic entity from two key dimensions of time and space. The pre-trained geographic space analysis model is called to perform dynamic feature fusion processing on the time-correlation features and the spatial topology features, fully considering the internal relationship between the time and space features, so that the feature set after fusion is more comprehensive and representative. Based on the fused feature set, a geographic space optimization strategy is generated, which contains a geographic entity state prediction result and a resource allocation priority configuration. Finally, according to the geographic space optimization strategy, a geographic information system update operation of the target area is triggered, realizing dynamic management and timely update of geographic space data, improving the accuracy, timeliness and effectiveness of geographic space data management, and helping to better serve practical application scenarios such as urban planning and resource allocation. BRIEF DESCRIPTION OF DRAWINGS

[0010] Figure 1 is the execution flow diagram of the geographic space data management method based on artificial intelligence provided by the embodiment of the present application.

[0011] Figure 2 is the hardware architecture diagram of the geographic space data management system based on artificial intelligence provided by the embodiment of the present application. DETAILED DESCRIPTION

[0012] The present application will be specifically described below in conjunction with the drawings of the specification, Figure 1 is the flow diagram of the geographic space data management method based on artificial intelligence provided by an embodiment of the present application, and the geographic space data management method based on artificial intelligence will be described in detail below.

[0013] Step S110: acquiring a geographic space data set of a target area, the geographic space data set comprising time-varying geographic entity observation records and corresponding spatial position identifiers.

[0014] In this embodiment, the target area can be a piece of urban new area to be developed, a large mining area or a natural reserve area that needs to be monitored, etc. The geographical space data set can be obtained from a wide range of sources, such as satellite remote sensing images, unmanned aerial vehicle survey data, and data collected by ground surveying equipment. For satellite remote sensing images, they can periodically capture the target area to obtain geographical information at different time points, which can reflect the time sequence changes of geographical entities, such as changes in land use types and changes in vegetation coverage. Ground surveying equipment, such as total station and GPS receiver, can accurately measure the spatial position of geographical entities to obtain corresponding spatial position identifiers, which are usually presented in the form of latitude and longitude coordinates or plane rectangular coordinates.

[0015] Suppose the target area is observed at different times t1, t2, t3, etc. by satellite remote sensing, and the obtained geographical entity observation records are represented by set R, R = {r1, r2, r3, …}, where each ri contains various attribute observation values of the geographical entity at a specific time point, such as reflectivity and elevation. The corresponding spatial position identifiers are represented by set P, P = {p1, p2, p3, …}, pi is the specific position coordinates of the geographical entity in space. In this way, the geographical space data set D can be represented as D = (R, P), which integrates the time sequence change information and spatial position information of the geographical entity.

[0016] Step S120: Feature extraction is performed on the geographical space data set to generate time sequence correlation features and spatial topology features of the geographical entity observation records.

[0017] After obtaining the geographical space data set, in order to deeply mine the useful information in the data, feature extraction operation is needed to generate time sequence correlation features and spatial topology features. Time sequence correlation features can reveal the change rule and mutual relationship of geographical entity observation records over time, while spatial topology features can reflect the connection and distribution characteristics of geographical entities in space.

[0018] Step S121: Data integrity check operation is performed on the geographical entity observation records in the geographical space data set to detect missing data timestamps and spatial position identifiers, and interpolation algorithm based on adjacent observation records is used to generate supplementary data to form a continuous spatio-temporal coverage preprocessed geographical entity data set.

[0019] During the collection of geographical space data, due to equipment failure, weather conditions and other factors, the geographical entity observation records may have missing data. In order to ensure the integrity and continuity of the data, data integrity check operation is needed.

[0020] First, check each observation record in the geospatial data set, for each timestamp and spatial location identifier in the record, determine whether there is a corresponding observation value. By traversing the observation record set R and the spatial location identifier set P in the geospatial data set D, for each timestamp t and spatial location p, check whether there is a corresponding observation value r. If a certain timestamp t and spatial location p are found to have no corresponding observation value, mark it as missing data and record the timestamp and spatial location identifier of the missing data.

[0021] Suppose in the time series T = {t1, t2, …, tn}, it is found that the observation value of spatial location p5 at time point t3 is missing. Next, based on the interpolation algorithm of adjacent observation records, the supplementary data is generated. Taking linear interpolation as an example, for the missing data point, find its adjacent time points t2 and t4 and the observation values r2 and r4 of the corresponding spatial location p5. Let the missing data be r3, according to the linear interpolation formula, r3 = r2 + (r4 - r2) * ((t3 - t2) / (t4 - t2)). In this way, all missing data is supplemented, and finally a continuous spatio-temporal coverage preprocessed geospatial entity data set D' is formed, so that the data has continuity in time and space.

[0022] Step S122: Perform outlier detection processing on the preprocessed geospatial entity data set, calculate the numerical distribution range of each geospatial entity observation record, remove data points outside the preset distribution range, and perform sliding window mean filling processing on the removed gaps to generate a standardized geospatial entity data set.

[0023] After obtaining the preprocessed geospatial entity data set, there may be some outliers in it, which may be caused by measurement errors, data transmission errors, etc. In order to ensure the quality of the data, outlier detection processing is needed.

[0024] Calculate the numerical distribution range of each geospatial entity observation record. For each attribute feature, its distribution range can be determined by statistical methods. For example, calculate the mean μ and standard deviation σ of each attribute feature, and then set a preset multiple k, and take the mean plus or minus k times the standard deviation as the normal distribution range of the data, i.e. [μ-kσ, μ+kσ]. For each observation record r in the preprocessed geospatial entity data set D', check whether the value of each attribute feature is within the corresponding normal distribution range. If the value of a certain attribute feature is outside the range, mark the data point as an outlier.

[0025] Assuming that for a certain attribute feature A, the mean μA and the standard deviation σA are calculated, and the preset multiple k = 3. If the value of the attribute feature A in the observation record r is not within the range of [μA-3σA, μA+3σA], the observation record is marked as an outlier. After removing these data points that exceed the preset distribution range, there will be a vacancy position left in the data. In order to fill in these vacancy positions, a sliding window mean filling process is adopted.

[0026] Taking a sliding window of length m as an example, for the vacancy position, take the m / 2 adjacent normal data points before and after it (if there are not enough m / 2 data points before and after, take as many data points as possible), calculate the mean of these data points as the filling value of the vacancy position. Assuming that the vacancy position is rj, and the adjacent normal data points before and after it are rj-i, …, rj-1, rj+1, …, rj+i (i = m / 2), then the filling value is the mean of the attribute feature values corresponding to these data points. In this way, all vacancy positions are filled, and the standardized geographic entity data set D” is finally generated, making the data more standardized and reliable.

[0027] Step S123: performing a time series segmentation operation on the standardized geographic entity data set, dividing the geographic entity observation records into multiple time window subsequences according to a preset period length, each time window subsequence containing a fixed number of consecutive observation records, and performing an overlapping cutting process on the observation records at the time window boundaries to avoid data truncation.

[0028] In order to better analyze the time sequence characteristics of the geographic entity observation records, a time series segmentation operation needs to be performed on the standardized geographic entity data set. According to a preset period length, the geographic entity observation records are divided into multiple time window subsequences.

[0029] Step S1231: obtaining the timestamp sequence of all geographic entity observation records in the standardized geographic entity data set, sorting the observation records according to the chronological order of the timestamps, and generating an ordered time series data stream.

[0030] For the standardized geographic entity data set D”, first extract the timestamps of all observation records to form a timestamp sequence T = {t1, t2, …, tn}. Then, according to the chronological order of the timestamps, the observation records are sorted so that they are arranged in chronological order, forming an ordered time series data stream S. Sorting can be done using common sorting algorithms such as quicksort, mergesort, etc. Through sorting, the time sequence of the geographic entity observation records can be clearly presented.

[0031] Step S1232: determining the time window length and sliding step parameters according to a preset division rule, and dividing the ordered time series data stream into multiple initial time window subsequences in a sliding window manner.

[0032] The preset division rule is determined according to specific analysis requirements and variation characteristics of geographical entities. It is assumed that the length of a time window is w and the sliding step is s. The ordered time series data stream S is segmented in a sliding window manner. Starting from the beginning of the data stream, each time, a continuous observation record with a length of w is selected as a time window subsequence. Then, the window is slid forward by s, and the next time window subsequence is continuously selected until the entire ordered time series data stream is covered. For example, the first time window subsequence S1 contains observation records from the 1st to the wth, the second time window subsequence S2 contains observation records from the s+1th to the s+wth, and so on, to generate a plurality of initial time window subsequences.

[0033] Step S1233: The number of observation records in each initial time window subsequence is detected. If the number is lower than a preset minimum record threshold, a window expansion operation is performed to dynamically adjust the window length to contain a sufficient number of observation records.

[0034] In order to ensure that each time window subsequence contains a sufficient number of observation records for effective analysis, the number of observation records in each initial time window subsequence needs to be detected. A minimum record threshold th is preset. For each initial time window subsequence Si, the number of observation records ni in it is counted. If ni is lower than th, a window expansion operation is performed.

[0035] The window expansion operation can be realized by dynamically adjusting the window length. For example, each time, the window length is increased by a fixed increment δ, and the observation records are reselected until the number of observation records in the new time window subsequence reaches or exceeds the minimum record threshold th. It is assumed that the number of observation records n1 in the initial time window subsequence S1 is lower than th. The window length is increased from w to w+δ, and the observation records from the 1st to the w+δth are reselected as a new time window subsequence. The number of observation records is checked again until the requirement is met.

[0036] Step S1234: The expanded time window subsequences are subjected to time alignment processing, and a dynamic time warping algorithm is used to align the time offsets of observation records in different time windows to eliminate the time sequence misalignment caused by inconsistent data collection frequencies.

[0037] Because the data collection frequencies can be inconsistent, the observation records in different time windows can have time offsets, resulting in time sequence misalignment. In order to eliminate this influence, the expanded time window subsequences are subjected to time alignment processing, and a dynamic time warping (DTW) algorithm is used.

[0038] The DTW algorithm aligns two time series in time by finding the optimal matching path between them. For two different time window subsequences Sj and Sk, a distance matrix is first calculated, where each element in the distance matrix represents the distance between two observation records. Then, by dynamic programming, the optimal path from the top-left corner to the bottom-right corner of the distance matrix is found, and the elements on the optimal path correspond to the optimal matching between the two time series. According to the optimal path, the observation records in the time window subsequences are adjusted so that they are aligned in time. In this way, the time series misalignment caused by inconsistent data collection frequencies is eliminated, ensuring the time consistency between time window subsequences.

[0039] Step S1235: Normalize the aligned time window subsequences, associate the normalized time window subsequences with the corresponding spatial position identifiers, and establish a bidirectional mapping relationship between the time window index and the spatial region identifier.

[0040] To eliminate the dimensional differences between different attribute features, the aligned time window subsequences are normalized. Common normalization methods can be used, such as min-max normalization or Z-score normalization. Taking min-max normalization as an example, for each attribute feature in each time window subsequence, find its minimum value min and maximum value max, and normalize each observation value x of the attribute feature to the interval [0, 1], with the normalization formula x' = (x-min) / (max-min).

[0041] The normalized time window subsequences are associated with the corresponding spatial position identifiers. For each time window subsequence Si, record its corresponding spatial position identifier pi. At the same time, a bidirectional mapping relationship between the time window index and the spatial region identifier is established. The time window index can be used to easily find a certain time window subsequence, while the spatial region identifier is used to determine the geographical spatial position corresponding to the time window subsequence. Through this bidirectional mapping relationship, data can be quickly queried and analyzed in time and spatial dimensions.

[0042] Step S124: Perform trend decomposition processing on each time window subsequence, separate the long-term trend component, seasonal period component, and residual noise component of the geographical entity observation records using the seasonal and trend decomposition algorithm, and extract the slope change feature of the long-term trend component and the amplitude fluctuation feature of the seasonal period component as the time series correlation feature.

[0043] After completing the division and processing of the time window subsequences, perform trend decomposition processing on each time window subsequence to extract the long-term trend component, seasonal period component, and residual noise component.

[0044] The seasonal and trend decomposition algorithm, such as the STL (Seasonal-Trend decomposition using Loess) algorithm, is adopted. The STL algorithm decomposes the time series into long-term trend components, seasonal cycle components, and residual noise components through a local weighted regression method. For each time window subsequence Si, it is input into the STL algorithm to obtain the long-term trend component Ti, the seasonal cycle component Si, and the residual noise component Ri, that is, Si = Ti + Si + Ri.

[0045] The slope change feature of the long-term trend component is extracted, and the slope change reflects the change rate of the observation record of the geographic entity in the long term. The slope change feature can be obtained by calculating the slope of the long-term trend component at different time points. For example, for the adjacent two time points ti and ti+1 in the long-term trend component Ti, the slope ki = (Ti+1-Ti) / (ti+1-ti) is calculated, and these slope values are taken as the slope change feature of the long-term trend component.

[0046] The amplitude fluctuation feature of the seasonal cycle component is extracted, and the amplitude fluctuation reflects the change amplitude of the observation record of the geographic entity in the seasonal cycle. The amplitude fluctuation feature can be obtained by calculating the difference between the maximum and minimum values of the seasonal cycle component in a seasonal cycle. For example, for the seasonal cycle component Si, find its maximum value maxi and minimum value mini in a seasonal cycle, and the amplitude fluctuation feature is ai = maxi-mini. The slope change feature of the long-term trend component and the amplitude fluctuation feature of the seasonal cycle component are taken as the time series correlation features, which can reflect the change law and mutual relationship of the observation record of the geographic entity over time.

[0047] Step S125: performing spatial clustering analysis processing based on the spatial position identifier, constructing a topological network containing spatial adjacency relationship of geographic entities, and extracting spatial topological features of the topological network.

[0048] In order to analyze the connection and distribution characteristics of geographic entities in space, spatial clustering analysis processing is performed based on the spatial position identifier, and a topological network is constructed and spatial topological features are extracted.

[0049] Step S1251: analyzing the latitude and longitude coordinate information of the spatial position identifier of each geographic entity, and constructing a geographic entity coordinate set.

[0050] For each geographic entity observation record in the standardized geographic entity data set D", parse the longitude and latitude coordinate information in its corresponding spatial location identifier. Assume that the spatial location identifier is represented by a set P = {p1, p2, …, pn}, each pi contains longitude and latitude coordinates (xi, yi). Extract these longitude and latitude coordinates to construct a geographic entity coordinate set C = {(x1, y1), (x2, y2), …, (xn, yn)}, which contains the specific location information of all geographic entities in space.

[0051] Step S1252: Calculate the spatial distance matrix between each pair of geographic entity coordinates, dynamically adjust the adjacency threshold based on the distribution density of geographic entity coordinates, and establish the corresponding connection edge in the topological network if the spatial distance between two geographic entities is less than the dynamic adjacency threshold.

[0052] According to the geographic entity coordinate set C, the spatial distance between each pair of geographic entity coordinates is calculated. Common distance measurement methods such as Euclidean distance and Manhattan distance can be used. Assuming that Euclidean distance is used, for geographic entity coordinates (xi, yi) and (xj, yj), the spatial distance between them is dij = √((xi-xj)^2+(yi-yj)^2). Calculate the spatial distance between all pairs of geographic entities to construct a spatial distance matrix D = [dij], where i, j = 1, 2, …, n.

[0053] Based on the distribution density of geographic entity coordinates, the adjacency threshold is dynamically adjusted. The distribution density of geographic entity coordinates can be measured by calculating the number of other geographic entities within a certain range around each geographic entity. For areas with higher distribution density, the adjacency threshold is appropriately reduced; for areas with lower distribution density, the adjacency threshold is appropriately increased. Assuming that the adjacency threshold is thd, for each element dij in the spatial distance matrix D, if dij < thd, then a connection edge between geographic entities i and j is established in the topological network. In this way, a topological network containing the spatial adjacency relationship of geographic entities is constructed.

[0054] Step S1253: Perform weight assignment operation on the connection edge, the weight value is inversely proportional to the spatial distance, the smaller the spatial distance, the greater the weight value of the connection edge.

[0055] After the topological network is constructed, weight assignment operations are performed on the connection edges in the network. Since the connection between two geographic entities with smaller spatial distance is closer, the weight value is inversely proportional to the spatial distance. For a connection edge (i, j) in the topological network, the corresponding spatial distance is dij, and the weight value is wij. The weight assignment formula can be wij = 1 / dij (in order to avoid a zero denominator, a very small constant ε can be added to the denominator). In this way, each connection edge in the topological network is assigned a corresponding weight, reflecting the spatial connection strength between geographic entities.

[0056] Step S1254: All geographic entity nodes are traversed to construct a weighted connection edge-free topological network, and the degree centrality index of each node is calculated to represent the spatial connection density.

[0057] All geographic entity nodes in the geographic entity coordinate set C are traversed, and the previously established connection edges and weight assignments are combined to construct a weighted connection edge-free topological network G=(V, E, W), where V is the node set, i.e., the geographic entity set; E is the connection edge set; and W is the weight set of the connection edges.

[0058] The degree centrality index of each node is calculated, which is used to represent the spatial connection density of the node in the topological network. For node i, its degree centrality index Ci can be obtained by calculating the sum of the weights of all connection edges connected to it, i.e., Ci = ∑wij, where j is all nodes connected to node i. The greater the degree centrality index, the closer the connection in the topological network, and the higher the spatial connection density.

[0059] Step S1255: A community detection algorithm is performed on the topological network to identify a high-cohesion geographic entity sub-group, and the spatial coverage range and internal connection strength index of each sub-group are extracted.

[0060] In order to further analyze the structure of the topological network, a community detection algorithm such as the Louvain algorithm is performed on the topological network G. The Louvain algorithm divides the topological network into multiple high-cohesion geographic entity sub-groups by iteratively optimizing the modularity. The nodes within each sub-group are closely connected, while the connections between sub-groups are relatively weak.

[0061] For each identified geographic entity sub-group, its spatial coverage range and internal connection strength index are extracted. The spatial coverage range can be determined by calculating the minimum bounding rectangle or convex hull of all geographic entity coordinates within the sub-group. The internal connection strength index can be measured by calculating the sum of the weights of all connection edges within the sub-group. The greater the sum of the weights of the connection edges, the closer the connection within the sub-group.

[0062] Step S1256: generating a spatial topology feature vector based on the degree centrality index and the sub-group features, the feature vector including node connection density, sub-group coverage radius and cross-group connection strength.

[0063] In this embodiment, the spatial topology feature vector is generated based on the degree centrality index and the sub-group features obtained in the foregoing. First, for the node connection density, it is embodied by the degree centrality index of each node calculated in the foregoing. The degree centrality index reflects the connection tightness of the node with other nodes in the topology network, that is, the spatial connection density of the node. The degree centrality indexes of all nodes are arranged in a set order to form a feature sequence about the node connection density.

[0064] For the sub-group coverage radius, after the geographical entity sub-groups are identified in the foregoing, the spatial coverage range of each sub-group has been calculated. The sub-group coverage radius can be obtained by calculating a certain characteristic distance of the spatial coverage range of the sub-group. For example, if the spatial coverage range is determined by the minimum circumscribed rectangle, half of the length of the diagonal of the rectangle can be taken as the sub-group coverage radius; if determined by the convex hull, the center of the convex hull can be taken as the reference to calculate the distance to the farthest point on the boundary of the convex hull as the sub-group coverage radius. The coverage radii of all sub-groups are arranged in a certain numbering order of the sub-groups to form a feature sequence about the sub-group coverage radius.

[0065] The cross-group connection strength measures the connection tightness between different sub-groups. It can be obtained by calculating the sum of the weights of the connection edges connecting the nodes of different sub-groups. For each pair of sub-groups, the total weight sum of the connection edges between them is counted, and then these cross-group connection strength values are arranged in a certain arrangement order of the sub-group pairs to form a feature sequence.

[0066] The node connection density feature sequence, the sub-group coverage radius feature sequence and the cross-group connection strength feature sequence are spliced to generate the spatial topology feature vector. The spatial topology feature vector contains multiple dimensions of information, which reflects the topology structure and connection characteristics of the geographical entity in space from different aspects.

[0067] Step S126: performing feature alignment processing on the time sequence correlation features and the spatial topology features, generating a feature dimension consistent with the number of spatial sub-regions by aggregating the time sequence correlation features of all time window sub-sequences in the spatial sub-region, so that the mapping relationship between the time window sub-sequences and the spatial sub-regions is consistent, and combining the aligned features into a multi-dimensional feature matrix.

[0068] After obtaining the time series correlation features and spatial topology features, in order to facilitate subsequent analysis and processing, the two kinds of features need to be aligned. First, determine the division of the spatial sub-regions. The spatial sub-regions can be divided according to the geographical entity sub-groups obtained from the previous spatial clustering analysis, and each sub-group corresponds to a spatial sub-region.

[0069] For each spatial sub-region, aggregate the time series correlation features of all time window sub-sequences in the region. Suppose there are multiple time window sub-sequences in the spatial sub-region A, each time window sub-sequence has corresponding time series correlation features, such as the slope change feature of the long-term trend component and the amplitude fluctuation feature of the seasonal period component. Perform aggregation operation on these time series correlation features to obtain a comprehensive time series correlation feature corresponding to the spatial sub-region.

[0070] In this way, a comprehensive time series correlation feature is generated for each spatial sub-region, so that the generated feature dimension is consistent with the number of spatial sub-regions. This ensures that the mapping relationship between the time window sub-sequences and the spatial sub-regions is consistent, that is, each spatial sub-region has a corresponding comprehensive time series correlation feature.

[0071] Combine the aligned time series correlation features and spatial topology features into a multi-dimensional feature matrix. The comprehensive time series correlation feature and the corresponding spatial topology feature vector of each spatial sub-region can be arranged in a set order to form a row of the multi-dimensional feature matrix. For example, the first row can be the splicing of the comprehensive time series correlation feature and the spatial topology feature vector of the first spatial sub-region, the second row is the splicing of the corresponding features of the second spatial sub-region, and so on. The multi-dimensional feature matrix integrates the time series and spatial feature information of the geographical entity observation records.

[0072] Step S130: calling a pre-trained geographical space analysis model to perform dynamic feature fusion processing on the time series correlation features and the spatial topology features to generate a fused feature set.

[0073] In order to fully exploit the potential relationship between the time series correlation features and the spatial topology features, a pre-trained geographical space analysis model is called to perform dynamic feature fusion processing on the two kinds of features. The geographical space analysis model is trained on a large amount of data and can effectively learn the complex relationship between features.

[0074] Step S131: inputting the time series correlation features into the time series feature encoder of the geographical space analysis model, extracting local time series patterns of different time scales through time convolution layers, and capturing global dependency relationships across time windows using a multi-head attention mechanism to generate a time series encoding feature vector.

[0075] The time-series correlation features obtained above are input into the time-series feature encoder of the geographic spatial analysis model. The time-series feature encoder mainly consists of a time convolution layer and a multi-head attention mechanism. The time convolution layer performs sliding convolution operations on the time-series correlation features using convolution kernels of different sizes to extract local time-series patterns at different time scales. For example, using smaller convolution kernels can extract short-term time-series change patterns, while using larger convolution kernels can capture time-series trends over a longer time range.

[0076] The multi-head attention mechanism is used to capture global dependencies across time windows. It divides the time-series correlation features into multiple sub-features, and each sub-feature is processed by a different attention head. Each attention head calculates attention scores between sub-features, which reflect the degree of correlation between features at different time points. By weighted summing these attention scores, an attention representation is obtained for each sub-feature. Finally, the attention representations of all sub-features are concatenated to obtain a representation of global dependencies.

[0077] After processing by the time convolution layer and the multi-head attention mechanism, a time-series encoding feature vector is generated. This time-series encoding feature vector contains information about local patterns at different time scales and global dependencies across time windows.

[0078] Step S132: Input the spatial topology features into the spatial feature encoder of the geographic spatial analysis model, aggregate the topology features of adjacent nodes through a graph convolution layer, and generate a spatial encoding feature vector using a spatial pooling layer.

[0079] The spatial topology features are input into the spatial feature encoder of the geographic spatial analysis model. The spatial feature encoder mainly consists of a graph convolution layer and a spatial pooling layer. The graph convolution layer aggregates the topology features of adjacent nodes based on the topology network constructed earlier, which contains the spatial adjacency relationships of geographic entities. The features of each node are influenced by the features of its adjacent nodes, and through graph convolution operations, the feature information of adjacent nodes is transmitted to the current node to update the feature representation of the current node.

[0080] The spatial pooling layer is used to compress the feature dimension. Since the spatial topology features may have a high dimension, in order to reduce the computational load and improve the efficiency of the model, dimension reduction processing is needed. The spatial pooling layer can use methods such as max pooling and average pooling to sample and compress the features, retaining important feature information.

[0081] After processing by the graph convolution layer and the spatial pooling layer, a spatial encoding feature vector is generated. This spatial encoding feature vector integrates the spatial topology structure information of geographic entities and has undergone dimension compression, facilitating subsequent feature fusion operations.

[0082] Step S133: Perform feature alignment operation on the time-series encoded feature vector and the spatial encoded feature vector, adjust the time dimension of the time-series encoded feature vector and the spatial dimension of the spatial encoded feature vector to the same order of magnitude by spatial sub-region aggregation or time interpolation method, and fill in the null values of the missing feature positions.

[0083] Since the dimensions of the time-series encoded feature vector and the spatial encoded feature vector may not be consistent, feature alignment operation is needed. Spatial sub-region aggregation or time interpolation method can be used to adjust the dimensions. If the time dimension of the time-series encoded feature vector and the spatial dimension of the spatial encoded feature vector differ greatly, spatial sub-region aggregation can be used to group the time-series encoded feature vector according to spatial sub-region, and then aggregate the features in each group, such as taking the average, sum, etc., to make the dimensions of the time-series encoded feature vector close to the dimensions of the spatial encoded feature vector.

[0084] If the difference between the time dimension and the spatial dimension is caused by uneven time sampling, time interpolation method can be used. Time interpolation method adjusts the time dimension of the time-series encoded feature vector to the same order of magnitude as the spatial dimension by interpolation. During the feature alignment process, there may be missing feature positions. For these missing positions, appropriate methods can be used for filling, such as using the average, median, etc. of adjacent feature values for filling.

[0085] Through feature alignment operation, the time-series encoded feature vector and the spatial encoded feature vector are matched in dimension.

[0086] Step S134: Construct a cross-modal attention mechanism, calculate the correlation matrix between the time-series encoded feature vector and the spatial encoded feature vector, and generate dynamic weight coefficients based on the correlation matrix.

[0087] Step S1341: Map the time-series encoded feature vector and the spatial encoded feature vector to a high-dimensional hidden space respectively to generate a time-series query vector and a spatial key vector.

[0088] In order to better calculate the correlation between the time-series encoded feature vector and the spatial encoded feature vector, they are first mapped to a high-dimensional hidden space respectively. The time-series encoded feature vector is linearly transformed by a fully connected layer to map it to a high-dimensional hidden space, obtaining a time-series query vector. Similarly, the spatial encoded feature vector is linearly transformed to obtain a spatial key vector. This mapping operation can convert the original feature vector to a space more suitable for calculating correlation, so that features of different modalities can be compared and correlated in the same space.

[0089] Step S1342: Calculate the dot product similarity between the time-series query vector and the spatial key vector to generate an initial attention score matrix.

[0090] After obtaining the time sequence query vector and the spatial key vector, the dot product similarity between them is calculated. For each element in the time sequence query vector and each element in the spatial key vector, the dot product is calculated, and all the dot product results are combined to form an initial attention score matrix. The dot product similarity reflects the degree of similarity between the time sequence features and the spatial features, and the higher the score indicates the closer the association between the two features.

[0091] Step S1343: Standardizing the initial attention score matrix, using the Softmax function to normalize the scores into a probability distribution, generating an attention weight matrix.

[0092] Since the score values in the initial attention score matrix may be large, in order to facilitate subsequent weighted calculation, it needs to be standardized. Using the Softmax function to convert each score value in the initial attention score matrix to a probability value, so that the sum of all elements in the matrix is 1. After the Softmax function processing, the attention weight matrix is obtained. Each element in the attention weight matrix represents the association probability between the time sequence features and the spatial features, and the larger the weight indicates the stronger the association.

[0093] Step S1344: Weighted aggregation of the spatial encoding feature vector according to the attention weight matrix, generating a spatial attention feature vector.

[0094] According to the attention weight matrix, the spatial encoding feature vector is weighted and aggregated. For each element in the spatial encoding feature vector, multiply the corresponding weight value in the attention weight matrix, and then fuse all the weighted elements to obtain the spatial attention feature vector. The spatial attention feature vector integrates the association information between the spatial encoding feature vector and the time sequence encoding feature vector, highlighting the spatial features with strong association with the time sequence features.

[0095] Step S1345: Concatenate the spatial attention feature vector with the time sequence encoding feature vector, and fuse the cross-modal information through a fully connected layer to generate a joint feature representation.

[0096] The spatial attention feature vector and the time sequence encoding feature vector are concatenated, and the two vectors are connected together in a set order to form a new vector. Then, the new vector is input into a fully connected layer. The fully connected layer processes the concatenated vector through a series of linear transformations and nonlinear activation functions, fusing cross-modal information. After the processing of the fully connected layer, a joint feature representation is generated, which integrates the feature information of the time sequence and spatial modalities.

[0097] Step S1346: Perform layer normalization on the joint feature representation to eliminate the scale difference between features, and use residual connection to retain the original feature information to generate the final dynamic weight coefficient.

[0098] Perform layer normalization on the joint feature representation. Layer normalization can standardize each element in the joint feature representation, eliminating the scale difference between features and allowing features to be compared and processed at the same scale. At the same time, the information of the original time encoding feature vector and the spatial encoding feature vector is retained through residual connection. Residual connection adds the original feature vector to the processed feature vector, which can avoid losing important original information during feature fusion. After layer normalization and residual connection processing, the final dynamic weight coefficient is generated. These coefficients reflect the importance of the time encoding feature vector and the spatial encoding feature vector in the feature fusion process, and will be dynamically adjusted according to different input data.

[0099] Step S135: Perform weighted fusion on the time encoding feature vector and the spatial encoding feature vector according to the dynamic weight coefficient to generate a preliminary fusion feature matrix.

[0100] According to the dynamic weight coefficient generated in the previous step, perform weighted fusion operation on the time encoding feature vector and the spatial encoding feature vector. For each element in the time encoding feature vector, multiply it by the corresponding dynamic weight coefficient; similarly, multiply each element in the spatial encoding feature vector by the corresponding dynamic weight coefficient. Then, concatenate the weighted time encoding feature vector and the spatial encoding feature vector to form a preliminary fusion feature matrix. The preliminary fusion feature matrix integrates time and spatial feature information and highlights the importance of different features according to the dynamic weight coefficient.

[0101] Step S136: Perform feature enhancement operation on the preliminary fusion feature matrix to enhance the interaction between features through a nonlinear transformation layer and remove redundant noise using a denoising autoencoder to generate an optimized fusion feature set.

[0102] In order to further improve the quality of the preliminary fusion feature matrix, perform feature enhancement operation on it. First, process the preliminary fusion feature matrix through a nonlinear transformation layer. The nonlinear transformation layer can use activation functions such as ReLU, Sigmoid, etc. to perform nonlinear transformation on each element in the feature matrix. This nonlinear transformation can enhance the interaction between features and mine the potential relationship between features.

[0103] Then, the preliminary fusion feature matrix is removed of redundant noise by using a denoising autoencoder. The denoising autoencoder is an unsupervised learning model that removes noise by adding noise to the input data and then learning to reconstruct the original data. The preliminary fusion feature matrix is input into the denoising autoencoder, and after processing by the encoder and decoder, a feature matrix removed of noise is obtained. After processing by the nonlinear transformation layer and the denoising autoencoder, an optimized post-fusion feature set is generated, which contains richer and more valuable geospatial feature information.

[0104] The training process of the above geospatial analysis model can be: using a large set of geospatial data, the time series correlation features and spatial topology features obtained after feature extraction are used as training data, and the training data is divided into a training set, a validation set and a test set.

[0105] The convolution kernel weights of the time convolution layer, the parameters of the multi-head attention mechanism, the weights of the graph convolution layer, and the related parameters of the spatial pooling layer are initialized. The training data is input into the geospatial analysis model, sequentially passes through the time series feature encoder and the spatial feature encoder, obtains the time series encoding feature vector and the spatial encoding feature vector, and then performs feature alignment and cross-modal attention fusion to output a preliminary fusion feature matrix. In this process, a suitable loss function, such as a mean square error loss function, is defined to calculate the loss value between the preliminary fusion feature matrix and the true label. According to the loss value, an optimization algorithm (such as the stochastic gradient descent algorithm) is used to update the parameters of the geospatial analysis model, so that the loss value gradually decreases. The performance of the model is evaluated on the validation set, and the parameters and structure of the geospatial analysis model are adjusted according to the validation results, such as adjusting the size of the convolution kernel and the number of attention heads. The trained geospatial analysis model is finally tested using the test set to evaluate the generalization ability of the geospatial analysis model.

[0106] Step S140: generating a geospatial optimization strategy based on the post-fusion feature set, the geospatial optimization strategy containing geospatial entity state prediction results and resource allocation priority configuration.

[0107] After obtaining the post-fusion feature set, geospatial optimization strategies are generated using these feature information, which aims to predict the state of geospatial entities and reasonably configure the resource allocation priority.

[0108] Step S141: inputting the post-fusion feature set into a geospatial entity state prediction model to calculate the state change probability of the geospatial entity in the future time window through a multi-layer perception network, and generating a probability distribution prediction result.

[0109] The fused feature set is input into a geographic entity state prediction model, which mainly consists of a multi-layer perceptron network. The multi-layer perceptron network is a kind of feedforward neural network, which consists of an input layer, a hidden layer and an output layer. The fused feature set is input into the input layer, and after a series of nonlinear transformation and weighted summation operations in the hidden layer, the state change probability of the geographic entity in the future time window is obtained in the output layer.

[0110] The neurons in the hidden layer perform nonlinear transformation on the input signal through an activation function, increasing the expression capacity of the model. Each neuron in the output layer corresponds to a state category of a geographic entity, and its output value represents the probability of the geographic entity being in this state category. The state change probabilities of all geographic entities are arranged in a set order to form a probability distribution prediction result. The probability distribution prediction result reflects the possible state changes of the geographic entity in the future time window.

[0111] In the training process of the geographic entity state prediction model, the geographic entity state prediction model mainly consists of a multi-layer perceptron network, including an input layer, multiple hidden layers and an output layer. The input layer receives the fused feature set, the hidden layer transforms and combines the input features through a nonlinear activation function (such as the ReLU function), and the output layer outputs the state change probability distribution of the geographic entity in the future time window. The layers are connected through full connection, i.e. each neuron is connected to all neurons in the previous layer. In detail, the fused feature set can be used as input data, and the corresponding real state of the geographic entity can be used as label data, which is also divided into training set, validation set and test set. After initializing the weights and biases of each layer of the multi-layer perceptron network, the training data is input into the model, and is sequentially passed through the input layer, the hidden layer and the output layer to obtain the predicted state change probability distribution. Then, the cross-entropy loss function is used to calculate the loss value between the predicted probability distribution and the real label, and the optimization algorithm (such as the Adam optimization algorithm) is used to update the parameters of the geographic entity state prediction model according to the loss value. The performance of the geographic entity state prediction model is evaluated on the validation set, and the number of neurons in the hidden layer, the learning rate and other parameters are adjusted. Finally, the test set is tested to evaluate the prediction accuracy of the geographic entity state prediction model.

[0112] Step S142: identifying a risk geographic entity according to the probability distribution prediction result, calculating a resource demand urgency score of the risk geographic entity, and generating a priority list based on the sorting of the resource demand urgency score.

[0113] According to the prediction result of the probability distribution, a geographical entity that may have a risk is identified. For example, if the probability of a certain geographical entity being in a dangerous state exceeds a preset threshold, it is marked as a risk geographical entity. For the identified risk geographical entity, a resource demand urgency score is calculated. The resource demand urgency score can comprehensively consider multiple factors, such as the current state of the geographical entity, the predicted state change trend, the influence on the surrounding environment, and the like.

[0114] A weight can be set for each factor, and then the score of each factor is multiplied by the corresponding weight and summed to obtain the resource demand urgency score. For example, the current state score of the geographical entity can be determined according to the deviation of its current attribute value from the normal state value; the predicted state change trend score can be determined according to the probability of the state change of the geographical entity in the probability distribution prediction result. After the resource demand urgency scores of all risk geographical entities are calculated, they are sorted in descending order of the scores to generate a priority list. The geographical entities in the priority list are arranged according to the urgency of resource demand, and the higher the score of a geographical entity, the more it needs to be allocated resources in priority.

[0115] Step S143: A multi-objective optimization model is constructed, with the maximum resource allocation efficiency and the minimum geographical entity protection cost as objective functions, and constraint conditions are introduced to limit the total amount of resources, the allocation time window, and the resource allocation order set based on the priority list.

[0116] In order to allocate resources reasonably, a multi-objective optimization model is constructed, which has two objective functions, one is to maximize the resource allocation efficiency, and the other is to minimize the geographical entity protection cost. The resource allocation efficiency can be measured by the effective utilization rate of resources, such as whether the resources can be allocated to the geographical entities in need in time and accurately; the geographical entity protection cost can include the procurement cost, transportation cost, maintenance cost, and the like of resources.

[0117] Constraint conditions are introduced to limit the solution range of the model. First, the total amount of resources is limited to ensure that the allocated resources do not exceed the upper limit of available resources. Second, the allocation time window is set to specify that the resource allocation must be completed within the set time range. Finally, the resource allocation order is set according to the priority list generated in the foregoing to ensure that the geographical entities with high priority obtain resource allocation in priority.

[0118] Step S144: An evolutionary algorithm is used to solve the multi-objective optimization model, a Pareto optimal solution set is generated, and an optimal solution that meets a preset strategy is selected through an interactive decision interface.

[0119] Step S1441: The population initialization parameters, crossover probability parameters, and mutation probability parameters of the evolutionary algorithm are set.

[0120] Before solving the multi-objective optimization model using evolutionary algorithm, some key parameters need to be set. Population initialization parameters are used to determine the size of the initial population and the encoding method of individuals. The size of the initial population determines the coverage of the search space of the algorithm. A larger population can increase the likelihood of finding the optimal solution, but it will also increase the amount of calculation. The encoding method of individuals represents the resource allocation scheme in a form that the algorithm can handle, such as binary encoding, real number encoding, etc.

[0121] The crossover probability parameter controls the probability of crossover operation between individuals in the evolution process. Crossover operation is to exchange part of the encoding of two individuals to generate new individuals, thereby increasing the diversity of the population. The mutation probability parameter controls the probability of mutation of individuals in the evolution process. Mutation operation is a random modification of the encoding of individuals to prevent the algorithm from falling into a local optimal solution.

[0122] Step S1442: Generate an initial population based on the resource allocation priority configuration and the constraint conditions of the multi-objective optimization model, where each individual represents a resource allocation scheme. Use a constraint satisfaction algorithm to verify the legality of each individual. Remove individuals that violate the total resource constraint, allocation time window constraint, and resource allocation sequence constraint, and generate new individuals that meet the constraints.

[0123] An initial population is generated based on the resource allocation priority configuration and the constraint conditions of the multi-objective optimization model. Each individual represents a resource allocation scheme that allocates resources to different geographic entities. A constraint satisfaction algorithm is used to verify the legality of each individual in the initial population. Check whether each individual violates the total resource constraint, allocation time window constraint, and resource allocation sequence constraint.

[0124] If an individual violates any of the constraint conditions, it is removed from the population. At the same time, in order to maintain the size of the population unchanged, new individuals that meet the constraint conditions are generated. New individuals can be generated by random generation or mutation based on existing legal individuals.

[0125] Step S1443: Perform fitness evaluation on each individual in the initial population, and calculate the multi-objective fitness value based on the resource allocation efficiency maximization objective function and the geographic entity protection cost minimization objective function.

[0126] The fitness of each individual in the initial population is evaluated, and a multi-objective fitness value is calculated based on the resource allocation efficiency maximization objective function and the geographical entity protection cost minimization objective function. The fitness value reflects the degree of excellence of each individual in the multi-objective optimization problem. For each resource allocation scheme represented by an individual, its resource allocation efficiency and geographical entity protection cost are calculated respectively. The resource allocation efficiency can be measured by calculating the effective benefits that the resources allocated to each geographical entity can generate, such as the degree of improvement in the state of the geographical entity; the geographical entity protection cost takes into account the costs of resource acquisition, transportation, maintenance, etc.

[0127] When calculating the multi-objective fitness value, the resource allocation efficiency and the geographical entity protection cost need to be considered comprehensively. A weighted sum method can be used to set weights for the resource allocation efficiency and the geographical entity protection cost, then multiply the scores of the two by the corresponding weights and add them together to get the multi-objective fitness value of the individual. The weights are determined according to the specific application scenario and requirements. If more emphasis is placed on resource allocation efficiency, the weight of resource allocation efficiency can be set higher; if more attention is paid to geographical entity protection cost, the weight of geographical entity protection cost can be set higher.

[0128] Step S1444: Perform genetic operations on the initial population, including selection operation based on fitness value, crossover operation based on crossover probability parameter, and mutation operation based on mutation probability parameter, to generate offspring population.

[0129] The selection operation is based on the fitness value of the individual. The higher the fitness value, the greater the probability of being selected into the next generation. Roulette selection, tournament selection and other methods can be used for selection. The roulette selection method is to take the fitness value of each individual as the sector area it occupies on the roulette, the higher the fitness value, the larger the sector area, and by randomly rotating the roulette, the individual is selected according to the sector pointed by the pointer. The tournament selection method is to randomly select a certain number of individuals from the population, compare their fitness values, and select the individual with the highest fitness value into the next generation.

[0130] The crossover operation is based on the crossover probability parameter. According to the pre-set crossover probability, two individuals are randomly selected from the individuals obtained by the selection operation as parents, and then the two parent individuals are subjected to crossover operation. The crossover operation can use single-point crossover, multi-point crossover and other methods. Single-point crossover randomly selects a crossover point in the coding sequence of the individual, and exchanges the parts after the crossover point of the two parent individuals to generate two offspring individuals. Multi-point crossover selects multiple crossover points and exchanges the parts between the crossover points.

[0131] The mutation operation is based on a mutation probability parameter. For each offspring individual, certain positions in its encoding sequence are randomly selected for mutation according to the mutation probability. The mutation operation can be to change a value in the encoding sequence, for example, to change a 0 to 1 or 1 to 0 in a binary encoding, or to make a small adjustment to a value in a real number encoding. Through the mutation operation, the diversity of the population can be increased, and the algorithm can be prevented from falling into a local optimal solution.

[0132] After the selection, crossover, and mutation operations, an offspring population is generated. The individuals in the offspring population are generated from the individuals in the initial population through genetic operations, and they inherit some characteristics of the parent individuals while possibly introducing new characteristics.

[0133] Step S1445: The initial population and the offspring population are merged, a non-dominated sorting process is performed to determine the Pareto front ranks of the individuals, and the crowding distance is calculated to screen diverse individuals, to generate a new generation population.

[0134] The initial population and the offspring population are merged to form a larger population. The merged population is subjected to a non-dominated sorting process. Non-dominated sorting is a sorting method for multi-objective optimization problems, which divides the individuals in the population into different Pareto front ranks. If an individual is not worse than another individual in all objective functions and is better than the other individual in at least one objective function, the former individual is said to dominate the latter individual. By constantly finding non-dominated individuals, they are divided into the first Pareto front rank, then from the remaining individuals, non-dominated individuals are continuously found and divided into the second Pareto front rank, and so on, until all individuals are divided into different Pareto front ranks.

[0135] The crowding distance of each individual is calculated. The crowding distance is used to measure the crowding degree of an individual in its Pareto front rank. For the individuals in each Pareto front rank, the individuals are sorted according to the value of each objective function, and then the sum of the distances between the boundary individual and the adjacent individual of each objective function of each individual is calculated as the crowding distance of the individual. The larger the crowding distance, the less crowded the individual is in its Pareto front rank, and the better the diversity.

[0136] According to the Pareto front ranks and the crowding distances, the individuals are screened to generate a new generation population. Individuals with lower Pareto front ranks are preferentially selected because these individuals are more optimal in multi-objective optimization problems. In the same Pareto front rank, individuals with larger crowding distances are preferentially selected to ensure the diversity of the population. In this way, a certain number of individuals are selected from the merged population to form a new generation population.

[0137] Step S1446: iteratively perform the above steps until an evolutionary termination condition is reached, and output a set of Pareto optimal solutions.

[0138] The steps of selection, crossover, mutation, merging of population, non-dominated sorting, and screening of individuals are iteratively performed to constantly update the population. The evolutionary termination condition can be reaching a preset maximum number of iterations, or the fitness value of the population no longer being significantly improved. When the evolutionary termination condition is met, the iteration is stopped, and a set of Pareto optimal solutions in the current population is output. Each solution in the set of Pareto optimal solutions represents a resource allocation scheme that achieves a balance between resource allocation efficiency and geographic entity protection cost, and there is no solution that is superior to another solution in all objectives.

[0139] Step S1447: visualize the set of Pareto optimal solutions in an interactive decision interface, receive user input selection instructions, and determine the final optimal solution based on a preset strategy matching degree.

[0140] The set of Pareto optimal solutions is visualized in an interactive decision interface. Two-dimensional or three-dimensional graphics can be used to represent the values of each solution in resource allocation efficiency and geographic entity protection cost as points, forming a Pareto frontier. Users can intuitively observe the relationship and advantages and disadvantages between different solutions on the interface.

[0141] Receive user input selection instructions. Users can select one or more solutions from the set of Pareto optimal solutions according to their needs and preferences. For example, users may be more concerned about resource allocation efficiency and want to select a solution with higher resource allocation efficiency; or users may be more concerned about the cost of geographic entity protection and want to select a solution with lower cost.

[0142] Determine the final optimal solution based on a preset strategy matching degree. The preset strategy can be some rules or preferences set in advance according to the actual application scenario and needs. For example, the preset strategy can specify the weight ratio of resource allocation efficiency and geographic entity protection cost, or set a threshold for certain objectives. According to the user input selection instructions and the preset strategy, calculate the matching degree of each candidate solution with the preset strategy, and select the solution with the highest matching degree as the final optimal solution.

[0143] Step S145: integrate the resource allocation scheme corresponding to the optimal solution with the geographic entity state prediction result to generate a geospatial optimization strategy containing specific execution steps and a time schedule.

[0144] The final optimal solution corresponding to the resource allocation scheme is integrated with the geographical entity state prediction results. The resource allocation scheme specifies which geographical entities to allocate resources to and the amount of allocation, while the geographical entity state prediction results provide the state changes of the geographical entities in the future time window. Combining these two aspects of information, specific execution steps and timelines are developed.

[0145] The specific execution steps include resource acquisition, transportation, allocation, and other operations, as well as specific handling measures for different geographical entities. For example, if a certain geographical entity is predicted to have a certain risk state, the execution steps can include monitoring and taking protective measures for that geographical entity. The timeline specifies the start and end times of each execution step, ensuring that resource allocation and geographical entity protection work can be carried out in an orderly manner.

[0146] By integrating the resource allocation scheme and the geographical entity state prediction results, a geographical space optimization strategy containing specific execution steps and timelines is generated. This geographical space optimization strategy provides a detailed guidance scheme for the management and optimization of geographical space.

[0147] Step S146: Verify the feasibility of the geographical space optimization strategy, simulate the state changes of the geographical entities after the execution of the geographical space optimization strategy, and if the simulation results exceed the preset risk threshold, dynamically adjust the objective function weights or constraint condition relaxation range of the multi-objective optimization model according to the risk type, and regenerate the geographical space optimization strategy.

[0148] The feasibility of the generated geographical space optimization strategy is verified. Simulation methods can be used to simulate the state changes of the geographical entities after the execution of the geographical space optimization strategy based on the current state of the geographical entities, resource allocation scheme and execution steps. Various factors are considered in the simulation process, such as the effect of resources, changes in environmental factors, etc.

[0149] A preset risk threshold is set to measure the risk level of the state changes of the geographical entities. The preset risk threshold can be set according to different geographical entities and application scenarios, for example, for some important geographical entities, a lower risk threshold can be set. If the simulation results exceed the preset risk threshold, it means that the current geographical space optimization strategy may have risks and needs to be adjusted.

[0150] The objective function weights or constraint condition relaxation range of the multi-objective optimization model are dynamically adjusted according to the risk type. If the risk type is that the resource allocation efficiency is too low, the weight of the resource allocation efficiency objective function can be appropriately increased, or the resource total amount constraint condition can be relaxed; if the risk type is that the geographical entity protection cost is too high, the weight of the geographical entity protection cost objective function can be appropriately increased, or the resource allocation time window constraint condition can be tightened.

[0151] Regenerate the geospatial optimization strategy. According to the adjusted multi-objective optimization model, the evolutionary algorithm is used again to solve, and a new set of Pareto optimal solutions is obtained. Then, according to the previous steps, the final optimal solution is determined, the resource allocation scheme and the geospatial entity state prediction result are integrated, and a new geospatial optimization strategy is generated. Through continuous feasibility verification and adjustment, the effectiveness and reliability of the geospatial optimization strategy are ensured.

[0152] Step S150: Trigger the geographic information system update operation of the target area according to the geospatial optimization strategy.

[0153] After generating the geospatial optimization strategy, it is necessary to trigger the geographic information system update operation of the target area according to the geospatial optimization strategy, so as to ensure that the data and information in the geographic information system can reflect the latest geospatial state and resource allocation situation.

[0154] Step S151: Analyze the specific execution steps and time table in the geospatial optimization strategy, extract the resource allocation scheme parameters matched with the geospatial entity state prediction result and the spatial position identifier of the target geospatial entity.

[0155] The specific execution steps and time table in the geospatial optimization strategy are analyzed. The specific execution steps detail the allocation and processing process of resources, and the time table specifies the time arrangement of each step. From these information, the resource allocation scheme parameters matched with the geospatial entity state prediction result are extracted, including the number of resource allocation, the object of allocation, etc. At the same time, the spatial position identifier of the target geospatial entity is extracted, which is used to determine the specific position of the geospatial entity in the geospatial space.

[0156] For example, if the geospatial optimization strategy specifies that a certain amount of resources is allocated to a certain geospatial entity at a certain time point, the spatial position identifier of the geospatial entity and the number of allocated resources are extracted as parameters in the analysis process.

[0157] Step S152: Generate incremental update instructions for the geographic information system according to the resource allocation scheme parameters, and the incremental update instructions contain the spatial coordinate sequence of the resource allocation path, the time stamp constraint condition and the corresponding geospatial entity attribute modification value.

[0158] According to the extracted resource allocation scheme parameters, the incremental update instructions for the geographic information system are generated. The incremental update instructions are specific instructions for updating the data in the geographic information system, which contain the spatial coordinate sequence of the resource allocation path, the time stamp constraint condition and the corresponding geospatial entity attribute modification value.

[0159] The spatial coordinate sequence of the resource allocation path describes the path of the resource from the acquisition point to the allocation point, represented by a series of spatial coordinates. The timestamp constraint specifies the time range for the resource allocation operation, ensuring that the resource allocation is completed within the specified time. The corresponding geographical entity attribute modification value refers to the changes in the attributes of the geographical entity after the resource allocation, such as the modification value of the state, quantity, and other attributes of the geographical entity.

[0160] For example, if the resource allocation is to deliver supplies to a certain geographical entity, the incremental update instruction will include the path coordinates of the supply transportation, the time range of the transportation, and the modification value of the attributes of the geographical entity after receiving the supplies.

[0161] Step S153: Spatiotemporal conflict detection on the incremental update instruction, verifying whether the resource allocation path overlaps with the spatial coordinate sequence of existing tasks in the geographic information system under the timestamp constraint, and dynamically adjusting the priority order of the resource allocation path based on the conflict detection result.

[0162] Spatiotemporal conflict detection is performed on the generated incremental update instruction. In the geographic information system, there may already be some ongoing tasks, which also have corresponding spatial coordinate sequences and time ranges. It is necessary to verify whether the resource allocation path overlaps with the spatial coordinate sequence of existing tasks under the timestamp constraint.

[0163] The conflict can be determined by comparing the spatial coordinate sequence of the resource allocation path and the spatial coordinate sequence of the existing tasks, as well as their time ranges. If there is a conflict, the priority order of the resource allocation path needs to be dynamically adjusted based on the conflict detection result. For resource allocation paths with higher priority, they can be executed first; for resource allocation paths with lower priority, their time range or path can be adjusted to avoid conflicts.

[0164] For example, if the resource allocation path overlaps with the path of an existing task within a certain time period, and the priority of the existing task is higher, the time or path of the resource allocation path needs to be adjusted to ensure that no conflict occurs.

[0165] Step S154: Split the adjusted incremental update instruction into multiple atomic transaction operation units according to the time table, each atomic transaction operation unit binds the corresponding geographical entity spatial location identifier and resource allocation parameter verification rule, and adds a version lock identifier to each atomic transaction operation unit, so that concurrent operations on the same geographical entity are executed in priority order.

[0166] The adjusted incremental update instruction is split into multiple atomic transaction operation units according to the time table. The atomic transaction operation unit is an indivisible operation unit, and each unit corresponds to a specific resource allocation operation. For example, allocating a certain amount of resources to a geographic entity can be an atomic transaction operation unit.

[0167] The corresponding geographic entity spatial position identifier and resource allocation parameter verification rule are bound to each atomic transaction operation unit. The geographic entity spatial position identifier is used to determine the geographic entity to which the operation unit is directed, and the resource allocation parameter verification rule is used to verify whether the resource allocation parameter of the operation unit is legal. For example, verifying whether the amount of resource allocation is within a reasonable range.

[0168] A version lock identifier is added to each atomic transaction operation unit. The version lock identifier is used to control concurrent operations on the same geographic entity. When multiple atomic transaction operation units simultaneously operate on a geographic entity, the operation is ensured to be executed in order according to the version lock identifier and priority order, avoiding data conflicts and inconsistencies.

[0169] Step S155: Trigger the asynchronous execution queue of the atomic transaction operation unit based on the timestamp constraint condition of the time table, and monitor the execution state of each atomic transaction operation unit in real time and capture resource allocation parameter verification failure events.

[0170] The asynchronous execution queue of the atomic transaction operation unit is triggered based on the timestamp constraint condition of the time table. The asynchronous execution queue can handle multiple atomic transaction operation units at the same time, improving execution efficiency. When the execution time of a certain atomic transaction operation unit is reached, it is added to the asynchronous execution queue for execution.

[0171] The execution state of each atomic transaction operation unit is monitored in real time. The execution state can be monitored by recording the start time, end time, execution result, etc. of the operation unit. At the same time, resource allocation parameter verification failure events are captured. If the resource allocation parameter of a certain atomic transaction operation unit does not meet the verification rule, a verification failure event will be triggered.

[0172] Step S156: According to the resource allocation parameter verification failure event, backtrack to the corresponding geographic entity spatial position identifier, regenerate the local incremental update instruction and insert it into the preset fault tolerance time window of the asynchronous execution queue.

[0173] When a resource allocation parameter verification failure event is captured, the corresponding geographic entity spatial position identifier is traced back according to the event. The reason for the verification failure is analyzed, which may be that the resource allocation parameter is unreasonable, the state of the geographic entity has changed, etc.

[0174] According to the analysis result, a local incremental update instruction is regenerated. The local incremental update instruction only updates the geographic entity that fails the verification, adjusts the resource allocation parameter to meet the verification rule. The regenerated local incremental update instruction is inserted into the preset fault tolerance time window of the asynchronous execution queue. The preset fault tolerance time window is a time range reserved for handling abnormal situations such as verification failure, and the operation can be re-executed in this time range to ensure the smooth update of the geographic information system.

[0175] Step S157: rendering the resource allocation heat map corresponding to the executed atomic transaction operation unit in the visualization layer of the geographic information system, superimposing the change trajectory of the geographic entity state prediction result, and generating a dynamically updated spatial decision view.

[0176] The resource allocation heat map is rendered in the visualization layer of the geographic information system. The resource allocation heat map represents the density and intensity of resource allocation through the depth of color. The deeper the color, the more resource allocation. Through the heat map, the allocation of resources in the geographic space can be observed intuitively.

[0177] Superimpose the change trajectory of the geographic entity state prediction result. The geographic entity state prediction result provides the state change of the geographic entity in the future time window. By superimposing the change trajectory on the resource allocation heat map, a more comprehensive understanding of the dynamic changes of the geographic space can be obtained.

[0178] Generate a dynamically updated spatial decision view. With the continuous execution of atomic transaction operation units and the change of geographic entity state, the spatial decision view will be updated in real time to provide the latest geographic space information for decision makers and help them make more reasonable decisions.

[0179] Step S158: reverse mapping the execution log of the atomic transaction operation unit and the geographic entity spatial position identifier, generating an incremental version snapshot and synchronizing it to the geographic space data copy of the edge node.

[0180] Reverse mapping the execution log of the atomic transaction operation unit and the geographic entity spatial position identifier. The execution log records the execution of each atomic transaction operation unit, including the time of the operation, the content of the operation, the execution result and other information. Through reverse mapping, the execution log can be associated with the corresponding geographic entity spatial position identifier, facilitating subsequent query and analysis.

[0181] Generate an incremental version snapshot. The incremental version snapshot records the update of the geographic information system at a certain time point, only containing incremental data related to the atomic transaction operation unit. By generating an incremental version snapshot, the data transmission and storage overhead can be reduced.

[0182] Synchronize the incremental version snapshot to the geospatial data replica of the edge node. The edge node is a distributed node in the geographic information system that stores a copy of the geospatial data. Synchronizing the incremental version snapshot to the edge node ensures that the data of the edge node is consistent with the data of the master node, improving the reliability and availability of the geographic information system.

[0183] Throughout the management of geospatial data, data authorization is an important link to ensure the legal use of data. When obtaining a set of geospatial data of a target area, explicit data authorization needs to be obtained from the data provider. Data providers may include government departments, research institutions, enterprises, etc.

[0184] For data authorization, a detailed authorization management mechanism needs to be established. First, the identity of the data provider needs to be verified and the qualifications need to be audited. Identity verification can be done by checking its registration information, relevant certificates, etc. Qualification audit needs to check whether it has the legal ability to collect and provide data, such as whether it has relevant surveying and mapping qualification certificates, etc.

[0185] When obtaining authorization, the scope of authorization needs to be clear, including the purpose of use, the term of use, the right of use, etc. The purpose of use should be strictly limited to the scope of geospatial data management and analysis, and cannot be used for other illegal or unauthorized purposes. The term of use needs to be set according to the actual needs and the requirements of the data provider, and after the term expires, if the data needs to be used continuously, the authorization needs to be renewed in time. The right of use needs to be clear whether the data can be modified, shared, distributed, etc.

[0186] For example, if land use data of a city is obtained from a government department, the authorization agreement clearly states that the data can only be used for geographic spatial planning analysis of the city, the term of use is one year, and the data cannot be shared with third parties. During use, these authorization clauses must be strictly followed.

[0187] At the same time, a record and audit mechanism of data authorization needs to be established. Record the detailed information of each data authorization, including authorization time, authorized person, authorization scope, etc. Regularly audit the use of data authorization and check whether there are violations of authorization clauses. If violations are found, measures need to be taken, such as suspending the right of data use, investigating legal responsibility, etc.

[0188] Data collection is the foundation of geospatial data management and must ensure its legality and fairness. When choosing data collection methods and equipment, relevant laws and regulations and industry standards need to be followed.

[0189] For the collection of satellite remote sensing images, it is necessary to ensure that the operation of the satellite complies with relevant international and domestic regulations. The launch and operation of the satellite need to go through strict approval and supervision. When using satellite remote sensing images, they should be obtained from legal channels and comply with the usage regulations of the data provider.

[0190] The use of ground surveying equipment also needs to comply with relevant standards. For example, total station and GPS receivers need to be quality tested and calibrated to ensure the accuracy and reliability of the collected data. During the collection process, the operators of the equipment should be trained and examined to ensure that they have professional operating skills and knowledge.

[0191] During data collection, personal privacy and sensitive information should be avoided. Geospatial data may involve some personal information, such as the information of the occupants of a building. Necessary technical means and management measures should be taken to filter and protect such sensitive information during collection.

[0192] For example, when using drones for aerial surveying, it is necessary to avoid taking pictures of the interior of private residences. If it is impossible to avoid some possible sensitive information, such information should be encrypted and access to it should be strictly controlled during data use.

[0193] At the same time, the fairness of data collection should be ensured. When selecting the collection area and collection time, it is necessary to avoid data bias caused by human factors. For example, it is not allowed to only select certain specific areas or certain times for data collection, while ignoring other areas and times. It is necessary to ensure that the data can fully and objectively reflect the geospatial characteristics of the target area.

[0194] Figure 2 The hardware structure of the artificial intelligence-based geospatial data management system 100 for implementing the above-mentioned artificial intelligence-based geospatial data management method is shown in FIG. 1. Figure 2 As shown in FIG. 1, the artificial intelligence-based geospatial data management system 100 can include a processor 110, a machine-readable storage medium 120, a bus 130, and a communication unit 140.

[0195] In one possible design, the artificial intelligence-based geospatial data management system 100 can be a single server or a group of servers. The group of servers can be centralized or distributed (e.g., the artificial intelligence-based geospatial data management system 100 can be a distributed system). In some embodiments, the artificial intelligence-based geospatial data management system 100 can be local or remote. For example, the artificial intelligence-based geospatial data management system 100 can access information and / or data stored in the machine-readable storage medium 120 via a network. For another example, the artificial intelligence-based geospatial data management system 100 can be directly connected to the machine-readable storage medium 120 to access the stored information and / or data.

[0196] The machine-readable storage medium 120 can store data and / or instructions. In some embodiments, the machine-readable storage medium 120 can store data acquired from an external terminal. In some embodiments, the machine-readable storage medium 120 can store data and / or instructions used by the artificial intelligence-based geospatial data management system 100 to perform or use to complete the exemplary methods described in the present disclosure.

[0197] In the implementation process, the one or more processors 110 execute the computer-executable instructions stored in the machine-readable storage medium 120, so that the processor 110 can perform the artificial intelligence-based geospatial data management method of the method embodiments as described above. The processor 110, the machine-readable storage medium 120, and the communication unit 140 are connected through the bus 130, and the processor 110 can be used to control the transceiving action of the communication unit 140.

[0198] The implementation process of the processor 110 can refer to the various method embodiments performed by the artificial intelligence-based geospatial data management system 100 as described above, which have similar implementation principles and technical effects, and will not be described here again.

[0199] In addition, the present embodiment also provides a readable storage medium, which is provided with computer-executable instructions. When the processor runs the computer-executable instructions, the artificial intelligence-based geospatial data management method as described above is realized.

[0200] It should be noted that, in order to simplify the description of the present disclosure and help understand one or more embodiments of the present disclosure, in the foregoing description of the embodiments of the present disclosure, various features are sometimes combined into one embodiment, figure or description thereof. Similarly, it should be noted that, in order to simplify the description of the present disclosure and help understand one or more embodiments of the present disclosure, in the foregoing description of the embodiments of the present disclosure, various features are sometimes combined into one embodiment, figure or description thereof.

Claims

1. An artificial intelligence-based geospatial data management method, characterized by, The method comprises: acquiring a geospatial data set of a target area, the geospatial data set comprising time-varying geospatial observation records and corresponding spatial location identifiers; performing feature extraction on the geospatial data set to generate time-series correlation features and spatial topology features of the geospatial observation records; calling a pre-trained geospatial analysis model to perform dynamic feature fusion processing on the time-series correlation features and the spatial topology features to generate a fused feature set; generating a geospatial optimization strategy based on the fused feature set, the geospatial optimization strategy comprising geospatial entity state prediction results and resource allocation priority configurations; triggering a geospatial information system update operation of the target area according to the geospatial optimization strategy; the calling of the pre-trained geospatial analysis model to perform dynamic feature fusion processing on the time-series correlation features and the spatial topology features to generate a fused feature set comprises: inputting the time-series correlation features into a time-series feature encoder of the geospatial analysis model, extracting local time-series patterns of different time scales through a time convolution layer, and capturing global dependency relationships across time windows using a multi-head attention mechanism to generate time-series encoding feature vectors; inputting the spatial topology features into a spatial feature encoder of the geospatial analysis model, aggregating topology features of adjacent nodes through a graph convolution layer, and compressing feature dimensions using a spatial pooling layer to generate spatial encoding feature vectors; performing feature alignment operations on the time-series encoding feature vectors and the spatial encoding feature vectors, adjusting the time dimension of the time-series encoding feature vectors and the spatial dimension of the spatial encoding feature vectors to the same order of magnitude through spatial sub-region aggregation or time interpolation method, and filling in null values at feature missing positions; constructing a cross-modal attention mechanism to calculate a correlation matrix between the time-series encoding feature vectors and the spatial encoding feature vectors, and generating dynamic weight coefficients based on the correlation matrix; performing weighted fusion of the time-series encoding feature vectors and the spatial encoding feature vectors according to the dynamic weight coefficients to generate a preliminary fused feature matrix; performing feature enhancement operations on the preliminary fused feature matrix, enhancing the interaction between features through a nonlinear transformation layer, and removing redundant noise using a denoising autoencoder to generate an optimized fused feature set.

2. The method of claim 1, wherein, the feature extraction on the geospatial data set to generate time-series correlation features and spatial topology features of the geospatial observation records comprises: performing data integrity check operations on the geospatial observation records in the geospatial data set to detect missing data timestamps and spatial location identifiers, generating supplementary data based on interpolation algorithms of adjacent observation records to form a preprocessed geospatial entity data set with continuous spatio-temporal coverage; performing outlier detection processing on the preprocessed geospatial entity data set, calculating the numerical distribution range of each geospatial observation record, removing data points outside the preset distribution range, and performing sliding window mean filling processing on the removed gaps to generate a standardized geospatial entity data set; performing a time series segmentation operation on the standardized geographical entity data set, dividing geographical entity observation records into a plurality of time window subsequences according to a preset period length, each time window subsequence containing a fixed number of continuous observation records, and performing an overlapping cutting process on observation records at the boundaries of the time windows to avoid data truncation; performing a trend decomposition process on each time window subsequence, separating long-term trend components, seasonal period components, and residual noise components of the geographical entity observation records using a seasonal and trend decomposition algorithm, extracting slope change features of the long-term trend components and amplitude fluctuation features of the seasonal period components as time series correlation features; performing a spatial clustering analysis process based on the spatial position identifier, constructing a topological network containing spatial adjacency relationships of geographical entities, and extracting spatial topological features of the topological network; performing a feature alignment process on the time series correlation features and the spatial topological features, generating a feature dimension consistent with the number of spatial sub-regions by aggregating time series correlation features of all time window subsequences within a spatial sub-region, so that the mapping relationship between the time window subsequences and the spatial sub-regions is consistent, and combining the aligned features into a multi-dimensional feature matrix.

3. The method of claim 2, wherein, The time series segmentation operation on the standardized geographical entity data set, dividing geographical entity observation records into a plurality of time window subsequences according to a preset period length, includes: obtaining the timestamp sequence of all geographical entity observation records in the standardized geographical entity data set, sorting the observation records according to the chronological order of the timestamps, and generating an ordered time series data stream; determining the time window length and the sliding step length parameters according to the preset division rule, and dividing the ordered time series data stream into a plurality of initial time window subsequences in a sliding window manner; detecting the number of observation records in each initial time window subsequence, if the number is lower than a preset minimum record threshold, performing a window expansion operation to dynamically adjust the window length to contain a sufficient number of observation records; performing time alignment processing on the expanded time window subsequences, aligning the time offsets of observation records in different time windows using a dynamic time warping algorithm to eliminate time sequence misalignment caused by inconsistent data collection frequencies; performing normalization processing on the aligned time window subsequences, associating and storing the normalized time window subsequences with the corresponding spatial position identifiers, and establishing a bidirectional mapping relationship between the time window index and the spatial region identifier.

4. The method of claim 2, wherein, The spatial clustering analysis process based on the spatial position identifier, constructing a topological network containing spatial adjacency relationships of geographical entities, and extracting spatial topological features of the topological network, includes: analyzing the latitude and longitude coordinate information of the spatial position identifier of each geographical entity to construct a geographical entity coordinate set; calculating the spatial distance matrix between each pair of geographical entity coordinates, dynamically adjusting the adjacency threshold based on the distribution density of the geographical entity coordinates, and if the spatial distance between two geographical entities is less than the dynamic adjacency threshold, establishing a corresponding connection edge in the topological network; The connection edge is subjected to a weight assignment operation, and the weight value is inversely proportional to the spatial distance, that is, the smaller the spatial distance, the greater the weight value of the connection edge; All geographic entity nodes are traversed to build a weighted connection edge undirected topological network, and the degree centrality index of each node is calculated to represent its spatial connection density; A community detection algorithm is performed on the undirected topological network to identify a high-cohesion geographic entity subgroup, and the spatial coverage range and internal connection strength index of each subgroup are extracted; Based on the degree centrality index and subgroup characteristics, a spatial topological feature vector is generated, including node connection density, subgroup coverage radius, and cross-group connection strength.

5. The method of claim 1, wherein, The cross-modal attention mechanism is constructed, the correlation matrix between the time series encoding feature vector and the spatial encoding feature vector is calculated, and the dynamic weight coefficient is generated based on the correlation matrix, including: The time series encoding feature vector and the spatial encoding feature vector are respectively mapped to a high-dimensional hidden space to generate a time series query vector and a spatial key vector; The dot product similarity between the time series query vector and the spatial key vector is calculated to generate an initial attention score matrix; The initial attention score matrix is normalized, and the score is normalized to a probability distribution using the Softmax function to generate an attention weight matrix; The spatial encoding feature vector is weighted and aggregated according to the attention weight matrix to generate a spatial attention feature vector; The spatial attention feature vector and the time series encoding feature vector are spliced, and cross-modal information is fused through a fully connected layer to generate a joint feature representation; The joint feature representation is subjected to layer normalization processing to eliminate the scale difference between the features, and residual connection is used to retain the original feature information to generate the final dynamic weight coefficient.

6. The method of claim 1, wherein, The fusion feature set is input into a geographic entity state prediction model to calculate the state change probability of the geographic entity in a future time window through a multi-layer perception network, and a probability distribution prediction result is generated; According to the probability distribution prediction result, a risk geographic entity is identified, the resource demand urgency score of the risk geographic entity is calculated, and a priority list is generated based on the resource demand urgency score; A multi-objective optimization model is constructed, taking the maximum resource allocation efficiency and the minimum geographic entity protection cost as the objective function, and introducing constraint conditions to limit the total amount of resources, the allocation time window, and the resource allocation order based on the priority list; An evolutionary algorithm is used to solve the multi-objective optimization model to generate a Pareto optimal solution set, and an optimal solution that meets the preset strategy is selected through an interactive decision interface; The resource allocation scheme corresponding to the optimal solution is integrated with the geographic entity state prediction result to generate a geographic space optimization strategy including specific execution steps and a time table. ​ The geographic space optimization strategy is verified for feasibility, a state change of a geographic entity after execution of the geographic space optimization strategy is simulated, and if a simulation result exceeds a preset risk threshold, a target function weight or a constraint condition relaxation range of the multi-objective optimization model is dynamically adjusted according to a risk type, the geographic space optimization strategy is regenerated.

7. The method of claim 6, wherein, The multi-objective optimization model is solved by using an evolutionary algorithm, a Pareto optimal solution set is generated, and an optimal solution meeting a preset strategy is selected through an interactive decision interface, including: Population initialization parameters, a crossover probability parameter and a mutation probability parameter of the evolutionary algorithm are set; An initial population is generated based on the resource allocation priority configuration and a constraint condition of the multi-objective optimization model, wherein each individual represents a resource allocation scheme, a constraint satisfaction algorithm is used to verify the legality of each individual, individuals violating a total resource constraint, an allocation time window constraint and a resource allocation sequence constraint are removed, and new individuals meeting the constraints are supplemented; Fitness of each individual in the initial population is evaluated, multi-objective fitness values are calculated based on the resource allocation efficiency maximization objective function and the geographic entity protection cost minimization objective function; Genetic operations are performed on the initial population, including a selection operation based on the fitness values, a crossover operation based on the crossover probability parameter and a mutation operation based on the mutation probability parameter, and a child population is generated; The initial population and the child population are merged, non-dominated sorting processing is performed to determine a Pareto front level of an individual, and a crowding distance is calculated to screen a diversity individual, and a new generation population is generated; The above steps are iteratively performed until an evolutionary termination condition is reached, and the Pareto optimal solution set is output; The Pareto optimal solution set is visualized in the interactive decision interface, a selection instruction input by a user is received, and a final optimal solution is determined based on a preset strategy matching degree.

8. The method of claim 1, wherein, The geographic information system update operation of a target region triggered according to the geographic space optimization strategy includes: Specific execution steps and a time table in the geographic space optimization strategy are analyzed, resource allocation scheme parameters and spatial position identifiers of target geographic entities matching a geographic entity state prediction result are extracted; An incremental update instruction of the geographic information system is generated according to the resource allocation scheme parameters, the incremental update instruction contains a spatial coordinate sequence of a resource allocation path, a timestamp constraint condition and corresponding geographic entity attribute modification values; The incremental update instruction is subjected to spatio-temporal conflict detection, whether the resource allocation path under the timestamp constraint condition overlaps with a spatial coordinate sequence of an existing task in the geographic information system is verified, and a priority order of the resource allocation path is dynamically adjusted based on a conflict detection result; The adjusted incremental update instruction is split into multiple atomic transaction operation units according to the time table, each atomic transaction operation unit is bound to a corresponding geographic entity spatial position identifier and a resource allocation parameter verification rule, and a version lock identifier is added to each atomic transaction operation unit, so that concurrent operations on the same geographic entity are performed in the priority order; Triggering an asynchronous execution queue of the atomic transaction operation unit based on a timestamp constraint condition of the schedule, monitoring the execution state of each atomic transaction operation unit in real time and capturing a resource allocation parameter verification failure event; According to the resource allocation parameter verification failure event, backtracking to the corresponding geographical entity spatial position identifier, regenerating a local incremental update instruction and inserting it into a preset fault tolerance time window of the asynchronous execution queue; Rendering a resource allocation heat map corresponding to the executed atomic transaction operation unit in the visualization layer of the geographic information system, superimposing the change trajectory of the geographical entity state prediction result, and generating a dynamically updated spatial decision view; Reverse mapping the execution log of the atomic transaction operation unit with the geographical entity spatial position identifier, generating an incremental version snapshot and synchronizing it to the geographical spatial data copy of the edge node.

9. An artificial intelligence-based geospatial data management system, characterized by, The system comprises a processor and a memory, the memory and the processor are connected, the memory is used to store programs, instructions or codes, and the processor is used to run the programs, instructions or codes in the memory to realize the geographic spatial data management method based on artificial intelligence in any one of claims 1-8.

Images