Automatic early warning system for environmental anomaly in urban ecological sensitive area based on spatio-temporal data mining

Abstract

This invention relates to the field of environmental anomaly early warning technology, and discloses an automatic early warning system for environmental anomalies in urban ecologically sensitive areas based on spatiotemporal data mining. The system includes: extracting intermediate center scalars; calculating rainfall concentration coefficients; generating bottleneck boundary retention and collapse coefficients; generating a dynamic adjacency matrix; outputting predicted future anomaly intensity values; generating cumulative anomaly energy; and outputting automatic early warning values ​​for the entire area. This invention constructs structural segments by spatially overlapping the catchment area of ​​the total urban ecologically sensitive area with riverside or boundary strip areas. It uses intermediate center scalars to identify topological bottleneck segments, employs a dynamic adjacency matrix to allow the propagation weights between structural segments to change in real time with the boundary hydraulic conditions, and integrates the bottleneck boundary retention and collapse coefficients into the gating mechanism of the spatiotemporal graph convolutional network to achieve adaptive adjustment of the model's time memory. It outputs standardized early warning values ​​for the entire area, adapting to practical application scenarios for environmental management in urban ecologically sensitive areas.

Description

Technical Field

[0001] This invention relates to the field of environmental anomaly early warning technology, and more specifically, to an automatic early warning system for environmental anomalies in urban ecologically sensitive areas based on spatiotemporal data mining. Background Technology

[0002] Urban ecologically sensitive areas serve as crucial buffer zones and sites for urban water environment restoration, making them highly susceptible to disturbances from rainfall runoff and non-point source pollution. Therefore, early warning of environmental anomalies is a vital aspect of urban ecological management. Currently, most environmental anomaly monitoring and early warning systems in urban ecologically sensitive areas rely on single-point water quality and level monitoring data, determining anomalies solely based on real-time values ​​of a single indicator. This approach only reflects the instantaneous state of local locations and cannot analyze the spatial propagation process of anomalies, resulting in significant delays and limitations in early warning information.

[0003] Existing early warning technologies generally use administrative boundaries or regular grids to divide analysis units, failing to construct analysis units that align with anomaly propagation mechanisms by incorporating confluence control relationships and riverside boundary attributes. Most early warning models use static adjacency matrices to describe the spatial connectivity between structural segments, unable to dynamically adjust propagation weights based on boundary stagnation states and hydraulic conditions. These methods ignore the amplifying effect of boundary fragmentation and topological bottlenecks on anomaly propagation, and fail to integrate boundary stagnation and collapse characteristics into the model's time gating mechanism, making it impossible to adaptively adjust the weights of historical states and real-time monitoring data, resulting in low accuracy in anomaly prediction. Furthermore, most early warning systems ultimately output multiple scattered indicator results, failing to integrate them into a unified, quantified area-wide early warning value, making direct integration into urban environmental management workflows difficult. Early warning risk level classification lacks customized threshold rules matching the structural characteristics of ecologically sensitive areas, and early warning responses lack clear high-risk location indications. These various problems collectively prevent existing technologies from achieving early prediction and quantitative early warning of environmental anomalies in urban ecologically sensitive areas, failing to meet the actual needs of refined management and emergency response in these areas. Summary of the Invention

[0004] This invention provides an automatic early warning system for environmental anomalies in urban ecologically sensitive areas based on spatiotemporal data mining, which solves the technical problems mentioned in the background.

[0005] This invention provides an automatic early warning system for environmental anomalies in urban ecologically sensitive areas based on spatiotemporal data mining, comprising: The first module involves finding the intersection of the total area of ​​the urban ecologically sensitive zone, the water catchment area, and the riverside or boundary strip area to obtain the structural segment, and extracting the intermediate center scalar of the structural segment; The second module calculates the runoff volume using rainfall intensity and impermeability, calculates the fragmented release volume using boundary length and effective boundary retention area, and calculates the rainfall concentration coefficient by combining local rainfall intensity. The third module transforms the intermediate center scalar into a bottleneck factor, obtains the effective residence time from the confluence flow and effective water depth, and generates the bottleneck boundary storage and collapse coefficient by combining the rainfall concentration coefficient, bottleneck factor, fragmented release volume and effective residence time. The fourth module obtains the hydraulic drawdown factor from the effective water depth difference, and integrates the hydraulic drawdown factor, hydrological path distance and bottleneck boundary retention collapse coefficient to generate a dynamic adjacency matrix. The fifth module inputs the dynamic adjacency matrix into the spatiotemporal graph convolutional network, incorporates the bottleneck boundary stagnation and collapse coefficient into the gating mechanism, and outputs the predicted value of future anomaly intensity. The sixth module amplifies and accumulates the predicted values ​​of future anomaly intensity by combining time decay and bottleneck boundary stagnation and collapse coefficients, generating the cumulative anomaly energy of the structural segment. The seventh module determines the structural segment allocation weights based on the area of ​​sensitive ecological patches and bottleneck factors, uses the structural segment allocation weights to weight and normalize the cumulative anomalies, and outputs the automatic early warning value for the entire area.

[0006] The beneficial effects of this invention are as follows: This invention constructs structural segments by spatially overlapping the total catchment area of ​​an ecologically sensitive urban area with the riverside or boundary strip areas, making the analysis unit conform to the actual propagation mechanism of environmental anomalies. It uses a median center scalar to identify topological bottleneck segments and generates a bottleneck boundary retention and collapse coefficient by combining the concentration of fragmented rainfall at the boundary with the effective residence time, achieving a unified quantitative representation of multi-source influencing factors. It employs a dynamic adjacency matrix instead of static connection relationships, allowing the propagation weights between structural segments to change in real time with the boundary hydraulic conditions. It integrates the bottleneck boundary retention and collapse coefficient into the gating mechanism of a spatiotemporal graph convolutional network, achieving adaptive adjustment of the model's time memory. It generates cumulative anomaly energy through time decay and structural amplification processing, taking into account both the persistence characteristics of anomalies and structural risk characteristics. It completes weighted aggregation based on the area of ​​sensitive ecological patches and bottleneck factors, outputting a standardized warning value for the entire area. The warning results are presented in a unified format, can be integrated into the urban ecological environment management workflow, provide a reference for patrol arrangements and risk management in ecologically sensitive areas, and are suitable for practical application scenarios of environmental management in urban ecologically sensitive areas. Attached Figure Description

[0007] Figure 1 This is a flowchart of the calculation process of the automatic early warning system for environmental anomalies in urban ecologically sensitive areas based on spatiotemporal data mining, as described in this invention. Figure 2 This is a computational scenario diagram of the automatic early warning system for environmental anomalies in urban ecologically sensitive areas based on spatiotemporal data mining, as described in this invention. Detailed Implementation

[0008] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, features described in some examples may be combined in other examples.

[0009] It should be noted that, unless otherwise defined, the technical or scientific terms used in one or more embodiments of the present invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in one or more embodiments of the present invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" indicate that the element or object preceding the term encompasses the elements or objects listed following the term and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0010] like Figures 1-2 As shown, the automatic early warning system for environmental anomalies in urban ecologically sensitive areas based on spatiotemporal data mining includes: The first module involves finding the intersection of the total area of ​​the urban ecologically sensitive zone, the water catchment area, and the riverside or boundary strip area to obtain the structural segment, and extracting the intermediate center scalar of the structural segment; The second module calculates the runoff volume using rainfall intensity and impermeability, calculates the fragmented release volume using boundary length and effective boundary retention area, and calculates the rainfall concentration coefficient by combining local rainfall intensity. The third module transforms the intermediate center scalar into a bottleneck factor, obtains the effective residence time from the confluence flow and effective water depth, and generates the bottleneck boundary storage and collapse coefficient by combining the rainfall concentration coefficient, bottleneck factor, fragmented release volume and effective residence time. The fourth module obtains the hydraulic drawdown factor from the effective water depth difference, and integrates the hydraulic drawdown factor, hydrological path distance and bottleneck boundary retention collapse coefficient to generate a dynamic adjacency matrix. The fifth module inputs the dynamic adjacency matrix into the spatiotemporal graph convolutional network, incorporates the bottleneck boundary stagnation and collapse coefficient into the gating mechanism, and outputs the predicted value of future anomaly intensity. The sixth module amplifies and accumulates the predicted values ​​of future anomaly intensity by combining time decay and bottleneck boundary stagnation and collapse coefficients, generating the cumulative anomaly energy of the structural segment. The seventh module determines the structural segment allocation weights based on the area of ​​sensitive ecological patches and bottleneck factors, uses the structural segment allocation weights to weight and normalize the cumulative anomalies, and outputs the automatic early warning value for the entire area.

[0011] In one embodiment of the present invention, a structural segment is obtained by intersecting the total area of ​​the urban ecologically sensitive area, the water catchment area, and the riverside or boundary strip area. The intermediate center scalar of the structural segment is extracted, including: Structural segments are generated through spatial intersection operations. The calculation formula is as follows: In the formula, The total area of ​​the city's ecologically sensitive zone. The area affected by water runoff For riverside or boundary strip areas, For structural segments, To perform the intersection operation; With structural segments As a statistical unit, statistical arbitrary source structure segments To the target structural segment The total number of shortest propagation paths between And count the shortest propagation paths that pass through the current structural segment. Path count ; Based on the total number of shortest propagation paths The shortest propagation path passes through the current structural segment Path count Extracting the scalar of the intermediary center The calculation formula is as follows: In the formula, For the intermediary center's metric, For source structure segment, For the target structure segment, For the current structural segment, The total number of shortest propagation paths, For the shortest propagation path passing through the current structural segment The number of paths, This function performs a summation operation on the combination of the source structure segment, the target structure segment, and the current structure segment that meet the conditions.

[0012] It should be noted that the total extent of the urban ecologically sensitive area is the overall spatial extent data used to define the boundaries of all structural segments, which can be obtained through ecologically sensitive area delineation results, remote sensing interpretation results, and spatial layer acquisition. The catchment area refers to the spatial extent that can influence the corresponding structural segment through surface runoff, pipe network confluence, or tributary inflow, which can be obtained through catchment area delineation results and spatial layer acquisition. The riverside or boundary strip area refers to a strip-shaped functional area laid out along the river channel, water body edge, or ecologically sensitive boundary, which can be obtained through riverside or boundary strip delineation results and spatial layer acquisition. A structural segment is the smallest analytical unit formed by overlaying the total extent of the urban ecologically sensitive area, the catchment area, and the riverside or boundary strip area. The intermediate center scalar is a topological quantity used to characterize the criticality of the current structural segment in the entire propagation path. The source structural segment is the starting structural segment of the shortest propagation path. The target structural segment is the ending structural segment of the shortest propagation path. The current structural segment is the structural segment currently undergoing path statistics and centrality calculation. The total number of shortest propagation paths is the number of paths between the source and target structural segments that satisfy the minimum propagation cost condition. The number of paths in the shortest propagation path that pass through the current structural segment is the total number of shortest propagation paths that pass through the current structural segment.

[0013] It should be noted that, since anomalous propagation in urban ecologically sensitive areas is constrained by ecological boundaries, hydrological runoff, and riverside location, only by simultaneously intersecting the total area of ​​the urban ecologically sensitive area, the catchment area, and the riverside or boundary strip area can a structural segment possessing ecological sensitivity, runoff source, and boundary location attributes be obtained. Because anomalous diffusion between structural segments is not a simple geometric adjacency diffusion but occurs along actual propagation channels, using the total number of shortest propagation paths rather than ordinary adjacency counts is necessary to reflect the criticality of the current structural segment within the overall propagation network. Since whether a structural segment constitutes a bottleneck depends on how many shortest propagation paths pass through it, accumulating the ratio of the number of paths passing through the current structural segment to the total number of shortest propagation paths is necessary to form a mediator scalar and stably characterize the topological bottleneck nature of the structural segment.

[0014] It should be noted that the method for determining the catchment area is as follows: starting from the inflow control position corresponding to each structural segment, tracing upstream along the surface flow direction and existing flow boundary, a closed area that contributes flow only to that structural segment is obtained. The method for determining the riverside or boundary strip area is as follows: strips are extracted along the riverbank or ecological boundary where the structural segment is located, with a uniform width or functional limit, and the formation conditions being continuous strip centerlines and closed strip boundaries. The method for constructing the total number of shortest propagation paths between structural segments is as follows: using structural segments as nodes and actual propagable relationships as directed connections, the shortest path between any two structural segments is calculated according to the principle of minimum propagation cost, and all shortest paths of the same length are simultaneously included in the total. The method for counting the number of paths passing through the current structural segment in the shortest propagation path is as follows: all combinations of source and target structural segments are iterated one by one; when a shortest path includes the current structural segment and the current structural segment is neither the starting point nor the ending point of the path, the path is included in the number of paths passing through the current structural segment. The propagation direction, edge weight attributes, and handling of disconnected cases corresponding to the shortest propagation path are as follows: the propagation direction adopts a unidirectional relationship from the upstream structural segment to the downstream structural segment, the edge weight is uniformly quantified using the propagation cost, and the combination of disconnected structural segments does not participate in the summation statistics.

[0015] In one embodiment of the present invention, the runoff is calculated using rainfall intensity and impermeability, the fragmented release is calculated using boundary length and effective boundary retention area, and the rainfall concentration factor is calculated by combining local rainfall intensity, including: For the merging structure segment upstream catchment unit set Each upstream catchment unit Generate convergence weights The calculation formula is as follows: In the formula, To merge the structural segment The upstream catchment unit set, For upstream catchment units Any upstream catchment unit in, upstream catchment unit For structural segments The confluence weight, upstream catchment unit impermeability upstream catchment unit area, For upstream catchment units The sum of the impermeability and area products of all upstream catchment units within the area. It is a very small positive number; Using the confluence weight With upstream water catchment unit Rainfall intensity generation structure segment At any moment Exchange rate The calculation formula is as follows: In the formula, For structural segments At any moment Exchange rate liquidity upstream catchment unit For structural segments The confluence weight, upstream catchment unit At any moment The intensity of rainfall, upstream catchment unit To structural segment The delay in transmission, For the merging structure segment Perform a summation operation on all upstream catchment units; Utilizing structural segments boundary length With effective boundary storage area Generate fragmented amplification The calculation formula is as follows: In the formula, For structural segments The fragmentation is amplified. For structural segments The boundary length, For structural segments The effective boundary storage area Pi For natural logarithm operations; Utilizing structural segments Local rainfall intensity With structural segment Corresponding neighborhood range Average rainfall intensity generates rainfall concentration factor The calculation formula is as follows: In the formula, For structural segments At any moment Rainfall concentration coefficient For structural segments At any moment The intensity of localized rainfall, For the neighborhood range, Neighborhood range area, For spatial location At any moment The intensity of rainfall, Neighborhood range Integral amount of rainfall intensity within, It is a very small positive number.

[0016] It should be noted that the upstream catchment unit set refers to the set of all upstream units that can contribute to the current structural segment's flow. An upstream catchment unit is a single flow unit within the upstream catchment unit set. The flow weight is a weight value used to characterize the proportion of the flow contribution of a single upstream catchment unit to the current structural segment. Impermeability is a proportional parameter used to characterize the degree of surface hardening of the upstream catchment unit, which can be obtained through impermeable surface interpretation results, land use classification results, and spatial layer data collection. Area is a quantity used to characterize the geometric scale of the upstream catchment unit. The minimum positive number is a protective parameter used to avoid zero denominators and maintain computational stability; a preferred value is 0.000001, which is sufficient to eliminate the risk of division by zero and will not substantially disturb the normalization results. Flow rate is the total flow input received by the current structural segment at the current moment. Time is a time parameter used to identify the observation or calculation time point to which each time series quantity belongs. Rainfall intensity is the rainfall input intensity of the upstream catchment unit at the corresponding moment, which can be obtained through rainfall monitoring, radar inversion, or gridded rainfall data collection. Time delay is the propagation time required for rainfall from the upstream catchment unit to reach the current structural segment. Fragmentation amplification is an amplification used to characterize the degree of fragmentation and vulnerability of the structural segment boundary morphology. Boundary length is the total length of the outer boundary of the structural segment. Effective boundary retention area is the functional area within the structural segment boundary that actually participates in retention, buffering, and containment. Rainfall concentration factor is a ratio coefficient used to characterize whether the local rainfall in the current structural segment is higher than the average level of the neighborhood. Local rainfall intensity is the rainfall intensity at the current location of the structural segment at the current time, which can be obtained through rainfall monitoring, radar inversion, or gridded rainfall data acquisition. Neighborhood range is the local spatial range used to calculate the average rainfall intensity around the current structural segment. The area of ​​the neighborhood range is the total area corresponding to the neighborhood range. Spatial location is the specific location of the rainfall intensity in the spatial field. The rainfall intensity at a spatial location at a given time is the value of the rainfall intensity field at a specified spatial location and time, which can be obtained through rainfall monitoring, radar inversion, or gridded rainfall data acquisition.

[0017] It should be noted that the runoff contribution of urban ecologically sensitive areas is not determined solely by distance, but is simultaneously influenced by the surface hardening degree and unit size of upstream catchment units. Therefore, using impermeability and area together to generate runoff weights is necessary to reflect the actual runoff contribution capacity of each upstream catchment unit. Because rainfall has a propagation lag when it reaches a structural segment, runoff flux cannot be directly calculated by summing rainfall intensities at the same moment. Instead, it should be calculated by first extrapolating back to the moment when it actually affects the current structural segment, and then weighted according to the runoff weights. Since a larger boundary length and a smaller effective boundary retention area indicate a narrower, more fragmented, and less compact structural segment, using the ratio of the square of the boundary length to the effective boundary retention area and performing a natural logarithmic operation can stably characterize the structural effect of boundary fragmentation on anomalously amplified structures. Since anomalies often first appear at the core of local heavy rainfall rather than at the neighborhood average location, constructing a rainfall concentration coefficient using the ratio of local rainfall intensity to the neighborhood average rainfall intensity can reflect whether the current structural segment has experienced concentrated rainfall impacts higher than the surrounding background levels.

[0018] It should be noted that the upstream catchment unit set is divided as follows: taking the current structural segment as the endpoint of the flow, all spatial units contributing to the flow are decomposed upstream along the flow direction, with each unit belonging to only one structural segment as the dividing principle. The impermeability rate is obtained by statistically analyzing the proportion of impermeable surfaces within the upstream catchment unit at a uniform scale, using the classification results of the same time phase as the input for the current round of calculation. The time delay is calculated by calculating the propagation time based on the actual propagation path length and average propagation speed from the upstream catchment unit to the current structural segment, and then uniformly converting the propagation time to the time step unit used in this invention. The effective boundary retention area is determined by identifying boundary functional areas within the structural segment boundary that can actually accommodate, retain, and buffer the flow, and only areas that maintain continuous connectivity and have retention capacity are included in the effective boundary retention area. The neighborhood range is determined by selecting local areas centered on the current structural segment according to a uniform spatial radius or a uniform adjacent level, ensuring that all structural segments use consistent neighborhood selection rules. The calculation method for the integral of rainfall intensity within the neighborhood is as follows: the neighborhood is discretized into computational units with uniform spatial resolution, the rainfall intensity of all computational units at the current moment is summed, and the discrete summation result is used to approximate the continuous integral result. The method for distinguishing time is as follows: when time is used to represent temporal meaning, it is only used to identify the time position of observation and prediction, and is no longer used to represent the target structural segment. The method for setting the minimum positive number is as follows: a uniform value is used in all formulas containing division operations, and this value is set once during system initialization and then fixed for use throughout the entire process.

[0019] In one embodiment of the present invention, the intermediate center scalar is converted into a bottleneck factor, and the effective residence time is obtained from the sinking flux and effective water depth. The bottleneck boundary retention and collapse coefficient is generated by combining the rainfall concentration coefficient, bottleneck factor, fragmented release volume, and effective residence time, including: Using intermediary center scalar Generate bottleneck factor The calculation formula is as follows: In the formula, As a bottleneck factor, For the intermediary center's scalar, It is the minimum value among the intermediate center scalars corresponding to all structural segments. It is the maximum value among the intermediate center scalars corresponding to all structural segments. It is a very small positive number; Utilizing the flow of funds Generate valid stay time The calculation formula is as follows: In the formula, For the effective stay time, For the effective boundary storage area, For effective water depth, To increase circulation volume, It is a very small positive number; Using rainfall concentration factor Fragmentation and large-scale release and bottleneck factors With effective stay time Generate bottleneck boundary stagnation collapse coefficient The calculation formula is as follows: In the formula, The bottleneck boundary stagnation collapse coefficient, This is the rainfall concentration factor. As a bottleneck factor, To achieve large-scale fragmentation, For the effective stay time, It is a very small positive number. To eliminate the reference time constant in time dimension, its value can be set to one standard time step.

[0020] It should be noted that the bottleneck factor is a structural bottleneck degree parameter obtained by normalizing the intermediate center scalar. The minimum value among the intermediate center scalars corresponding to all structural segments is the statistic used to normalize the lower bound. The maximum value among the intermediate center scalars corresponding to all structural segments is the statistic used to normalize the upper bound. Effective residence time is a parameter characterizing the duration of the current structural segment's ability to retain confluence and anomalies. Effective water depth is the water depth parameter of the current structural segment that actually participates in the storage and propagation process. It can be obtained through water depth monitoring, water level conversion, and structural segment water depth data collection. The bottleneck boundary storage collapse coefficient is a core coefficient comprehensively characterizing whether the boundary transitions from a buffer state to an unstable state.

[0021] It should be noted that since the absolute value of the intermediate center scalar varies with the number of structural segments and the scale of the propagation network, the intermediate center scalar of the current structural segment must be normalized using the maximum and minimum values ​​among all structural segments. Only after normalization can the bottleneck factor remain comparable across different structural segments. Because the storage capacity of a structural segment is simultaneously affected by the boundary functional area that can participate in storage, the effective water depth that can participate in storage, and the intensity of external confluence impact, the effective residence time with clear physical meaning can be obtained by dividing the product of the effective boundary storage area and the effective water depth by the confluence flow. Since whether a structural segment enters a boundary instability state is not determined by a single factor but by the combined effects of concentrated rainfall, topological bottlenecks, boundary fragmentation, and residence time, the bottleneck boundary storage collapse coefficient must be generated by integrating these quantities to compress multi-source heterogeneous factors into a unified structural risk characterization quantity.

[0022] It should be noted that the effective water depth is obtained as follows: a representative water depth value is selected within each structural segment, and the portion of the water depth that can actually participate in storage and propagation at the current moment is used as the input for the effective water depth calculation. The maximum and minimum values ​​of the intermediate center scalars corresponding to all structural segments are statistically analyzed as follows: in the same round of structural segment generation results, a full traversal of the intermediate center scalars of all structural segments is performed, and the maximum and minimum values ​​are extracted and used for normalization of all subsequent structural segments. The time step and unit unification method for the effective residence time is as follows: the flow rate, effective boundary storage area, and effective water depth are uniformly converted to the same measurement time base before the ratio calculation is performed. The interpretation of the bottleneck boundary storage collapse coefficient is as follows: the larger the coefficient, the weaker the boundary buffering capacity and the stronger the propagation and propagation capacity of the current structural segment; the smaller the coefficient, the stronger the boundary storage capacity of the current structural segment.

[0023] In one embodiment of the present invention, a hydraulic drawdown factor is obtained from the effective water depth difference, and a dynamic adjacency matrix is ​​generated by fusing the hydraulic drawdown factor, hydrological path distance, and bottleneck boundary retention collapse coefficient, including: Utilizing structural segments Effective water depth With adjacent structural segments Effective water depth Generate hydraulic drawdown factor The calculation formula is as follows: In the formula, For structural segments Pointer to structure segment Hydraulic drawdown factor For structural segments Effective water depth, For structural segments Effective water depth, To be related to structural segments The set of adjacent structural segments, For the set of adjacent structural segments Any structural segment in, To perform the operation between the zero value and the larger value within the parentheses, It is a very small positive number; Utilizing hydrological path distance Hydraulic drawdown factor and the bottleneck boundary stagnation collapse coefficient Generate dynamic adjacency matrix elements The calculation formula is as follows: In the formula, For the structural segment in the dynamic adjacency matrix Pointer to structure segment matrix elements, For structural segments With structural segment The hydrological path distance between them This is a parameter representing the hydrological path distance attenuation scale. This is for natural exponent calculations. For hydraulic drawdown factor, For structural segments At any moment The bottleneck boundary stagnation collapse coefficient, To be related to structural segments The set of adjacent structural segments, For structural segments and adjacent structural segments The hydrological path distance between them For structural segments Pointing to adjacent structural segments Hydraulic drawdown factor; Combine all dynamic adjacency matrix elements Generation time Dynamic adjacency matrix The calculation formula is as follows: In the formula, For a moment The dynamic adjacency matrix, For the structural segment in the dynamic adjacency matrix Pointer to structure segment Matrix elements.

[0024] It should be noted that the hydraulic head factor is a directional factor used to characterize the probability of forward propagation from the current structural segment to adjacent structural segments. Adjacent structural segments are those adjacent to the current structural segment that have a direct propagation relationship. The set of structural segments adjacent to a structural segment is the set of all directly adjacent propagation objects of the current structural segment. Any structural segment in the set of adjacent structural segments is a single adjacent structural segment used for normalization summation. The hydrological path distance is the propagation distance measured along the actual propagation channel between the current structural segment and adjacent structural segments. The hydrological path distance attenuation scale parameter is a custom parameter controlling the intensity of hydrological path distance attenuation, preferably ranging from 50 meters to 500 meters. This range can distinguish between near-neighbor propagation and far-neighbor propagation and is suitable for the scale of adjacent structural segments commonly found in urban ecologically sensitive areas. The dynamic adjacency matrix element is a single weighted element in the dynamic adjacency matrix that points from the current structural segment to adjacent structural segments. The dynamic adjacency matrix is ​​a structural segment propagation relationship matrix updated at regular intervals.

[0025] It should be noted that since the propagation between structural segments is affected not only by their adjacency but also by the effective water depth difference between the current structural segment and its adjacent segments, normalizing the positive effective water depth difference into a hydraulic drawdown factor can characterize the directional strength of anomaly propagation from high to low potential. Because the actual propagation capacity is simultaneously affected by channel length, forward propagation capacity, and boundary instability, the hydrological path distance, hydraulic drawdown factor, and bottleneck boundary retention and collapse coefficient must be integrated into dynamic adjacency matrix elements to obtain edge weights that reflect the actual propagation state. Since the propagation relationship between structural segments changes over time, all dynamic adjacency matrix elements need to be recombined into a dynamic adjacency matrix at each time step to avoid static adjacency relationships masking real-time propagation changes.

[0026] It should be noted that the set of adjacent structural segments is determined as follows: only structural segments that have a direct connection with the current structural segment in the propagation channel and do not have intermediate jumping structural segments are included in the set of adjacent structural segments. The hydrological path distance is calculated by accumulating the path length segment by segment along the actual propagation channel from the current structural segment to the adjacent structural segment, and using the total path length as the hydrological path distance. The hydrological path distance attenuation scale parameter is set by pre-setting a uniform value based on the statistical results of the average path distance of the structural segments, and keeping it unchanged within the same model version. The normalization method for the dynamic adjacency matrix elements is as follows: summing the unnormalized weights of the current structural segment pointing to all adjacent structural segments, and then dividing each unnormalized weight by the summation result to obtain the normalized dynamic adjacency matrix element. The handling of whether the dynamic adjacency matrix includes self-connections and its directionality is as follows: the dynamic adjacency matrix uses a directed matrix, and self-connected elements do not participate in the adjacency propagation calculation in the current claim.

[0027] In one embodiment of the present invention, a dynamic adjacency matrix is ​​input into a spatiotemporal graph convolutional network, and the bottleneck boundary stagnation and collapse coefficient is connected to a gating mechanism to output a predicted value of future anomaly intensity, including: Using dynamic adjacency matrix Generate spatial aggregation representation The calculation formula is as follows: In the formula, For the current structural segment In the Layer, Time Spatial aggregation representation, For activation function, To be consistent with the current structural segment The set of adjacent structural segments, For dynamic adjacency matrix From the current structural segment Pointing to adjacent structural segments matrix elements, For the first The spatial convolution weight matrix of the layer, Adjacent structural segments In the Layer, Time The hidden state; Using the bottleneck boundary storage collapse coefficient Generate gating coefficients The calculation formula is as follows: In the formula, For the current structural segment At any moment The gating coefficient, For the gated weight vector, To aggregate representation of space With the bottleneck boundary stagnation collapse coefficient The resulting vector after concatenation For gated bias terms, For activation functions; Fusion Spatial Aggregation Representation Generate candidate hidden states The calculation formula is as follows: In the formula, For the current structural segment In the Layer, Time The candidate hidden state, For the first The time-update weight matrix of the layer, This is the time-recursive weight matrix corresponding to the hidden state at the previous time step. For the current structural segment In the previous hidden state, Update the bias term for time. It is the hyperbolic tangent activation function; Using gating coefficient Execute gated fusion to generate hidden states The calculation formula is as follows: In the formula, For the current structural segment In the Layer, Time The hidden state, This is element-wise multiplication; Output the predicted value of future anomaly intensity using the hidden state of the final layer. The calculation formula is as follows: In the formula, For the current structural segment In the future The predicted value of future anomaly intensity, For smooth non-negative activation functions, This is the output layer weight vector. For the final layer In the future The hidden state, For output layer bias terms, To predict the step size for the future, Predict the upper limit of the step size for the future.

[0028] It should be noted that the spatial aggregation representation is the spatial feature representation obtained by weighted aggregation of the hidden states of adjacent structural segments at the current layer and current time step. Layer numbering is a parameter used to distinguish different layers of the spatiotemporal graph convolutional network. The spatial convolution weight matrix of layer l is a trainable matrix used to map the hidden states of adjacent structural segments to the spatial aggregation representation of the current structural segment. The hidden state of an adjacent structural segment at time t in layer l-1 is the state representation of the adjacent structural segment at the same time step in the previous layer. Gating coefficients are coefficients used to control the fusion ratio between candidate hidden states and the hidden state at the previous time step. The gating weight vector is a trainable vector used to calculate the gating coefficients from the concatenated vector. The concatenated vector is a joint vector formed by connecting the spatial aggregation representation and the bottleneck boundary stagnation collapse coefficient end-to-end. The gating bias term is a trainable bias used to correct the calculated gating coefficients. The candidate hidden state is a candidate state generated by the current structural segment at the current layer and current time step based on the spatial aggregation representation and the hidden state at the previous time step. The time-update weight matrix of layer l is a trainable matrix used to map the spatial aggregation representation to the candidate hidden state. The time-recursive weight matrix corresponding to the hidden state at the previous time step is a trainable matrix used to map the hidden state at the previous time step to the candidate hidden state. The hidden state of the current structure segment at the previous time step is the state representation of the current structure segment retained at the previous time position. The time-update bias term is a trainable bias used to correct the calculation result of the candidate hidden state. The hidden state of the current structure segment at time t in layer l is the current layer state representation obtained after gating fusion update. The predicted future anomaly intensity value is the anomaly intensity output value of the current structure segment at the time corresponding to the future prediction step. The output layer weight vector is a trainable vector used to map the hidden state of the final layer to the predicted future anomaly intensity value. The hidden state of the final layer at a future time is the state representation of the final layer at the corresponding future time. The output layer bias term is a trainable bias used to correct the predicted future anomaly intensity value. The future prediction step is a step size parameter used to identify which prediction position is in the future. The upper limit of the future prediction step size is a custom parameter used to limit the output range of the predicted values ​​of future anomalies. It is preferred to take 3 to 12 time steps. This range can cover the short-term warning window and will not cause excessive error accumulation due to predictions that are too far away. The final layer is the last hidden layer used to output the predicted values ​​of future anomalies. It is preferred to take 2 to 4 layers. A shallower number of layers is sufficient to complete the local propagation modeling of structural segments, while a deeper number of layers can easily lead to overly smooth states.

[0029] It should be noted that since spatiotemporal graph convolutional networks can simultaneously handle the connection relationships and state evolution relationships between structural segments, inputting the dynamic adjacency matrix into the spatiotemporal graph convolutional network allows for joint modeling of propagation and temporal relationships within the same network. Because the state of the current structural segment is influenced by the states of adjacent structural segments, and the intensity of this influence varies with the elements of the dynamic adjacency matrix, the spatial aggregation representation obtained by weighted aggregation of the hidden states of the previous layer of adjacent structural segments can characterize the spatial propagation input of the current structural segment. Since the bottleneck boundary stagnation collapse coefficient reflects whether the current structural segment boundary is in an unstable state, concatenating the spatial aggregation representation with the bottleneck boundary stagnation collapse coefficient and inputting it into the gating mechanism allows the gating coefficient to simultaneously perceive both spatial input and the degree of boundary instability. Because candidate hidden states need to simultaneously absorb the current spatial propagation input and the existing temporal memory of the current structural segment, a reasonable candidate hidden state can only be formed by fusing the spatial aggregation representation with the hidden state of the previous time step through a time-updated weight matrix and a time-recursive weight matrix. Since the impact of the current shock on state updates should be increased when the boundary is unstable, and more historical memory should be retained when the boundary is stable, gating fusion updates of candidate hidden states and previous hidden states using gating coefficients can achieve adaptive time memory adjustment. Since the final warning is based on the anomaly intensity at future moments rather than the hidden state itself, the output layer weight vector needs to map the hidden states at future moments of the final layer to predicted values ​​of future anomaly intensity.

[0030] It should be noted that the input method of the spatiotemporal graph convolutional network is as follows: at each time step, the dynamic adjacency matrix is ​​used as the spatial connection input, and the hidden state of the current structural segment at the previous time step and the hidden state of the adjacent structural segments at the previous layer are used as the state input, and the process is advanced step by step at a uniform time step. The activation function is selected as follows: the spatial aggregation representation and gating coefficients use monotonically bounded activation functions, the candidate hidden states use bidirectional compression activation functions, and the predicted future anomaly intensity uses non-negative output activation functions. The gating weight vector, the spatial convolution weight matrix of the l-th layer, the time update weight matrix of the l-th layer, the time recursive weight matrix corresponding to the hidden state at the previous time step, and the output layer weight vector are obtained by using historical time series data as training input and performing iterative optimization with the goal of minimizing the error between the predicted future anomaly intensity and the actual anomaly intensity. The hidden state is initialized as follows: at the first time position, the hidden states of all structural segments are uniformly initialized to zero, and stable hidden states are gradually formed through gating fusion updates in subsequent time steps. The training method for future anomaly intensity prediction values ​​is as follows: supervised samples are constructed segment by segment using continuous historical time slices, and multiple future anomaly intensity prediction values ​​are output simultaneously in each training round according to the upper limit of the future prediction step size. The upper limit of the future prediction step size and the final layer are set as follows: the upper limit of the future prediction step size is set according to the warning lead time requirement, and the final layer is set according to a trade-off between prediction accuracy and computational complexity.

[0031] In one embodiment of the present invention, the predicted value of future anomaly intensity is amplified and accumulated by combining time decay and bottleneck boundary stagnation collapse coefficient to generate the cumulative anomaly energy of the structural segment, including: Using the upper limit of future prediction step size Generate time decay factor The calculation formula is as follows: In the formula, The time decay factor, It is a natural constant. Predict the upper limit of the step size for the future; Using time decay factor Predicted future anomaly intensity and the bottleneck boundary stagnation collapse coefficient Generate cumulative anomaly energy The calculation formula is as follows: In the formula, To accumulate abnormal energy, For predicting step size variables in the future, For structural segments In the future The predicted value of future anomaly intensity, For structural segments At the present moment The bottleneck boundary stagnation collapse coefficient, For natural logarithm operations, For future moments to The summation operation is performed on all calculation results within the range.

[0032] It should be noted that the time decay factor is used to control the gradual decay of the predicted intensity of future anomalies as the predicted location increases. The natural constant is a fixed constant used to generate the time decay factor. The cumulative anomaly energy is a comprehensive quantity obtained by processing all predicted intensities of future anomalies within the future prediction window according to time decay and structural amplification. Since the predicted intensities of future anomalies closer to the current time are more important for short-term early warning, the time decay factor should be generated based on the upper limit of the future prediction step size, and closer prediction locations should be given higher weights. Since a single predicted intensity of future anomalies cannot simultaneously reflect persistence and structural vulnerability, all predicted intensities of future anomalies should be weighted by power decay and a natural logarithmic amplification should be performed in conjunction with the bottleneck boundary stagnation and collapse coefficient to form a cumulative anomaly energy that is more suitable for early warning comparison.

[0033] It should be noted that the time decay factor is applied as follows: weights are successively assigned decreasing weights from the first future time position to the manyth future time position, and these weights are directly multiplied into the corresponding predicted future anomaly intensity value. The cumulative anomaly energy is updated by rereading all predicted future anomaly intensities at each current time point and summing them again; the cumulative anomaly energy from the previous time point is not directly extrapolated. The time alignment between the cumulative anomaly energy and subsequent automatic warning values ​​for the entire region is as follows: the cumulative anomaly energy at the current time point only participates in the weighted summation of the structural segment allocation at the current time point, and is not used across time points.

[0034] In one embodiment of the present invention, structural segment allocation weights are determined based on the area of ​​sensitive ecological patches and bottleneck factors. The cumulative abnormal energy is then weighted and normalized using these structural segment allocation weights, and an automatic early warning value for the entire region is output, including: Utilizing structural segments Corresponding sensitive ecological patch area Bottleneck Factors Generate structural segments and assign weights The calculation formula is as follows: In the formula, Assign weights to the structural segments. For the area of ​​sensitive ecological patches, As a bottleneck factor, The total number of all structural segments. It is the sum of the products of the sensitive ecological patch area and the bottleneck factor corresponding to all structural segments. It is a very small positive number; Weight allocation using structural segments With cumulative abnormal energy Generate automatic early warning values ​​for the entire region The calculation formula is as follows: In the formula, This is the automatic early warning value for the entire region. For the structural segment numbering, To accumulate abnormal energy, For a moment The weighted sum of the weights assigned to all structural segments and the cumulative anomaly energy is the result of the weighted calculation of the structural segment assignments and the cumulative anomaly energy. As a historical reference window The maximum value of the weighted summation result corresponding to all structural segments within the structure. These are the scaling transformation coefficients. It is a very small positive number.

[0035] It should be noted that the sensitive ecological patch area refers to the area within the current structural segment that truly belongs to a sensitive ecological patch. The structural segment allocation weight is the contribution weight of the current structural segment in the overall automatic early warning value aggregation. The total number of all structural segments is the number of structural segments participating in the overall automatic early warning value calculation in the current round. The overall automatic early warning value is the unique output value obtained by weighting and normalizing the cumulative anomalies of all structural segments according to their allocated weights. The historical reference window is a custom time window used to extract the maximum value of the historical weighted summation result, preferably a 30- to 90-day historical window within the same season. This range balances seasonal comparability, sample stability, and extreme value representativeness. The scale transformation coefficient is a custom parameter used to map the overall automatic early warning value to a unified dimensional range, preferably 100, which facilitates the formation of a standardized early warning scale from 0 to 100.

[0036] It should be noted that since different structural segments contribute differently to the overall ecological risk of the region, it is necessary to simultaneously utilize the area of ​​sensitive ecological patches and bottleneck factors to generate structural segment allocation weights, in order to highlight structural segments that are both sensitive and critical. Because the weighted summation result at the current moment lacks a unified dimension and is difficult to use directly for cross-time period comparisons, it is necessary to normalize the weighted summation result at the current moment using the maximum value of the weighted summation result within the historical reference window in order to obtain stable and comparable automatic early warning values ​​for the entire region.

[0037] It should be noted that the area of ​​sensitive ecological patches is determined as follows: all sensitive ecological patches are extracted within each structural segment, and their areas are summed; only the area of ​​sensitive ecological patches located within the current structural segment is included in the result. The historical reference window is determined as follows: a fixed-length historical period is selected backward from the current time, ensuring that the samples within the window have the same seasonal attributes and calculation methods as the current time. The maximum value of the weighted summation within the historical reference window is updated as follows: whenever the current time advances by one time step, the start and end positions of the window are updated synchronously, and the maximum value of the weighted summation for all times within the window is recalculated. The interpretation of the automatic warning value for the entire region is as follows: a larger automatic warning value for the entire region indicates a higher risk of anomalies at the current time, while a smaller automatic warning value indicates a lower risk of anomalies at the current time.

[0038] It should be noted that the training samples of the spatiotemporal graph convolutional network are constructed using historical continuous time slices. This means that historical data across the entire region is first time-aligned at a uniform time step. At each time point, three types of input data are simultaneously generated: the first type is the dynamic adjacency matrix corresponding to that time point; the second type is the state data of all structural segments at that time point; and the third type is the bottleneck boundary stagnation and collapse coefficient calculated for that time point. In other words, the network uses a dynamic adjacency matrix to express the spatial connectivity between structural segments at each time step, recursively uses the hidden state of the previous time step of a structural segment and the hidden state of the previous layer of adjacent structural segments for state inference, and incorporates the bottleneck boundary stagnation and collapse coefficient into a gating mechanism. The supervision label is the true value of the anomaly intensity of the corresponding structural segment in the future time period. The true anomaly intensity value is calculated based on historical monitoring data from the same period and is completely consistent with the calculation caliber of the model's output future anomaly intensity prediction value. The time step of the label is consistent with the time step of the input sequence, and the length of the label sequence for a single sample is consistent with the upper limit of the future prediction step, covering all future time periods that need to be predicted after the end of the input sequence. Furthermore, all constructed samples can be divided into three parts in chronological order: a training set, a validation set, and a test set. The training set should account for no less than 70% of the total sample size and be used for iterative optimization of model parameters; the validation set should account for no less than 15% of the total sample size and be used for verifying model performance and adjusting hyperparameters during training; the test set should account for no more than 15% of the total sample size and be used for testing the model's generalization ability after training. During sample selection, samples with a data integrity rate of less than 80% are removed, and samples from periods of continuous calm without rainfall or abnormal fluctuations should account for no more than 30% of the total sample size to ensure that the sample set contains sufficient rainfall impacts and boundary state change scenarios, which will not be elaborated further here.

[0039] It should be noted that during the training of the spatiotemporal graph convolutional network, the mean squared error function can be specified as the loss function. The calculation object is the deviation between the model's predicted future anomaly intensity and the corresponding supervision label. The optimization objective during training is to minimize the value of the loss function. Furthermore, the optimizer uses the adaptive moment estimator (Adam), with an initial learning rate of 0.001, a weight decay coefficient of 0.0001, an exponential decay rate of 0.9 for the first moment estimation, and an exponential decay rate of 0.999 for the second moment estimation. The training batch size is set between 8 and 32, adjusted according to the total number of samples and computational resources. Each training round traverses all batches of samples in the training set, and the maximum number of iterations is set between 200 and 500, with a default of 300.

[0040] It should be noted that this invention relies on the urban ecological environment monitoring system and geographic information platform for deployment and implementation. Data collection covers three categories: spatial basic data, real-time monitoring data, and meteorological and hydrological data. Spatial basic data is acquired through satellite remote sensing, UAV aerial surveying, and urban basic geographic information databases, including the boundaries of urban ecologically sensitive areas, water catchment unit divisions, riverbank strip ranges, impermeability, land use types, and hydrological propagation channels. Real-time monitoring data is collected through water quality sensors, water level gauges, water temperature sensors, turbidity sensors, and conductivity sensors deployed in ecologically sensitive areas. Sensors upload data at a uniform time step, covering key monitoring points across all structural sections. Meteorological and hydrological data is collected through regional meteorological stations and rainfall monitoring stations, including hourly rainfall intensity, watershed water level, and runoff velocity. After all collected data is transmitted to the system server, it undergoes spatiotemporal alignment, outlier removal, missing value filling, and numerical normalization preprocessing in sequence, before being input into various modules of the system for continuous calculation at a fixed time step. The system supports two modes: scheduled automatic operation and condition-triggered operation. In scheduled mode, the entire process calculation is executed in a loop according to a preset step size. In condition-triggered mode, the calculation is started immediately in scenarios such as rainfall or rapid changes in water level, ensuring the efficiency of early warning response.

[0041] It should be noted that this invention ultimately outputs two distinct types of results: the cumulative anomaly energy of a single structural segment and the quantitative automatic early warning value for the entire region. The cumulative anomaly energy of a single structural segment is a non-negative real number, reflecting the overall anomaly load of a single structural segment within the prediction window. The quantitative automatic early warning value for the entire region is a standardized value from 0 to 100, and its magnitude is positively correlated with the degree of environmental anomaly risk. Taking a riverside wetland ecologically sensitive area in a certain city as an example, this area is divided into 26 structural segments. After a localized heavy rainfall, the system completed the entire process calculation. The cumulative anomaly energy of the bottleneck structural segment No. 12 was 13.2, and the weighted normalized early warning value for the entire region was 72. This result can intuitively show that the ecologically sensitive area as a whole is in a medium-to-high risk state, with structural segment No. 12 being a high-risk location, and anomalies easily spreading to surrounding areas along this segment. In addition, the system can simultaneously output a risk ranking list for each structural segment, providing clear guidance for on-site response.

[0042] Specifically, after the quantitative automatic early warning value is output for the entire region, it needs to be compared with the preset risk threshold. The threshold is determined based on the region's historical environmental anomalies, the control level of ecologically sensitive areas, and hydrological and meteorological patterns. The threshold divides the early warning value into three risk levels: 0 to 30 corresponds to low risk, 31 to 60 corresponds to medium risk, and 61 to 100 corresponds to high risk. After the threshold is set, it is verified using historical real anomaly cases to ensure that the threshold division matches the actual risk occurrence. The system performs threshold calibration and update once a year based on changes in the regional ecological boundary status, rainfall distribution patterns, and runoff conditions to maintain threshold adaptability. After the system completes the early warning value calculation, it automatically performs threshold comparison to determine the current risk level of the entire region. In the low-risk state, the regular monitoring frequency is maintained; in the medium-risk state, the frequency of data monitoring and patrol of structural sections is increased; in the high-risk state, the emergency response process for ecologically sensitive areas is immediately activated, and information on high-risk structural sections is pushed to the control terminal to guide on-site response work. Further details are omitted here.

[0043] It should be noted that the interval and threshold sizes are set for ease of comparison. The size of the threshold depends on the amount of sample data and the base number set by those skilled in the art for each set of sample data, as long as it does not affect the proportional relationship between the parameter and the quantized value. Furthermore, the above formulas are all dimensionless calculations, and the formulas are derived from software simulations using a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0044] The embodiments of this example have been described above. However, this example is not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms based on the guidance of this example, and all of them are within the protection scope of this example.

Claims

1. An automatic early warning system for environmental anomalies in urban ecologically sensitive areas based on spatiotemporal data mining, characterized in that, include: The first module involves finding the intersection of the total area of ​​the urban ecologically sensitive zone, the water catchment area, and the riverside or boundary strip area to obtain the structural segment, and extracting the intermediate center scalar of the structural segment; The second module calculates the runoff volume using rainfall intensity and impermeability, calculates the fragmented release volume using boundary length and effective boundary retention area, and calculates the rainfall concentration coefficient by combining local rainfall intensity. The third module transforms the intermediate center scalar into a bottleneck factor, obtains the effective residence time from the confluence flow and effective water depth, and generates the bottleneck boundary storage and collapse coefficient by combining the rainfall concentration coefficient, bottleneck factor, fragmented release volume and effective residence time. The fourth module obtains the hydraulic drawdown factor from the effective water depth difference, and integrates the hydraulic drawdown factor, hydrological path distance and bottleneck boundary retention collapse coefficient to generate a dynamic adjacency matrix. The fifth module inputs the dynamic adjacency matrix into the spatiotemporal graph convolutional network, incorporates the bottleneck boundary stagnation and collapse coefficient into the gating mechanism, and outputs the predicted value of future anomaly intensity. The sixth module amplifies and accumulates the predicted values ​​of future anomaly intensity by combining time decay and bottleneck boundary stagnation and collapse coefficients, generating the cumulative anomaly energy of the structural segment. The seventh module determines the structural segment allocation weights based on the area of ​​sensitive ecological patches and bottleneck factors, uses the structural segment allocation weights to weight and normalize the cumulative anomalies, and outputs the automatic early warning value for the entire area.

2. The automatic early warning system for environmental anomalies in urban ecologically sensitive areas based on spatiotemporal data mining according to claim 1, characterized in that, Perform spatial intersection operations on the total area of ​​the urban ecologically sensitive area, the water catchment area, and the riverside or boundary strip area to generate structural segments; Using all structural segments as the statistical unit, count the total number of shortest propagation paths between any source structural segment and the target structural segment, and count the number of paths that pass through the current structural segment in the shortest propagation path. Calculate the ratio of the number of paths passing through the current structural segment in the shortest propagation path to the total number of shortest propagation paths, and extract the intermediate center scalar by summing all ratios.

3. The automatic early warning system for environmental anomalies in urban ecologically sensitive areas based on spatiotemporal data mining according to claim 1, characterized in that, For the set of upstream water catchment units that flow into the structural segment, a flow weight is generated based on the impermeability and area of ​​each upstream water catchment unit. The confluence of the structural segment is generated by summing the confluence weight, the rainfall intensity of the upstream catchment unit, and the time delay of the upstream catchment unit propagating to the structural segment.

4. The automatic early warning system for environmental anomalies in urban ecologically sensitive areas based on spatiotemporal data mining according to claim 1, characterized in that, Fragmented amplification is generated by using the square relationship between the boundary length of the structural segment and the effective boundary storage area and its natural logarithm. The ratio of the local rainfall intensity of a structural segment to the average rainfall intensity within the corresponding neighborhood of the structural segment is calculated to generate the rainfall concentration factor.

5. The automatic early warning system for environmental anomalies in urban ecologically sensitive areas based on spatiotemporal data mining according to claim 1, characterized in that, By using the maximum and minimum values ​​of the intermediate center scalars corresponding to all structural segments, a normalization transformation is performed on the intermediate center scalars corresponding to the current structural segment to generate the bottleneck factor. The effective residence time is calculated by using the product of the effective boundary retention area and the effective water depth as the numerator and the product of the catchment flow and the total effective catchment area upstream as the denominator. The total effective catchment area upstream is the sum of the products of the impermeability and the area of ​​each upstream catchment unit. By integrating rainfall concentration coefficient, bottleneck factor, fragmented release volume, and standardized effective residence time, the bottleneck boundary retention and collapse coefficient is generated through ratio calculation.

6. The automatic early warning system for environmental anomalies in urban ecologically sensitive areas based on spatiotemporal data mining according to claim 1, characterized in that, For a structural segment and its adjacent structural segments, the hydraulic head drop factor from the structural segment to the adjacent structural segment is generated by utilizing the difference between the effective water depth of the structural segment and the effective water depth of the adjacent structural segment. By combining the hydrological path distance between the structural segment and the adjacent structural segment, the hydraulic drawdown factor, and the bottleneck boundary retention collapse coefficient, dynamic adjacency matrix elements pointing from the structural segment to the adjacent structural segment are generated in the dynamic adjacency matrix. Based on the adjacency relationships between structural segments, all elements of the dynamic adjacency matrix are combined to generate a dynamic adjacency matrix.

7. The automatic early warning system for environmental anomalies in urban ecologically sensitive areas based on spatiotemporal data mining according to claim 1, characterized in that, Based on the matrix elements in the dynamic adjacency matrix that point from the current structural segment to the adjacent structural segment, a weighted aggregation is performed on the hidden state of the adjacent structural segment in the previous layer to generate a spatial aggregation representation. The spatial aggregation representation is concatenated with the bottleneck boundary stagnation and collapse coefficient and then input into the gating mechanism. The gating coefficients are generated by the gating weight vector and the activation function. The fusion space aggregation representation is related to the hidden state of the current structural segment in the previous time step, and candidate hidden states are generated by updating the weight matrix over time. The candidate hidden state and the hidden state of the current structural segment in the previous time step are updated by gating coefficients to generate the hidden state of the current structural segment in the current time step. The output layer weight vector maps the hidden state of the final layer at future time points to predicted values ​​of future anomaly intensity.

8. The automatic early warning system for environmental anomalies in urban ecologically sensitive areas based on spatiotemporal data mining according to claim 1, characterized in that, A time decay factor is generated based on the upper limit of the future prediction step size; The time decay factor is used to perform power decay weighting on each future anomaly intensity prediction value within the future prediction step range, and the natural logarithmic amplification is performed using the bottleneck boundary stagnation collapse coefficient. The cumulative anomaly energy is generated by summing the full processing results.

9. The automatic early warning system for environmental anomalies in urban ecologically sensitive areas based on spatiotemporal data mining according to claim 1, characterized in that, The ratio is calculated by using the product of the sensitive ecological patch area corresponding to each structural segment and the bottleneck factor as the numerator, and the sum of the products of the sensitive ecological patch areas corresponding to all structural segments and the bottleneck factor as the denominator, thus generating the structural segment allocation weight.

10. The automatic early warning system for environmental anomalies in urban ecologically sensitive areas based on spatiotemporal data mining according to claim 9, characterized in that, By assigning weights to structural segments, a weighted summation operation is performed on the cumulative anomalies corresponding to each structural segment to obtain the weighted summation result at the current moment. The maximum value of the weighted summation result within the historical reference window is used to perform normalization processing on the weighted summation result at the current moment, generating an automatic early warning value for the entire region.

Images