CN-121998179-A - Regional hazardous waste risk pattern prediction method and system based on spatial clustering and time sequence evolution

CN121998179ACN 121998179 ACN121998179 ACN 121998179ACN-121998179-A

Abstract

The invention provides a regional hazardous waste risk pattern prediction method and system based on spatial clustering and time sequence evolution, and relates to the technical field of environmental big data. The method comprises the steps of obtaining multi-source business data to extract node yield load and storage allowance attributes and construct risk potential energy vectors, combining history transfer records to calculate topological association strength and form transfer path impedance coefficients, coupling potential energy descending quantity to obtain transfer resistance parameters and map the transfer resistance parameters to transfer probability to generate time sequence dynamic transfer tensors, dividing the transfer resistance parameters to obtain high-risk aggregation clusters, constructing risk conduction topology maps based on the centroid of the aggregation clusters, identifying key hinges and searching risk cascade key paths, extracting hot spot drift vectors, building a dual-drive evolution prediction model, and outputting regional risk pattern prediction results. The method solves the problems of insufficient modeling of the risk dynamic conduction mechanism, mutual fracture of space-time analysis dimensions and lack of physical network constraint of the prediction model in the prior art.

Inventors

WANG LIQUN
WANG YUTING
CHENG DANDAN
WU BEN
PANG WEILIANG
YANG JING
MENG JIANLI
TAN CUILING
ZHAO JIE
ZHANG WEI

Assignees

天津市生态环境科学研究院(天津市环境规划院、天津市低碳发展研究中心)

Dates

Publication Date: 20260508
Application Date: 20260116

Claims (10)

1. The regional hazardous waste risk pattern prediction method based on spatial clustering and time sequence evolution is characterized by comprising the following steps of: Acquiring multi-source service data of a target area, and extracting attributes of nodes in the target area based on the multi-source service data to obtain a node attribute set, wherein the nodes in the node attribute set have yield load attributes and storage allowance attributes; Calculating risk potential energy vectors of all the nodes according to the coupling relation between the yield load attribute and the storage allowance attribute of all the nodes in the node attribute set; Calculating transfer path impedance coefficients among nodes according to historical transfer records extracted from the multi-source service data and the risk potential energy vector; Determining a circulation resistance parameter among the nodes according to the risk potential energy vector and the transfer path impedance coefficient, and acquiring node geographic coordinates from the multi-source business data to construct regional space distance constraint; Generating a time sequence dynamic circulation tensor according to the regional space distance constraint and the circulation resistance parameter; Constructing a dangerous waste risk evolution space-time cube based on the time sequence dynamic flow tensor in a preset three-dimensional mapping domain, and carrying out voxel density segmentation on the dangerous waste risk evolution space-time cube to obtain a high-risk cluster; constructing a risk conduction topological graph based on the high-risk cluster, and analyzing the risk conduction topological graph by utilizing a network centrality algorithm to obtain a risk cascade key path; Sequentially tracking mass center displacement of the high-risk cluster on a time sequence according to the risk cascade critical path to extract a risk hot spot drift vector, and carrying out topology constraint correction and time sequence inertia extrapolation on the risk hot spot drift vector to obtain a dual-drive evolution prediction model; And carrying out evolution calculation on the regional risk according to the dual-drive evolution prediction model to obtain a regional dangerous waste risk pattern prediction result.
2. The regional hazardous waste risk pattern prediction method based on spatial clustering and time sequence evolution according to claim 1, wherein the obtaining multi-source service data of a target region, and extracting attributes of nodes in the target region based on the multi-source service data, to obtain a node attribute set, comprises: Performing field analysis and duplication removal processing on the multi-source service data to separate electronic transfer list data and enterprise management plan record data, and identifying a target node from the multi-source service data by utilizing an enterprise unique identification code; extracting waste generation time sequence records of the target node in a preset history period from the electronic transfer list data, and carrying out time attenuation weighted accumulation on the waste generation time sequence records to obtain the yield load attribute; Analyzing the upper limit value of the verification storage of the target node from the enterprise management plan record data, extracting the real-time stock quantity of the target node at the current moment, and calculating the difference value between the upper limit value of the verification storage and the real-time stock quantity to obtain the storage allowance attribute; And mapping the yield load attribute and the storage allowance attribute to corresponding target nodes to obtain the node attribute set.
3. The regional hazardous waste risk pattern prediction method based on spatial clustering and time-series evolution according to claim 1, wherein the calculation expression of the risk potential energy vector is: ; Wherein, the Indexing for the nodes; a node index for taking the maximum value of all nodes; Is a node Yield load attribute values of (2); Is a node A storage margin attribute value of (2); Is a node Risk potential energy vector of (a); Is a node Risk potential energy model length; To take the maximum value operation; An extremely small positive number for preventing the denominator from being zero; is a transposition operation; Is a two-norm operation.
4. The regional hazardous waste risk pattern prediction method based on spatial clustering and time-series evolution according to claim 1, wherein the calculating the transfer path impedance coefficient between the nodes according to the history transfer record extracted from the multi-source service data and the risk potential energy vector comprises: analyzing the history transfer record to obtain a topology incidence matrix, and performing reciprocal operation and normalization processing on the topology incidence matrix to obtain a basic channel resistance value representing service viscosity between nodes; Acquiring a risk potential energy vector of an output node serving as a transfer starting point and a risk potential energy vector of an input node serving as a transfer end point, and calculating a corresponding module length difference value and a direction included angle cosine value; Constructing an inverse repulsive force function of potential energy gradient according to the module length difference value and the cosine value of the direction included angle so as to calculate a potential energy resistance value; The basic channel resistance value and the potential energy resistance value are weighted and summed to obtain a transfer path impedance coefficient among the nodes; the expression of the reverse repulsive force function is as follows: ; Wherein, the Indexing for the nodes; Indexing for the nodes; a node index for taking the maximum value of all nodes; a node index for taking the maximum value of all nodes; For a node in a history transfer record To the node The number of times of transfer or the topology association strength formed by the accumulated amount of the orders; Is a node Is a transfer path impedance coefficient of (a); Is a node Risk potential energy vector of (a); Is a node Risk potential energy vector of (a); Is a node Risk potential energy model length; Is a node Max is the maximum value operation; An extremely small positive number for preventing the denominator from being zero; Is an exponential function; is a transposition operation; Is a two-norm operation.
5. The regional hazardous waste risk pattern prediction method based on spatial clustering and time-series evolution according to claim 1, wherein the determining the circulation resistance parameter between the nodes according to the risk potential energy vector and the transfer path impedance coefficient, and obtaining the node geographic coordinates from the multi-source service data to construct the regional spatial distance constraint comprises: analyzing longitude and latitude geographical coordinates of each node from the multi-source service data, and calculating spherical transmission distance of the longitude and latitude geographical coordinates between any two nodes to obtain a global distance matrix; Acquiring a preset dangerous waste transportation limit radius, performing space weighting calculation on the global distance matrix by using a Gaussian attenuation function to obtain a distance weight, and setting the distance weight exceeding the limit radius to zero to obtain the regional space distance constraint; Calculating gradient projection values of the risk potential energy vectors of the starting node and the ending node in the direction of the transfer path, and substituting the gradient projection values and the impedance coefficients of the transfer path into a preset circulation dynamics equation to carry out coupling solution; outputting quantized values through the flow dynamics equation to obtain flow resistance parameters among the nodes, wherein the flow resistance parameters are positively correlated with the transfer path impedance coefficients and negatively correlated with the gradient projection values; the computational expression of the flow dynamics equation is as follows: ; Wherein, the Indexing for the nodes; Indexing for the nodes; Is a node Potential energy decrease amount of (2); Is a node Risk potential energy model length; Is a node Risk potential energy model length; Is a node Is a transfer path impedance coefficient of (a); Is a node Max is the maximum value operation; To prevent the denominator from being a very small positive number of zero.
6. The regional hazardous waste risk pattern prediction method based on spatial clustering and time-series evolution according to claim 1, wherein the generating a time-series dynamic circulation tensor according to the regional spatial distance constraint and the circulation resistance parameter comprises: Constructing a resistance-probability mapping function, inputting the circulation resistance parameter into the resistance-probability mapping function to carry out negative exponential decay operation, and obtaining a basic circulation probability matrix for representing potential interaction possibility between nodes in an unconstrained state; Performing matrix Hadamard product operation on the basic circulation probability matrix and the regional space distance constraint to obtain a space correction circulation matrix; Setting a time window length and a time step length, and constructing a three-dimensional tensor structure comprising a starting node dimension, a terminating node dimension and a time dimension according to the time window length and the time step length; Mapping the space correction circulation matrix as a reference slice to each time step of the three-dimensional tensor structure, and dynamically modulating the slice values of each time step by the change rate of the node attribute set at different moments to obtain the time sequence dynamic circulation tensor; The drag-probability mapping function has the expression: ; Wherein, the Indexing for the nodes; Indexing for the nodes; For nodes without space distance constraint Is a basic circulation probability of (1); Is a node A flow resistance parameter of (2); is an exponential function.
7. The regional hazardous waste risk pattern prediction method based on spatial clustering and time sequence evolution according to claim 1, wherein the constructing a hazardous waste risk evolution space-time cube based on the time sequence dynamic flow tensor in a preset three-dimensional mapping domain, and performing voxel density segmentation on the hazardous waste risk evolution space-time cube to obtain a high risk aggregation cluster comprises: Constructing a three-dimensional mapping domain taking geographic longitude and latitude as a plane coordinate axis and taking a time sequence as a vertical coordinate axis, and dividing the three-dimensional mapping domain into standard voxel units with fixed length, width and height dimensions; resolving a circulation intensity value in the time sequence dynamic circulation tensor, and projecting the circulation intensity value as a weight to a space coordinate point corresponding to the three-dimensional mapping domain; Performing smooth interpolation calculation on all the standard voxel units by adopting a space-time kernel density estimation algorithm to obtain a risk density value of each standard voxel unit so as to construct a space-time cube for risk evolution of the hazardous waste; Calculating the global background noise threshold value of the dangerous waste risk evolution space-time cube by adopting a self-adaptive threshold value segmentation algorithm, and removing standard voxel units with the risk density value lower than the global background noise threshold value to obtain a high risk retention voxel set; And executing three-dimensional connected domain marking on the high-risk reserved voxel set so as to combine standard voxel units with adjacent spatial positions and continuous risk density values into an independent connected region, thereby obtaining the high-risk cluster.
8. The regional hazardous waste risk pattern prediction method based on spatial clustering and time sequence evolution according to claim 1, wherein the method is characterized by constructing a risk conduction topological graph based on the high-risk clustering, analyzing the risk conduction topological graph by using a network centrality algorithm to obtain a risk cascade critical path, and comprises the following steps: Extracting the geometric centroid of each high-risk cluster as a network topology node, and calculating a circulation flux accumulation value between each network topology node based on the time sequence dynamic circulation tensor; Normalizing the circulation flux accumulated value to obtain a risk transmission probability matrix; Constructing the risk conduction topological graph, wherein the top point of the risk conduction topological graph is a network topological node, and the directed edge of the risk conduction topological graph is a non-zero element in the risk transmission probability matrix; And calculating the intermediation centrality of each network topology node in the risk conduction topology map by using a weighted betweenness centrality algorithm, and identifying the node with the intermediation centrality higher than a preset threshold value. Obtaining a key pivot node; performing maximum probability gradient search in the risk conduction topological graph by taking the key pivot node as a starting point, and sequentially connecting downstream nodes along the direction of slowest decrease of the risk transmission probability matrix value to obtain a potential risk chain; And performing closed loop detection and redundant pruning on the potential risk chain to obtain the risk cascade critical path.
9. The regional hazardous waste risk pattern prediction method based on spatial clustering and time sequence evolution according to claim 1, wherein the sequentially tracking centroid displacement of the high-risk cluster on a time sequence according to the risk cascade critical path to extract a risk hot spot drift vector, and performing topology constraint correction and time sequence inertia extrapolation on the risk hot spot drift vector to obtain a dual-drive evolution prediction model comprises the following steps: extracting the space cross sections of the high risk clusters at each moment, calculating the risk weighted centroids of the space cross sections by using a density weighting algorithm, and connecting the risk weighted centroids of adjacent time steps to obtain an original centroid displacement vector; matching the original centroid displacement vector with a risk cascade key path in the risk conduction topological graph to obtain the risk hot spot drift vector; acquiring the outcoming adjacent edges of the risk hot spot drift vector on the risk conduction topological graph, and calculating cosine similarity of the risk hot spot drift vector and each outcoming adjacent edge to determine a guiding constraint edge; projecting the risk hot spot drift vector to the guiding constraint edge to obtain a spatial topology correction component; constructing a time sequence sliding window, intercepting a risk hot spot drift vector sequence at a historical moment, inputting the risk hot spot drift vector sequence into a pre-trained long-term and short-term memory network, and extracting time sequence characteristics to obtain a time sequence inertia extrapolation component; Performing data assimilation and state estimation on the spatial topology correction component and the time sequence inertia extrapolation component by adopting an adaptive Kalman filtering algorithm to generate the dual-drive evolution prediction model; the expression of the dual-drive evolution prediction model is as follows: ; Wherein, the Is a time step index; Is a time step Hot spot drift vector of (2); Is a time step Is a time-series inertial extrapolation component of (a); Is a time step Risk weighted centroid coordinate vectors of high risk clusters; Is a time step Risk weighted centroid coordinate vectors of high risk clusters; Is a time step Guiding a unit direction vector of the constraint edge; Is a time step A posterior state covariance matrix of (2); a process noise covariance matrix; The covariance matrix is observed; is a transposition operation; and (5) inverting the matrix.
10. Regional hazardous waste risk pattern prediction system based on spatial clustering and time sequence evolution is characterized by comprising: The data acquisition and attribute extraction module is used for acquiring multi-source service data of a target area, and extracting attributes of nodes in the target area based on the multi-source service data to obtain a node attribute set, wherein the nodes in the node attribute set have yield load attributes and storage allowance attributes; The risk potential energy vector calculation module is used for calculating the risk potential energy vector of each node according to the coupling relation between the yield load attribute and the storage margin attribute of each node in the node attribute set; The transfer path impedance calculation module is used for calculating transfer path impedance coefficients among the nodes according to the historical transfer records extracted from the multi-source service data and the risk potential energy vector; The circulation resistance and space constraint construction module is used for determining circulation resistance parameters among the nodes according to the risk potential energy vector and the transfer path impedance coefficient, and acquiring node geographic coordinates from the multi-source business data to construct regional space distance constraint; The time sequence dynamic flow tensor generation module is used for generating a time sequence dynamic flow tensor according to the regional space distance constraint and the flow resistance parameter; the space-time cube construction and clustering module is used for constructing a dangerous waste risk evolution space-time cube based on the time sequence dynamic flow tensor in a preset three-dimensional mapping domain, and carrying out voxel density segmentation on the dangerous waste risk evolution space-time cube to obtain a high-risk aggregation cluster; The risk topology analysis and path identification module is used for constructing a risk conduction topology map based on the high-risk aggregation cluster, and analyzing the risk conduction topology map by utilizing a network centrality algorithm to obtain a risk cascade key path; The dual-drive evolution prediction model construction module is used for tracking centroid displacement of the high-risk cluster on a time sequence according to the risk cascade critical path in sequence to extract a risk hot spot drift vector, and carrying out topology constraint correction and time sequence inertia extrapolation on the risk hot spot drift vector to obtain a dual-drive evolution prediction model; and the regional risk pattern prediction module is used for carrying out evolution calculation on the regional risk according to the dual-drive evolution prediction model to obtain a regional dangerous waste risk pattern prediction result.

Description

Regional hazardous waste risk pattern prediction method and system based on spatial clustering and time sequence evolution Technical Field The invention relates to the technical field of environmental big data, in particular to a regional hazardous waste risk pattern prediction method and system based on spatial clustering and time sequence evolution. Background Along with the acceleration of the industrialized process, the generation amount of dangerous waste rises year by year, and the prevention and control of the environmental risk become the key points of regional environmental management. At present, hazardous waste supervision mainly relies on the combination of an electronic coupon system and a geographic information system, and informationized recording and inquiring of the full life cycle of hazardous waste are realized by collecting structured data such as declaration data, transfer coupons, treatment records and the like of enterprises. The existing supervision platform can generally display basic position information of enterprises, and utilizes a statistical method to collect and count the generation amount and the disposal amount of dangerous wastes in each administrative area, and generates a static thermodynamic diagram or a statistical report, so that point-to-point flow direction monitoring and post-tracing are realized. In order to improve the initiative and predictability of supervision, recent technical trends begin to develop toward big data fusion and intelligent prediction. Researchers have attempted to introduce complex network theory to analyze the structural characteristics of hazardous waste stream networks, identifying key nodes in the network, such as large disposal centers. Meanwhile, a machine learning algorithm, such as a random forest, linear regression or long-short-term memory neural network, is combined, and the future dangerous waste production is predicted based on historical data. Furthermore, with the development of digital twinning technology, a visual presentation based on one graph of a geographic information system has become industry standard, aiming at assisting regulatory decisions through multi-scale spatial presentation. However, existing regional hazardous waste risk prediction methods still have significant limitations in practical applications. First, there is a lack of deep modeling of risk dynamic conduction mechanisms. In the prior art, the dangerous waste network is regarded as static physical connection, and risk overflow and cascading effect between nodes caused by factors such as saturated handling capacity and blocked transportation are ignored. When a core disposal unit fails, risks are rapidly backlogged upstream to the waste production enterprise along the network, and conventional statistical models cannot capture such nonlinear dynamics based on network topology. Second, the space-time analysis dimensions fracture each other. The existing method generally uses spatial clustering to identify the current hot spot, uses time sequence prediction to predict total amount trend, and performs the two steps as two independent steps, and lacks a unified space-time coupling analysis framework. This makes it difficult for the model to accurately predict the continuous migration path of high risk areas over geospatial, and it is not possible to explain how risk hotspots flow from one area to another under time-shifting. Third, the predicted outcome lacks physical constraints. The pure data-driven algorithm is often used for carrying out numerical extrapolation based on a historical curve, capacity constraint of a real physical network, such as storage capacity upper limit and trans-regional transfer limit, are not fully considered, so that the phenomenon that a predicted result is accurate in numerical value but logic is contrary to normal is possible, and actual resource scheduling and risk blocking deployment are difficult to guide. Disclosure of Invention In order to overcome the defects of the prior art, the invention aims to provide a regional hazardous waste risk pattern prediction method and system based on spatial clustering and time sequence evolution, and solves the problems of insufficient modeling of a risk dynamic conduction mechanism, mutual cleavage of space-time analysis dimensions and lack of physical network constraint of a prediction model in the prior art. In order to achieve the above object, the present invention provides the following solutions: A regional hazardous waste risk pattern prediction method based on spatial clustering and time sequence evolution comprises the following steps: Acquiring multi-source service data of a target area, and extracting attributes of nodes in the target area based on the multi-source service data to obtain a node attribute set, wherein the nodes in the node attribute set have yield load attributes and storage allowance attributes; Calculating risk potential energy vectors of all the nodes accordi