Environmental protection monitoring method and system based on Internet of Things

Abstract

The invention discloses an environment protection monitoring method and system based on the Internet of Things, and relates to the technical field of environment monitoring of the Internet of Things, and the method comprises the steps: building a joint optimization model based on a directed causal network diagram and a constraint function, synchronously training the edge weight parameters of the joint optimization model through an alternating direction multiplier method, and obtaining a dynamic constraint coefficient; and inputting the real-time standardized time-space sequence data into the joint optimization model, extracting causal features through a dynamic constraint coefficient, calculating pollution contribution degree scores of each monitoring point, and generating a dynamic treatment optimization instruction set according to contribution degree sorting. According to the method, a data-driven initial causal network is generated through a PC algorithm, a correction direction is matched in combination with a rule of a Neo4j knowledge graph, redundant paths are eliminated, a process mechanism is fused, the problem of misjudgment of a causal direction in a complex scene is solved, and a directed network with high credibility is generated.

Description

Technical Field

[0001] The present invention relates to the technical field of Internet of Things environmental monitoring, and in particular to an Internet of Things-based environmental protection monitoring method and system. Background Art

[0002] Currently, IoT-based environmental monitoring technologies are gradually being applied to industrial pollution control and ecological regulation. Conventional methods often use distributed sensor networks to collect environmental parameters in real time, generate spatiotemporal correlation datasets through data fusion techniques, and employ statistical analysis models to locate pollution sources and predict their spread. At the causal inference level, existing solutions typically construct pollution propagation paths based on time series correlations, combine expert experience or rule engines to set threshold conditions, and generate remediation recommendations. These methods have achieved considerable progress in data collection density, real-time response speed, and rule matching efficiency, providing fundamental technical support for environmental monitoring.

[0003] However, existing methods still have room for improvement in causal modeling and dynamic constraint processing. Traditional statistical models focus on correlation analysis between variables, making it difficult to distinguish causal directions from spurious associations, resulting in pollution source tracing results that may deviate from the true physical mechanisms. Currently, IoT-based environmental monitoring systems have yet to develop effective solutions for robust causal reasoning and adaptability to dynamic constraints. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides an environmental protection monitoring method based on the Internet of Things to solve the problems of insufficient credibility of pollution tracing and poor dynamic adaptability of governance strategies in the existing technology.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: In the first aspect, the present invention provides an environmental protection monitoring method based on the Internet of Things, which includes: real-time collection of environmental parameters and preprocessing to generate standardized spatiotemporal series data; based on the standardized spatiotemporal series data, generating an initial causal network through the PC algorithm, and combining the rule matching verification of the Neo4j knowledge graph, marking the causal relationship and generating a directed causal network diagram; obtaining the process safety threshold, using the ReLU function and the Sigmoid function to perform derivable processing on the process safety threshold, and generating a constraint function in combination with the directed causal network diagram; constructing a joint optimization model based on the directed causal network diagram and the constraint function, synchronously training the edge weight parameters of the joint optimization model through the alternating direction multiplier method, and obtaining the dynamic constraint coefficient; inputting the real-time standardized spatiotemporal series data into the joint optimization model, extracting the causal characteristics through the dynamic constraint coefficient and calculating the pollution contribution score of each monitoring point, and generating a dynamic governance optimization instruction set based on the contribution ranking.

[0007] As a preferred solution of the Internet of Things-based environmental protection monitoring method described in the present invention, the preprocessing includes timestamp synchronization, spatial interpolation completion, denoising and data standardization processing.

[0008] As a preferred solution of the environmental protection monitoring method based on the Internet of Things of the present invention, the following specific steps are taken: based on the standardized spatiotemporal sequence data, the initial causal network is generated by the PC algorithm, and the rule matching verification is combined with the Neo4j knowledge graph to mark the causal relationship and generate a directed causal network diagram. Based on the standardized spatiotemporal series data, the PC algorithm is used to perform a conditional independence test, and the causal direction hypothesis of the test results is verified to generate the initial causal network structure; Based on the initial causal network structure, Neo4j is used to perform rule matching and causal path directionality correction on the knowledge graph storage to generate a causal network diagram with direction identification. Based on the causal network graph with direction identification, a directed causal network graph is generated through the feedback edge set optimization algorithm.

[0009] As a preferred solution of the environmental protection monitoring method based on the Internet of Things of the present invention, wherein: the process safety threshold is obtained, the process safety threshold is subjected to derivable processing using the ReLU function and the Sigmoid function, and a constraint function is generated in combination with a directed causal network graph. The specific steps are as follows: Extract the process safety threshold, perform piecewise derivable transformation through ReLU function and Sigmoid function, and generate a threshold function group; Based on the directed causal network graph, the threshold function group is multi-level topologically weighted fused through the graph constraint propagation algorithm to generate the constraint function.

[0010] As a preferred solution of the environmental protection monitoring method based on the Internet of Things of the present invention, wherein: the joint optimization model is constructed based on the directed causal network graph and the constraint function, and the specific steps are as follows: Based on the directed causal network graph, the causal network edge connection direction and edge weight parameters between nodes are initialized through the feedback edge set optimization algorithm; Based on the edge connection direction and edge weight parameters of the causal network, a multi-level constraint fusion algorithm is used to bind the constraint functions and establish a time-varying coupling relationship between the network edge weight parameters and the dynamic constraint coefficients. The alternating direction multiplier method is then combined to fuse the fitting error and constraint penalty to form a joint optimization model.

[0011] As a preferred solution of the environmental protection monitoring method based on the Internet of Things described in the present invention, wherein: the edge weight parameters of the joint optimization model are synchronously trained by the alternating direction multiplier method, and the dynamic constraint coefficient is obtained. The specific steps are as follows: The path integral tracing algorithm within the time window is used to perform multi-scale spatiotemporal correlation analysis on the temporal cumulative effects of the causal path, generating the initial iterative values ​​of the edge weight parameters. Based on the initial iterative values ​​of the edge weight parameters, a bivariate interaction optimization framework containing multiplier terms is generated through the alternating direction multiplier decomposition method. Based on the bivariate interactive optimization framework, the updated edge weight parameters are generated through the gradient backpropagation algorithm; Based on the updated edge weight parameters, dynamic constraint coefficients are generated through the proximal gradient algorithm and the residual convergence judgment method.

[0012] As a preferred solution of the Internet of Things-based environmental protection monitoring method described in the present invention, the following specific steps are used: inputting real-time standardized spatiotemporal series data into a joint optimization model, extracting causal characteristics through dynamic constraint coefficients, calculating the pollution contribution score of each monitoring point, and generating a dynamic governance optimization instruction set based on the contribution ranking. The real-time standardized spatiotemporal series data are input into the joint optimization model. Based on the time-varying coupling relationship between the network edge weight parameters and the dynamic constraint coefficients, the dynamic causal correlation characteristics of each monitoring node are extracted through the multi-path integral tracking algorithm. According to the dynamic causal association characteristics, the constrained spatiotemporal integral accumulation is performed along the hierarchical edge direction of the directed causal network graph, and the nonlinear modulation of the pollution propagation path intensity by the dynamic constraint coefficient is synchronously integrated to generate the pollution contribution score of each monitoring point. ; Based on the pollution contribution score, a descending list of pollution contribution scores of monitoring points is generated through a heap sort algorithm, and the priority levels are divided based on hash mapping; Based on the priority level and pollution contribution score of each monitoring point, the control intensity parameters of each monitoring point are calculated through a dynamic weight adjustment algorithm combined with real-time process safety thresholds, and a dynamic governance optimization instruction set is generated using linear programming; Environmental governance actions are driven by the dynamic control parameters in the dynamic governance optimization instruction set, and the post-governed environmental data is simultaneously collected to verify the effectiveness of the dynamic control parameters and fed back to the joint optimization model for iterative weight updates.

[0013] In the second aspect, the present invention provides an environmental protection monitoring system based on the Internet of Things, including a preprocessing module, a causal network construction module, a constraint function generation module, a joint optimization model construction module, and a dynamic decision instruction generation module; the preprocessing module collects environmental parameters in real time and performs preprocessing to generate standardized spatiotemporal sequence data; the causal network construction module generates an initial causal network based on the standardized spatiotemporal sequence data through the PC algorithm, and combines the rule matching verification of the Neo4j knowledge graph to mark the causal relationship and generate a directed causal network graph; the constraint function generation module obtains the process safety threshold, uses the ReLU function and the Sigmoid function to perform derivable processing on the process safety threshold, and generates the constraint function in combination with the directed causal network graph; the joint optimization model construction module constructs a joint optimization model based on the directed causal network graph and the constraint function, synchronously trains the edge weight parameters of the joint optimization model through the alternating direction multiplier method, and obtains the dynamic constraint coefficient; the dynamic decision instruction generation module inputs the real-time standardized spatiotemporal sequence data into the joint optimization model, extracts the causal characteristics through the dynamic constraint coefficient and calculates the pollution contribution score of each monitoring point, and generates a dynamic governance optimization instruction set based on the contribution ranking.

[0014] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the environmental protection monitoring method based on the Internet of Things as described in the first aspect of the present invention is implemented.

[0015] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, any step of the environmental protection monitoring method based on the Internet of Things as described in the first aspect of the present invention is implemented.

[0016] The beneficial effects of the present invention are as follows: a data-driven initial causal network is generated through the PC algorithm, and the rule matching correction direction is combined with the Neo4j knowledge graph to eliminate redundant paths and integrate process mechanisms, thereby solving the problem of misjudgment of causal direction in complex scenarios and generating a high-confidence directed network; discrete process thresholds are converted into continuous and differentiable constraint functions through the ReLU / Sigmoid function, and hierarchical dynamic constraint fusion is achieved through the graph constraint propagation algorithm, so that the optimization model strictly follows safety rules. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 The figure is a flow chart of the environmental protection monitoring method based on the Internet of Things.

[0019] Figure 2 Schematic diagram of the environmental protection monitoring system based on the Internet of Things.

[0020] Figure 3 Flowchart of process safety threshold processing.

[0021] Figure 4 Flowchart constructed for the joint optimization model. DETAILED DESCRIPTION

[0022] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0023] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0025] Reference Figures 1 to 4 , is an embodiment of the present invention, which provides an environmental protection monitoring method based on the Internet of Things, comprising the following steps: S1, real-time collection of environmental parameters and preprocessing to generate standardized spatiotemporal series data; Preprocessing includes timestamp synchronization, spatial interpolation completion, denoising and data normalization.

[0026] It should be noted that environmental parameters, including gas concentration data, temperature and humidity data, and particulate matter distribution data, are collected through multi-source sensing terminals; timestamp synchronization is based on the timestamp tags of gas concentration data, temperature and humidity data, and particulate matter distribution data, and the time phase difference of different sensing terminals is eliminated through the time interpolation algorithm to generate standardized time series data with time axis alignment; spatial interpolation completion is based on the geographic coordinate information of missing data, and the data distribution characteristics of neighboring sensing nodes are weightedly fitted through the spatial interpolation algorithm to generate complete spatially continuous geographic coordinate data; denoising uses wavelet transform decomposition and reconstruction technology to filter out high-frequency noise components in time series data and retain low-frequency effective signal components; data standardization processing maps gas concentration data, temperature and humidity data, and particulate matter distribution data of different dimensions to a unified numerical interval through the mean-variance normalization method to generate standardized spatiotemporal series data.

[0027] S2. Based on the standardized spatiotemporal sequence data, the PC algorithm is used to generate an initial causal network. Combined with the rule matching verification of the Neo4j knowledge graph, the causal relationships are annotated and a directed causal network diagram is generated. Based on the standardized spatiotemporal series data, the PC algorithm is used to perform a conditional independence test, and the causal direction hypothesis of the test results is verified to generate the initial causal network structure; It should be noted that first, a fully connected undirected graph containing all environmental parameter nodes is initialized, and the order of the conditional variable set is set to zero and gradually increased; at each order, a partial correlation coefficient test is performed for each pair of node combinations given the current conditional variable set. The test result is the partial correlation coefficient p value of each pair of nodes under a specific conditional variable set. When the p value exceeds the conditional independence judgment threshold in the general standard of statistical hypothesis testing (the p value critical value for judging the conditional independence of node pairs in statistical hypothesis testing), the node pairs are judged to be conditionally independent and the corresponding edges are removed. The iteration is carried out until no new edges are deleted, and an undirected causal network skeleton is generated; then the causal directionality hypothesis verification is based on the standard The timestamp continuity of the standardized spatiotemporal series data is calculated, the lag cross coefficient is calculated through the cross-correlation function in time series analysis, and the time delay characteristics of information transmission between nodes are determined. The Pearson time-lag cross-correlation coefficient of the standardized spatiotemporal series data is calculated and the maximum absolute value lag order is extracted. The lag correlation is judged by combining the p-value lower than the conditional independence judgment threshold in the statistical significance verification, and the direction from the first node to the later node is assigned to the edges with lag correlation. At the same time, the direction of the remaining edges without time delay is inferred by combining the directed separation rule in causal discovery theory, and finally an initial causal network structure containing directional causal relationships is generated.

[0028] Based on the initial causal network structure, Neo4j is used to perform rule matching and causal path directionality correction on the knowledge graph storage to generate a causal network diagram with direction identification. It should be noted that based on the initial causal network structure, Neo4j creates a node and relationship database to store the environmental parameter nodes and causal edges in the initial causal network structure. The node types include gas concentration nodes, particulate matter nodes, and water quality nodes. The edge type is a causal link with an associated direction attribute. Rule matching uses Neo4j's Cypher query language to retrieve environmental protection regulations and process flow heuristic rules in the knowledge graph, and uses regular expressions to match the causal relationship pattern in the process mechanism description. The constraint condition is "If the process parameter change of node A in the rule base logically causes the pollution index of node B to fluctuate, then the edge direction from node A to node B should be consistent with the rule." Causal path directionality correction traverses the directed edges in the initial causal network, and reverses the direction of conflicting edges or deletes redundant edges based on the rule matching results. For example, when the rule base clearly states that "an increase in reactor temperature will inevitably lead to an increase in VOCs concentration", the edge direction consistency from the temperature node to the concentration node is forced to be corrected, and finally a directed causal network graph containing a rule verification mark is output.

[0029] Based on the causal network graph with direction identification, a directed causal network graph is generated through the feedback edge set optimization algorithm.

[0030] It should be noted that the causal network diagram is first detected for directional conflicting edges (such as bidirectional edges or loop paths) and redundant paths (such as the parallel paths of node A→node B→node C and node A→node C). For directional conflicting edges, the directional weights of the conflicting edges are compared based on the edge directional attributes verified by Neo4j knowledge graph rule matching and the logical constraints generated by the process safety threshold conversion (such as the concentration limit defined in the equipment operating manual). Causal links that are consistent with the knowledge graph rules and have higher edge weight parameters (derived from the causal cumulative strength calculated by the path integral tracing algorithm) are preferentially retained. For redundant paths, the path integral tracing algorithm is used to calculate the cumulative impact strength of multi-hop paths in the standardized spatiotemporal series data (such as the path strength of temperature node→humidity node→concentration node). If the causal cumulative strength of the direct path (temperature node→concentration node) is higher than that of the indirect path, or if the indirect path violates the logical rules of the process safety threshold conversion (such as the concentration node value exceeds the limit defined in the emission standard document), the redundant path is deleted. Finally, the processing results of the conflicting edges and redundant paths are integrated to generate a directed causal network diagram with a streamlined topology and consistent directional logic. Among them, the logical rules for process safety threshold conversion are derived from the concentration limits defined in the equipment operation manual and emission standard documents.

[0031] S3. Obtain the process safety threshold, use the ReLU function and the Sigmoid function to make it derivable, and generate a constraint function in combination with the directed causal network graph; Extract the process safety threshold, perform piecewise derivable transformation through ReLU function and Sigmoid function, and generate a threshold function group; It should be noted that the process safety threshold is derived from the structured analysis of process texts such as equipment operation manuals and emission standard documents. Numerical process safety threshold parameters (such as reactor temperature threshold, gas concentration emission limit, and particulate matter concentration threshold) are extracted through regular expression matching, and the parameter names are strictly aligned with the node names in the standardized spatiotemporal series data; the ReLU function generates a piecewise linear differentiable function for the numerical threshold, such as the temperature threshold is converted into a constraint condition that generates linear growth when the process safety threshold parameter is exceeded; the Sigmoid function generates a smooth differentiable function for the logical threshold, such as the concentration change trend threshold is converted into a probabilistic transition constraint; the piecewise differentiable transformation converts the discrete process safety threshold into a continuously differentiable threshold function group.

[0032] Based on the directed causal network graph, the threshold function group is multi-level topologically weighted fused through the graph constraint propagation algorithm to generate the constraint function.

[0033] It should be noted that the environmental parameter nodes (such as temperature nodes and concentration nodes) in the directed causal network graph are traversed, and the ReLU function and Sigmoid function in the threshold function group are bound to the corresponding nodes respectively; hierarchical constraint propagation is performed along the causal network path direction (such as node A→node B→node C), and the first layer starts from the root node (such as the temperature node), and the ReLU function output of the root node is used as the initial constraint value, and the threshold function of the downstream node (such as the concentration node) is weighted superimposed according to the edge weight parameter; the second and subsequent layers are based on the constraint propagation results of the previous nodes, and the constraint function output of the parent node is nonlinearly multiplied with the Sigmoid function of the child node to generate a composite constraint value; in the weighted fusion process, the edge weight parameter dynamically adjusts the constraint contribution of different paths through the weight attribute in the causal network graph. For example, when the edge weight of node A→node B is 0.8, the ReLU function constraint value of node A is transferred to the Sigmoid function constraint calculation of node B with a coefficient of 0.8; finally, after completing multi-level fusion along all causal paths, a global constraint function is generated.

[0034] S4. Construct a joint optimization model based on the directed causal network graph and the constraint function, synchronously train the edge weight parameters of the joint optimization model through the alternating direction multiplier method, and obtain the dynamic constraint coefficient; Based on the directed causal network graph, the causal network edge connection direction and edge weight parameters between nodes are initialized through the feedback edge set optimization algorithm; It should be noted that the causal links in the directed causal network graph are traversed to detect edges with direction conflicts (such as bidirectional edges or loop paths). According to the logical constraints generated by the knowledge graph rule matching verification results and the process safety threshold conversion, the edge directions that are consistent with the causal relationship pattern in the rule base are retained first; for redundant paths, the cumulative causal influence intensity of different paths is calculated through the path integral tracing algorithm, and redundant edges with low weight or violating the threshold constraint logic are deleted; the edge weight parameter is initialized based on the lagged cross-correlation coefficient of the standardized spatiotemporal series data and the causal propagation intensity generated by the path integral tracing algorithm, and the lagged correlation coefficient between nodes in the historical data and the path integral result are weighted and summed to generate the edge weight parameter.

[0035] Based on the edge connection direction and edge weight parameters of the causal network, a multi-level constraint fusion algorithm is used to bind the constraint functions and establish a time-varying coupling relationship between the network edge weight parameters and the dynamic constraint coefficients. The alternating direction multiplier method is then used to fuse the fitting error and constraint penalty to form a joint optimization model. It should be noted that along the hierarchical topological path of the directed causal network graph, the ReLU function and the Sigmoid function in the threshold function group are bound at the node level, the ReLU function output of the root node is used as the initial constraint value, and the Sigmoid function of the child node is nonlinearly modulated based on the constraint value of the parent node; the edge weight parameter extracts the causal accumulation strength within the time window (such as the historical propagation strength from the temperature node to the concentration node) through the path integral tracking algorithm, and the dynamic constraint coefficient responds to real-time data fluctuations (such as the current concentration exceeding the standard) through the proximal gradient algorithm. The two form a closed-loop feedback through the gradient backpropagation mechanism, so that the adjustment of the edge weight parameter drives the adaptive update of the dynamic constraint coefficient. At the same time, the dynamic constraint coefficient imposes a reverse constraint on the optimization direction of the edge weight parameter to establish a time-varying coupling relationship. The time-varying coupling relationship refers to the mutual dependence mechanism between the edge weight parameter and the dynamic constraint coefficient that is dynamically adjusted over time. For example, when the edge weight parameter of the temperature node is updated based on the historical causal accumulation strength, the dynamic constraint coefficient synchronously reversely corrects the optimization direction of the edge weight parameter according to the current concentration exceeding the standard. The alternating direction multiplier method decomposes the joint optimization problem into two sub-problems: edge weight parameter optimization and dynamic constraint coefficient optimization. The edge weight parameters (based on minimizing the fitting error of standardized spatiotemporal series data) and the dynamic constraint coefficients (based on minimizing the threshold function constraint penalty term) are updated alternately and iteratively, and finally merged into a joint optimization model.

[0036] The path integral tracing algorithm within the time window is used to perform multi-scale spatiotemporal correlation analysis on the temporal cumulative effects of the causal path, generating the initial iterative values ​​of the edge weight parameters. Based on the initial iterative values ​​of the edge weight parameters, a bivariate interaction optimization framework containing multiplier terms is generated through the alternating direction multiplier decomposition method. It should be noted that within the sliding time window, along the hierarchical path direction of the directed causal network graph, for example: temperature node → concentration node → emission node, the path integral operation is used to calculate the cumulative intensity of causal influence at different time scales, for example, the hourly window captures short-term fluctuation correlation, and the daily window captures steady-state propagation correlation; the cumulative intensity calculation adopts the weighted integral method, and the real-time values ​​of the nodes and the edge weight parameters in the standardized spatiotemporal series data are integrated and accumulated according to the time dimension to generate the initial iterative value of the edge weight parameter reflecting the historical causal intensity; the alternating direction multiplier decomposition method splits the edge weight parameter optimization problem and the dynamic constraint coefficient optimization problem into a two-variable interaction framework, in which the edge weight parameter update subproblem constructs a loss function based on the path integral tracking result and the fitting error term, and the dynamic constraint coefficient update subproblem constructs a regularization term based on the constraint function penalty term and the multiplier term; the two-variable interactive optimization framework solves the edge weight parameter (fixed dynamic constraint coefficient) and the dynamic constraint coefficient (fixed edge weight parameter) through alternating iteration, and the multiplier term is updated synchronously to balance the convergence direction of the two variables, and finally generates the edge weight parameter update rule of the joint optimization model.

[0037] Based on the bivariate interactive optimization framework, the updated edge weight parameters are generated through the gradient backpropagation algorithm; It should be noted that, under the condition that the dynamic constraint coefficient is fixed, the partial derivative of the edge weight parameter is calculated based on the fitting error of the standardized spatiotemporal series data; along the hierarchical topological path direction of the directed causal network graph (such as temperature node → concentration node → emission node), the gradient value of each edge weight parameter is calculated layer by layer from the terminal node in reverse. During the calculation process, the causal cumulative effect weight within the time window of the standardized spatiotemporal series data is extracted by the path integral tracking algorithm, and it is fused with the gradient value to generate a corrected gradient; based on the multiplier term adjustment amount defined by the corrected gradient direction and the alternating direction multiplier decomposition method, the edge weight parameter is updated by the gradient descent algorithm, and the update logic is the current edge weight parameter minus the product of the linear combination of the learning rate, the corrected gradient value and the multiplier term; during the iterative process, the edge weight parameter gradually approaches the optimal solution that balances the fitting error and the constraint penalty term through gradient back propagation, and finally generates the updated edge weight parameter that dynamically adapts to the real-time data fluctuations (such as gas concentration monitoring values) and process safety thresholds (such as concentration limits defined in emission standard documents).

[0038] Based on the updated edge weight parameters, dynamic constraint coefficients are generated through the proximal gradient algorithm and the residual convergence judgment method.

[0039] It should be noted that, under the condition that the edge weight parameters are fixed, the gradient direction of the dynamic constraint coefficient is calculated based on the constraint penalty term of the threshold function group, wherein the logical constraint condition generated by the process safety threshold conversion is modulated by the update step size through the Sigmoid function gradient term; the proximal gradient algorithm performs gradient descent update on the dynamic constraint coefficient, and the update formula is the current dynamic constraint coefficient minus the product of the linear combination of the learning rate and the gradient value and the multiplier term defined by the alternating direction multiplier method decomposition method; the residual convergence judgment method calculates the residual norm of the current dynamic constraint coefficient and the result of the previous iteration. If the residual norm is lower than the preset threshold, it is judged to converge and the optimized dynamic constraint coefficient is output; if it does not converge, the constraint penalty term is recalculated based on the updated dynamic constraint coefficient, and the proximal gradient update and residual judgment are iteratively performed until the convergence conditions are met; finally, a dynamic constraint coefficient is generated that is coordinated with the edge weight parameters and meets the process safety threshold constraint.

[0040] S5. Input the real-time standardized spatiotemporal series data into the joint optimization model, extract the causal characteristics through the dynamic constraint coefficient and calculate the pollution contribution score of each monitoring point, and generate a dynamic governance optimization instruction set based on the pollution contribution score ranking.

[0041] The real-time standardized spatiotemporal series data are input into the joint optimization model. Based on the time-varying coupling relationship between the network edge weight parameters and the dynamic constraint coefficients, the dynamic causal correlation characteristics of each monitoring node are extracted through the multi-path integral tracking algorithm. It should be noted that along the hierarchical propagation paths of a directed causal network graph (e.g., temperature node → concentration node → emission node), a path integral tracing algorithm is executed within a sliding time window. Real-time node data values ​​and edge weight parameters are weighted and integrated along the time dimension to calculate the historical causal cumulative strength. The dynamic causal association feature is composed of a path integral value, a dynamic constraint coefficient adjustment factor, and a gradient feedback factor. The path integral value reflects the strength of the historical causal propagation effect between nodes. The dynamic constraint coefficient adjustment factor dynamically modulates the causal strength weight gradient based on the real-time process safety threshold constraint. The gradient feedback factor updates the causal influence ratio through the gradient back-propagation chain of the edge weight parameter and the dynamic constraint coefficient. For example, the dynamic causal association feature from a temperature node to a concentration node can be represented as the product of the historical cumulative strength of the path integral tracing algorithm and the real-time modulation of the dynamic constraint coefficient, representing the causal contribution of the direct causal link from the temperature node to the concentration node under time-varying coupling. Ultimately, a dynamic causal association feature set is generated for each monitoring node under multiple time scales and propagation paths, which is used to characterize the real-time causal evolution of environmental parameters.

[0042] According to the dynamic causal association characteristics, the constrained spatiotemporal integral accumulation is performed along the hierarchical edge direction of the directed causal network graph, and the nonlinear modulation of the pollution propagation path intensity by the dynamic constraint coefficient is synchronously integrated to generate the pollution contribution score of each monitoring point. Its expression is: ; in, A timestamp reference point for real-time monitoring. is a directed causal path from the pollution source node to the target monitoring node in the directed causal network graph, For all the nodes in the directed causal network graph pointing to The set of causal transmission paths of monitoring points, is the length of the sliding time window, For path At the moment The edge weight parameter, Is the path In time of Real-time normalization of spatiotemporal series data values, For path At the moment The dynamic constraint coefficient, is the mean modulation value of all monitoring points in the current time window, is the Sigmoid function, is the standard deviation of the modulation values ​​of all monitoring points in the current time window, It is Pollution contribution score of each monitoring point; It should be noted that along the hierarchical edges of a directed causal network graph, starting from the root node (e.g., temperature node), all downstream causal paths (e.g., temperature node → concentration node → emission node) are traversed layer by layer in topological order, with the traversal order determined by the hierarchical edge directions. For each path, a spatiotemporal integration is performed within a sliding time window: the node values ​​in the real-time standardized spatiotemporal series data (derived from the standardized output of the preprocessing module) are multiplied by edge weight parameters (generated by the path integral tracking algorithm based on a weighted calculation of historical lagged cross-correlation coefficients and causal transmission strength) point by point along the time dimension and the integral is accumulated to generate the pollution transmission path strength, which directly quantifies the cumulative pollution transmission impact of the path within the time window. Subsequently, the dynamic constraint coefficient (optimized using the alternating direction multiplication method) is converted to a nonlinear weight using a sigmoid function and multiplied by the pollution transmission path strength to generate a modulation value. This process enhances the contribution of paths that exceed the process safety threshold (amplified when the weight approaches 1) and suppresses the contribution of paths that do not exceed the threshold (decreased when the weight approaches 0). Finally, the real-time statistical mean and standard deviation of the modulation values ​​of all monitoring points in the current time window are normalized to generate the pollution contribution score of each monitoring point.

[0043] Based on the pollution contribution score, a descending list of pollution contribution scores of monitoring points is generated through a heap sort algorithm, and the priority levels are divided based on hash mapping; Based on the priority level and pollution contribution score of each monitoring point, the control intensity parameters of each monitoring point are calculated through a dynamic weight adjustment algorithm combined with real-time process safety thresholds, and a dynamic governance optimization instruction set is generated using linear programming; It should be noted that the pollution contribution score set of each monitoring point is taken as input, and the maximum value of the current remaining pollution contribution score is output in sequence through the extraction of the top element of the max-heap data structure and the heap structure adjustment operation, until all the elements in the heap are sorted, and a descending list arranged from high to low by the pollution contribution score is generated; based on the predefined score interval and priority level mapping relationship table (for example, a pollution contribution score higher than 0.8 is mapped to high priority, between 0.5 and 0.8 is mapped to medium priority, and below 0.5 is mapped to low priority), each pollution contribution score in the descending list is quickly mapped to the corresponding priority level through a hash function, and an independent identifier is assigned to each level; finally, a pollution contribution descending hierarchical list containing priority identifiers is generated, which is used to guide the update frequency of the edge weight parameters and dynamic constraint coefficients of the joint optimization model and the priority execution order of the pollution tracing strategy.

[0044] Environmental governance actions are driven by the dynamic control parameters in the dynamic governance optimization instruction set, and the post-governed environmental data is simultaneously collected to verify the effectiveness of the dynamic control parameters and fed back to the joint optimization model for iterative weight updates.

[0045] It should be noted that the order of treatment is determined based on the priority identifiers in the descending hierarchy of pollution contribution, with high-priority pollution sources being treated first. Treatment intensity is then calculated based on the dynamic control parameters (edge ​​weight parameters and dynamic constraint coefficients) output by the joint optimization model. The edge weight parameters reflect the historical pollution transmission intensity through a path integral tracing algorithm, and the dynamic constraint coefficients reflect the real-time degree of exceedance through a proximal gradient algorithm. The dynamic control parameters are then input into the control equations for the environmental treatment equipment, which are derived from the parameter-action conversion formulas specified in the equipment operating manual. The final output of the environmental treatment action includes specific operational instructions (such as ventilation volume adjustment amplitude and purification equipment power setting value), which strictly match the current pollution contribution score and process safety threshold requirements. After the treatment action is implemented, real-time environmental monitoring data is immediately used to verify the treatment effect, and deviations are fed back to the joint optimization model for iterative parameter updates.

[0046] This embodiment also provides an environmental protection monitoring system based on the Internet of Things, including: a preprocessing module, a causal network construction module, a constraint function generation module, a joint optimization model construction module, and a dynamic decision instruction generation module; The preprocessing module collects environmental parameters in real time and performs preprocessing to generate standardized spatiotemporal series data; The causal network construction module generates an initial causal network based on standardized spatiotemporal sequence data through the PC algorithm, and combines it with the rule matching verification of the Neo4j knowledge graph to annotate causal relationships and generate a directed causal network diagram; The constraint function generation module obtains the process safety threshold, uses the ReLU function and the Sigmoid function to make the process safety threshold derivable, and generates the constraint function in combination with the directed causal network graph; The joint optimization model construction module builds a joint optimization model based on a directed causal network graph and constraint functions, synchronously trains the edge weight parameters of the joint optimization model through the alternating direction multiplier method, and obtains the dynamic constraint coefficients; The dynamic decision-making instruction generation module inputs real-time standardized spatiotemporal series data into the joint optimization model, extracts causal characteristics through dynamic constraint coefficients, calculates the pollution contribution score of each monitoring point, and generates a dynamic governance optimization instruction set based on the contribution ranking.

[0047] This embodiment also provides a computer device, which is suitable for the environmental protection monitoring method based on the Internet of Things, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the environmental protection monitoring method based on the Internet of Things proposed in the above embodiment.

[0048] The computer device may be a terminal, comprising a processor, memory, a communication interface, a display, and an input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores an operating system and computer programs. The internal memory provides an environment for the operating system and computer programs stored in the non-volatile storage media. The communication interface of the computer device is used to communicate with external terminals via wired or wireless communication. Wireless communication may be achieved via Wi-Fi, a carrier network, NFC (near-field communication), or other technologies. The display of the computer device may be a liquid crystal display or an electronic ink display. The input device may be a touchscreen overlay on the display, buttons, a trackball, or a touchpad on the computer device housing, or an external keyboard, touchpad, or mouse.

[0049] This embodiment also provides a storage medium having a computer program stored thereon. When the program is executed by a processor, the method for implementing environmental protection monitoring based on the Internet of Things as proposed in the above embodiment is implemented. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0050] In summary, the present invention uses: the PC algorithm to generate a data-driven initial causal network, combines the rule matching correction direction of the Neo4j knowledge graph, eliminates redundant paths and integrates process mechanisms, solves the problem of misjudgment of causal direction in complex scenarios, and generates a high-confidence directed network; uses the ReLU / Sigmoid function to convert discrete process thresholds into continuous and differentiable constraint functions, and realizes hierarchical dynamic constraint fusion through the graph constraint propagation algorithm, so that the optimization model strictly follows safety rules.

[0051] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. An environmental protection monitoring method based on the Internet of Things, characterized by: include, Collect environmental parameters in real time and perform preprocessing to generate standardized spatiotemporal series data; Based on standardized spatiotemporal sequence data, the PC algorithm is used to generate an initial causal network. Combined with the rule matching verification of the Neo4j knowledge graph, the causal relationships are annotated and a directed causal network diagram is generated. Obtain the process safety threshold, use the ReLU function and Sigmoid function to make it derivable, and generate the constraint function in combination with the directed causal network graph; A joint optimization model is constructed based on a directed causal network graph and a constraint function. The edge weight parameters of the joint optimization model are trained synchronously through the alternating direction multiplier method, and the dynamic constraint coefficients are obtained. The real-time standardized spatiotemporal series data are input into the joint optimization model, the causal characteristics are extracted through the dynamic constraint coefficient, and the pollution contribution score of each monitoring point is calculated. The dynamic governance optimization instruction set is generated based on the contribution ranking.

2. The environmental protection monitoring method based on the Internet of Things according to claim 1, characterized in that: The preprocessing includes timestamp synchronization, spatial interpolation completion, denoising and data standardization.

3. The environmental protection monitoring method based on the Internet of Things according to claim 2, characterized in that: Based on the standardized spatiotemporal sequence data, the initial causal network is generated by the PC algorithm, and the rule matching verification of the Neo4j knowledge graph is combined to mark the causal relationship and generate a directed causal network diagram. The specific steps are as follows: Based on the standardized spatiotemporal series data, the PC algorithm is used to perform a conditional independence test, and the causal direction hypothesis of the test results is verified to generate the initial causal network structure; Based on the initial causal network structure, Neo4j is used to perform rule matching and causal path directionality correction on the knowledge graph storage to generate a causal network diagram with direction identification. Based on the causal network graph with direction identification, a directed causal network graph is generated through the feedback edge set optimization algorithm.

4. The environmental protection monitoring method based on the Internet of Things according to claim 3, characterized in that: The process safety threshold is obtained by using the ReLU function and the Sigmoid function to perform derivable processing on the process safety threshold, and a constraint function is generated in combination with a directed causal network diagram. The specific steps are as follows: Extract the process safety threshold, perform piecewise derivable transformation through ReLU function and Sigmoid function, and generate a threshold function group; Based on the directed causal network graph, the threshold function group is multi-level topologically weighted fused through the graph constraint propagation algorithm to generate the constraint function.

5. The environmental protection monitoring method based on the Internet of Things according to claim 4, characterized in that: The specific steps of constructing a joint optimization model based on a directed causal network graph and a constraint function are as follows: Based on the directed causal network graph, the causal network edge connection direction and edge weight parameters between nodes are initialized through the feedback edge set optimization algorithm; Based on the edge connection direction and edge weight parameters of the causal network, a multi-level constraint fusion algorithm is used to bind the constraint functions and establish a time-varying coupling relationship between the network edge weight parameters and the dynamic constraint coefficients. The alternating direction multiplier method is then combined to fuse the fitting error and constraint penalty to form a joint optimization model.

6. The environmental protection monitoring method based on the Internet of Things according to claim 5, characterized in that: The alternating direction multiplier method is used to synchronously train the edge weight parameters of the joint optimization model and obtain the dynamic constraint coefficient. The specific steps are as follows: The path integral tracing algorithm within the time window is used to perform multi-scale spatiotemporal correlation analysis on the temporal cumulative effect of the causal path to generate the initial iterative value of the edge weight parameter; Based on the initial iteration value of the edge weight parameter, a bivariate interactive optimization framework with multiplier terms is generated through the alternating direction multiplier decomposition method; Based on the bivariate interactive optimization framework, the updated edge weight parameters are generated through the gradient backpropagation algorithm; Based on the updated edge weight parameters, dynamic constraint coefficients are generated through the proximal gradient algorithm and the residual convergence judgment method.

7. The environmental protection monitoring method based on the Internet of Things according to claim 6, characterized in that: The specific steps of inputting real-time standardized spatiotemporal series data into the joint optimization model, extracting causal characteristics through dynamic constraint coefficients and calculating the pollution contribution score of each monitoring point, and generating a dynamic governance optimization instruction set based on the contribution ranking are as follows: The real-time standardized spatiotemporal series data are input into the joint optimization model. Based on the time-varying coupling relationship between the network edge weight parameters and the dynamic constraint coefficients, the dynamic causal correlation characteristics of each monitoring node are extracted through the multi-path integral tracking algorithm. Based on the dynamic causal association characteristics, constrained spatiotemporal integral accumulation is performed along the hierarchical edges of the directed causal network graph, and the nonlinear modulation of the pollution propagation path intensity by the dynamic constraint coefficient is simultaneously integrated to generate the pollution contribution score of each monitoring point. Based on the pollution contribution score, a descending list of pollution contribution scores of monitoring points is generated through a heap sort algorithm, and the priority levels are divided based on hash mapping; Based on the priority level and pollution contribution score of each monitoring point, the control intensity parameters of each monitoring point are calculated through a dynamic weight adjustment algorithm combined with real-time process safety thresholds, and a dynamic governance optimization instruction set is generated using linear programming; Environmental governance actions are driven by the dynamic control parameters in the dynamic governance optimization instruction set, and the post-governed environmental data is simultaneously collected to verify the effectiveness of the dynamic control parameters and fed back to the joint optimization model for iterative weight updates.

8. An Internet of Things-based environmental protection monitoring system, based on the Internet of Things-based environmental protection monitoring method according to any one of claims 1 to 7, characterized in that: Including, pre-processing module, causal network construction module, constraint function generation module, joint optimization model construction module, dynamic decision instruction generation module; The preprocessing module collects environmental parameters in real time and performs preprocessing to generate standardized spatiotemporal series data; The causal network construction module generates an initial causal network based on standardized spatiotemporal sequence data through the PC algorithm, and combines it with the rule matching verification of the Neo4j knowledge graph to annotate causal relationships and generate a directed causal network diagram; The constraint function generation module obtains the process safety threshold, uses the ReLU function and the Sigmoid function to make the process safety threshold derivable, and generates the constraint function in combination with the directed causal network graph; The joint optimization model construction module builds a joint optimization model based on a directed causal network graph and constraint functions, synchronously trains the edge weight parameters of the joint optimization model through the alternating direction multiplier method, and obtains the dynamic constraint coefficients; The dynamic decision-making instruction generation module inputs real-time standardized spatiotemporal series data into the joint optimization model, extracts causal characteristics through dynamic constraint coefficients, calculates the pollution contribution score of each monitoring point, and generates a dynamic governance optimization instruction set based on the contribution ranking.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the environmental protection monitoring method based on the Internet of Things according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the environmental protection monitoring method based on the Internet of Things according to any one of claims 1 to 7 are implemented.

Images