Intelligent monitoring system and method for hazardous waste based on the Internet of Things

Abstract

The present invention discloses an intelligent hazardous waste supervision system and method based on the Internet of Things, which specifically relates to the field of solid hazardous waste supervision and is used to solve the problems of data consistency and anomaly detection in solid hazardous waste management. The system integrates multi-source Internet of Things sensor data and uses a long short-term memory network model to build a normal operating baseline in real time and mark abnormal data; uses partial mutual information and betweenness centrality indicators to build a multi-node data association network, quickly identifies the key nodes that have the greatest impact on data consistency, and locks the source of data mutation; uses wavelet transform and partial directed coherence method, combined with a historical fault case library, to identify the root cause of data mutation; automatically performs isolation and repair operations based on the root cause analysis results, and calls backup monitoring devices to ensure monitoring continuity and rapid recovery; thereby significantly enhancing the stability and reliability of the intelligent solid hazardous waste supervision system.

Description

Technical Field

[0001] The present invention relates to the field of solid hazardous waste supervision, and more specifically, to an intelligent hazardous waste supervision system and method based on the Internet of Things. Background Art

[0002] With the widespread application of IoT technology in hazardous solid waste management, the collection, transportation, and storage processes have become highly digitized and intelligent. By deploying a variety of sensor devices at each stage, real-time data collection and monitoring are enabled, improving the efficiency and safety of hazardous waste management. For example, weight sensors are installed at collection points to monitor waste volume, GPS and load sensors track transportation status during transportation, and environmental sensors at storage nodes monitor key parameters such as temperature and humidity. This multi-node, multi-link data flow not only enhances regulatory transparency but also provides a solid foundation for data-driven decision-making. However, as system complexity increases, data consistency and accuracy become key factors affecting overall regulatory effectiveness. In practical applications, issues such as sensor failure, communication interruptions, or human intervention inevitably lead to missing or distorted data, which can lead to erroneous risk assessments and emergency responses, and in severe cases, even to environmental pollution and safety accidents.

[0003] In a multi-node hazardous waste data chain, identifying key nodes where data consistency mutations occur has become a technical challenge that urgently needs to be solved. Specifically, in the multi-link data flow of hazardous waste collection, transportation, and storage, factors such as sensor failure, human intervention, or communication interruption may lead to the loss or mutation of key data. For example, when waste from multiple hazardous waste collection points is centrally transferred to the same storage node, if there is an abnormal increase in the transfer weight, it is difficult to quickly determine whether this mutation is due to data distortion at a certain collection point or due to a deviation in the load distribution during transportation. In this case, there is a lack of effective algorithms and mechanisms to monitor and analyze data consistency in real time, making it difficult to quickly lock the mutation nodes, which in turn affects the reliability and response speed of the overall hazardous waste management system. Therefore, there is an urgent need to develop an innovative multi-node data consistency detection and key node identification method that can accurately locate abnormal data sources in complex data streams and distinguish the root causes of data mutations through intelligent analysis to ensure the stability and security of the hazardous waste management system.

[0004] In order to solve the above problems, a technical solution is now provided. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides an intelligent hazardous waste supervision system and method based on the Internet of Things. By integrating multi-source Internet of Things sensor data and utilizing a long short-term memory network model, a normal operating baseline is constructed in real time, and abnormal data is accurately marked; through partial mutual information and betweenness centrality indicators, a multi-node data association network is constructed to quickly identify the key nodes that have the greatest impact on data consistency and lock the source of data mutation; wavelet transform and partial directed coherence method are used, combined with a historical fault case library, to accurately identify the root cause of data mutation, such as sensor hardware failure or communication link interruption; isolation and repair operations are automatically performed based on the root cause analysis results, and backup monitoring devices are called to ensure monitoring continuity and rapid recovery; thereby significantly enhancing the stability and reliability of the solid hazardous waste intelligent supervision system to solve the problems raised in the above-mentioned background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] The intelligent management method for hazardous waste based on the Internet of Things includes the following steps:

[0008] S1: In each link of hazardous waste disposal, various sensors are deployed to collect data from corresponding nodes and store them synchronously in the central database according to timestamps;

[0009] S2: Based on the data set in the central database, a long short-term memory network model is used to build a normal operating baseline for each node. The data uploaded by the sensor in real time is predicted and compared. When the actual value deviates from the baseline prediction value by more than the preset threshold, the corresponding node data is marked as abnormal.

[0010] S3: For nodes marked as abnormal, a multi-node data association network is constructed and the correlation and betweenness centrality index between each node is calculated to determine the key nodes that have the greatest impact on the overall data consistency in the corresponding abnormal event, and form a key node list;

[0011] S4: Perform Granger causality analysis on the nodes in the key node list and match them with the node anomaly information in the historical fault case library to identify the root cause of the data mutation;

[0012] S5: Based on the root cause identification results of key nodes, isolation and repair operations are automatically performed and backup monitoring devices are called.

[0013] In a preferred embodiment, step S2 includes the following:

[0014] First, for each node's data set, the sliding window method is used to split the time series data into subsequences of fixed length as the input of the LSTM model; data of different scales are converted to a unified range through normalization:

[0015] Secondly, we build multivariate LSTM models, each of which is trained specifically on the dataset of one node.

[0016] Subsequently, in the real-time data monitoring stage, new data from each node is continuously received and input into the trained LSTM model to generate real-time prediction values; the residual between the predicted value and the actual observed value is used, and a dynamic threshold setting method based on information entropy is applied to calculate the anomaly score of each node.

[0017] In a preferred embodiment, step S3 includes the following contents:

[0018] S3.1. First, a multi-dimensional cross-analysis matrix is constructed based on the historical observation values and anomaly marks of each node in the same period. The partial mutual information method is used to remove redundancy in the mutual information between each pair of nodes and eliminate the confounding interference caused by third-party nodes. The specific logic is as follows:

[0019] Set the time window of abnormal events as T, extract all abnormal nodes N={n1, n2, ..., n k} data set D = {d1, d2, ..., d k}, where each data point d i Contains node n i The values of various parameters at time t;

[0020] For any two abnormal nodes n i and n j , given all other node sets C=N\{n i , n j}, the partial mutual information is defined as: Among them, P(d i , d j , d C ) is the joint probability distribution, P(d i , d j |d C ) is the conditional probability distribution; set the partial mutual information threshold θ PMI , only when PMI(n i ;n j |C)≥θ PMI When i and node n j Edge connections are established between them to form an adjacency matrix.

[0021] In a preferred embodiment, step S3 further includes the following:

[0022] S3.2. Next, based on the values obtained from the partial mutual information, an undirected edge is assigned to each pair of nodes and an adjacency matrix is constructed, i.e., a multi-node data association network. The network betweenness centrality index is then calculated to measure the importance of the node to information flow in the network. The specific logic is as follows:

[0023] Based on the partial mutual information calculation results, a multi-node data association network G = (N, E) is constructed, where the node set N is the abnormal node set, and the edge set E consists of node pairs (n i , n j )composition;

[0024] In the constructed multi-node data association network, the betweenness centrality of each node is calculated to measure the importance of the node as an information bridge in the network;

[0025] S3.3. Then, based on the calculated betweenness centrality values, all abnormal nodes are sorted from high to low according to betweenness centrality; nodes whose betweenness centrality values exceed the betweenness centrality threshold are screened out to form a key node list.

[0026] In a preferred embodiment, step S4 includes the following contents:

[0027] S4.1, first, perform continuous wavelet transform on the time series data of each key node in the key node list to capture the time-frequency characteristics of the data; then, according to the scale parameter of the wavelet coefficient, divide the frequency range into several frequency band sets, each frequency band corresponds to a specific subset of wavelet coefficients, so as to analyze the causal relationship between nodes at different frequencies;

[0028] S4.2, for each frequency band ω v , construct a multivariate model of the key node list and calculate partial directed coherence to quantify the causal relationship between nodes. The specific process is as follows:

[0029] S4.2.1, for each frequency band, assume that the time series data of the key nodes in the corresponding frequency band satisfy the multivariate autoregressive model, expressed as: X(ω v , t) = A(ω v )X(ω v ,t-1)+∈(ω v , t); where X(ω v , t) is the key node in the frequency band ω v The state vector under v ) is the coefficient matrix, ∈(ω v , t) is the noise vector;

[0030] S4.2.2, by solving the multivariate autoregressive model, obtain the frequency band ω vThe transfer function matrix H(ω v ): H(ω v )=[IA(ω v )] -1 ; Where I is the identity matrix;

[0031] S4.2.3, for key node k o With key node k v In the frequency band ω v The following partially directed coherent WPDC is defined as: Among them, H ov (ω v ) is the element in the oth row and vth column of the transfer function matrix;

[0032] S4.2.4, a node k is considered critical only if it is partially directed coherent or equal to the partially directed coherent threshold. v In the frequency band ω v The upper Granger leads to the key node k o ; At this time, construct the directed edge (k v →k o ) indicates a causal relationship;

[0033] S4.3, for each frequency band, construct a causal relationship matrix based on the partial directed coherence calculation results; perform a logical OR operation on the causal relationship matrices of all frequency bands to generate a comprehensive causal relationship matrix.

[0034] In a preferred embodiment, step S4 further includes the following:

[0035] S4.4, the fault case library contains comprehensive causal relationship matrix templates of various abnormal events identified in history and their corresponding fault categories;

[0036] For the comprehensive causal relationship matrix of the current abnormal event, calculate its Jaccard similarity index with each historical fault case;

[0037] Traverse all historical fault cases and calculate the Jaccard similarity index value between the current comprehensive causal relationship matrix and each historical case; select the fault case with the highest Jaccard similarity index value that exceeds the set threshold, and use its corresponding fault category as the root cause of the current abnormal event; if all Jaccard similarity index values do not exceed the set threshold, mark the current abnormal event as a new fault category and wait for subsequent addition to the fault case library.

[0038] In a preferred embodiment, step S5 includes the following contents:

[0039] Based on the list of key nodes and the root cause of the fault, the corresponding handling strategy script is first loaded. The script pre-defines the automated operation process corresponding to different fault categories.

[0040] The IoT-based intelligent hazardous waste supervision system includes: data acquisition module, anomaly detection module, key identification module, root cause analysis module and automatic response module;

[0041] Data collection module: In each link of hazardous waste disposal, various sensors are deployed to collect data from corresponding nodes, and the data is synchronously stored in the central database according to the timestamp. The collected data is then transmitted to the anomaly detection module;

[0042] Anomaly Detection Module: Based on the data set in the central database, the long-short-term memory network model is used to build the normal operation baseline of each node. The module predicts and compares the data uploaded by the sensor in real time. When the actual value deviates from the baseline prediction value by more than the preset threshold, the corresponding node data is marked as abnormal and passed to the key identification module.

[0043] Key Identification Module: For nodes marked as abnormal, by constructing a multi-node data association network and calculating the correlation and betweenness centrality indicators between each node, it determines the key nodes that have the greatest impact on the overall data consistency in the corresponding abnormal event, forms a key node list, and passes the key node list to the root cause analysis module;

[0044] Root cause analysis module: This module performs Granger causality analysis on the nodes in the key node list and matches them with the node anomaly information in the historical fault case library to identify the root cause of the data mutation and transmits the root cause identification results to the automatic response module.

[0045] Automatic response module: Based on the root cause identification results of key nodes, it automatically performs isolation and repair operations and calls backup monitoring devices.

[0046] The technical effects and advantages of the IoT-based hazardous waste intelligent supervision system and method of the present invention are as follows:

[0047] The present invention integrates multi-source IoT sensor data and utilizes a long-short-term memory network model to construct a normal operating baseline in real time and accurately mark abnormal data. It constructs a multi-node data association network through partial mutual information and betweenness centrality indicators, quickly identifies key nodes that have the greatest impact on data consistency, and locks in the source of data mutations. It uses wavelet transform and partial directed coherence methods, combined with a historical fault case library, to accurately identify the root cause of data mutations, such as sensor hardware failure or communication link interruption. It automatically performs isolation and repair operations based on the root cause analysis results, and calls backup monitoring devices to ensure monitoring continuity and rapid recovery. This significantly enhances the stability and reliability of the solid hazardous waste intelligent supervision system. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 This is a flow chart of the method for intelligent management of hazardous waste based on the Internet of Things of the present invention;

[0049] Figure 2 This is a schematic diagram of the structure of the intelligent hazardous waste supervision system based on the Internet of Things of the present invention. DETAILED DESCRIPTION

[0050] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0051] Example 1: Figure 1 The present invention provides a method for intelligently monitoring hazardous waste based on the Internet of Things, including:

[0052] S1: In each link of hazardous waste disposal, various sensors are deployed to collect data from corresponding nodes and store them synchronously in the central database according to timestamps;

[0053] S2: Based on the data set in the central database, a long short-term memory network model is used to build a normal operating baseline for each node. The data uploaded by the sensor in real time is predicted and compared. When the actual value deviates from the baseline prediction value by more than the preset threshold, the corresponding node data is marked as abnormal.

[0054] S3: For nodes marked as abnormal, a multi-node data association network is constructed and the correlation and betweenness centrality index between each node is calculated to determine the key nodes that have the greatest impact on the overall data consistency in the corresponding abnormal event, and form a key node list;

[0055] S4: Perform Granger causality analysis on the nodes in the key node list and match them with the node anomaly information in the historical fault case library to identify the root cause of the data mutation and clarify the anomaly type;

[0056] S5: Based on the root cause identification results of key nodes, isolation and repair operations are automatically performed and backup monitoring devices are called.

[0057] With the widespread application of Internet of Things (IoT) technology in solid hazardous waste management, the collection, transportation, and storage of hazardous waste have achieved a high degree of digitization and intelligence. By deploying weight sensors, GPS modules, and environmental sensors in each link, key data is collected in real time, improving the efficiency and safety of waste management. However, when it comes to multi-node and multi-link data integration, problems such as data noise, inconsistent formats, and time synchronization still restrict the overall performance of the system and the accuracy of data analysis. To this end, this solution proposes an innovative multi-source data integration and preprocessing technology. It aims to ensure the high quality and consistency of data in each link through a precise data processing process, providing a solid data foundation for subsequent anomaly detection and key node identification.

[0058] Step S1 includes the following contents:

[0059] Throughout the collection, transfer, and storage of hazardous waste, various sensors, including weight sensors, GPS modules, and environmental sensors (including temperature, humidity, and hazardous gas concentration sensors), are deployed to collect real-time weight, location, and environmental parameter data at each node. The collected data is first subjected to dynamic noise suppression using an adaptive Kalman filter to ensure the authenticity and stability of the weight and environmental data. For the location information obtained by the GPS module, the Geographically Weighted Regression (GWR) algorithm is applied to correct for spatial errors, improving the accuracy of the location data.

[0060] Next, missing values in the data were filled using polynomial regression interpolation to ensure the integrity of the dataset. To address the differences in data formats generated by different sensors, a unified data conversion protocol was designed and implemented, converting all data into a structured, unified format (such as JSON) for seamless integration into the central database. Furthermore, a distributed time synchronization algorithm based on the Precision Time Protocol (PTP) was used to calibrate the timestamps of data from all nodes, ensuring high temporal consistency across all links.

[0061] To achieve efficient data storage and retrieval, we selected a distributed database system (such as Apache Cassandra) and designed a database architecture specifically tailored to the needs of hazardous waste management, supporting high-throughput writes and low-latency queries for large-scale data. We implemented data compression and partitioning strategies to optimize storage space utilization and access performance. Furthermore, we built a real-time data stream processing pipeline based on Apache Kafka and Apache Flink to ensure rapid data transfer to the analysis stage, enabling real-time data processing and monitoring.

[0062] Through the above steps, S1 achieves comprehensive integration and efficient preprocessing of multi-source data across all stages of hazardous waste collection, transportation, and storage, eliminating noise and inconsistencies in the data and ensuring data integrity and time-series synchronization. This technical logic not only lays a solid data foundation for subsequent real-time anomaly detection and key node identification, but also significantly improves the data processing capabilities and monitoring accuracy of the overall hazardous waste management system, providing key support for the efficient operation of the smart supervision system.

[0063] After the multi-source data integration and preprocessing step (S1) ensures the high quality and consistency of weight, location, and environmental data from collection, transportation, and storage, the next challenge is to effectively identify and flag anomalies in this data so that timely countermeasures can be taken. To this end, step S2 proposes an intelligent anomaly detection method based on a long short-term memory network (LSTM). By constructing a baseline model of normal operation for each node, it predicts and compares real-time data to achieve accurate anomaly data labeling. This method not only improves the accuracy and real-time performance of anomaly detection but also provides reliable data support for subsequent key node identification.

[0064] Step S2 includes the following contents:

[0065] After integrating and storing the weight, location, and environmental data from hazardous waste collection, transportation, and storage in the central database, step S2 aims to build a normal operating baseline for each node using a long-short-term memory network model and perform real-time anomaly detection. The specific technical processing logic is as follows:

[0066] First, for each node's dataset, a sliding window approach is used to segment the time series data into fixed-length subsequences, which serve as inputs for the LSTM model. These subsequences include weight data, GPS location coordinates, and environmental parameters (temperature, humidity, hazardous gas concentration, etc.). Normalization (for example, z-score normalization) converts data of varying scales to a uniform range, improving model training efficiency and prediction accuracy.

[0067] Next, we constructed multivariate LSTM models, each trained specifically for a node's dataset. The model's input layer receives preprocessed time series data, which is then passed through multiple layers of LSTM units to capture temporal dependencies and sequence features. Finally, a fully connected layer outputs a prediction for the next time step. During training, mean squared error (MSE) is used as the loss function, and the model parameters are optimized through a backpropagation algorithm to ensure the model accurately predicts data trends for each node under normal operating conditions.

[0068] Subsequently, during the real-time data monitoring phase, new data from each node is continuously fed into the trained LSTM model to generate real-time predictions. Using the residuals between the predicted and observed values, a dynamic threshold setting method based on information entropy is applied to calculate an anomaly score for each node. Specifically, an entropy weighting method is used to weight the multivariate residuals to form a comprehensive anomaly score. When a node's anomaly score exceeds a preset threshold, the corresponding node's data is marked as an anomaly.

[0069] Finally, the data from nodes marked as abnormal is passed to step S3 for key node correlation analysis and location. The results and relevant parameters of each anomaly detection are recorded for continuous optimization and adjustment of the model. Step S2 not only enables efficient real-time data monitoring and accurate anomaly detection, but also lays a solid foundation for the intelligent advancement of the overall hazardous waste management system.

[0070] Step S2 systematically achieves accurate monitoring and anomaly detection of real-time data of each node in hazardous waste management by constructing and applying a multivariate long-short-term memory network model. Through sliding window data segmentation, standardization processing, multivariate LSTM model training and prediction, and dynamic threshold setting based on information entropy, step S2 ensures high-accuracy labeling of abnormal data. This technical logic not only effectively identifies potential data anomalies, but also provides reliable anomaly markers for subsequent key node correlation analysis (step S3), significantly improving the overall intelligence level and responsiveness of the solid hazardous waste intelligent supervision system based on the Internet of Things, ensuring that the system maintains a high degree of stability and security in a complex and changing operating environment.

[0071] After completing the anomaly detection based on the long short-term memory network model (corresponding to step S2), the real-time data of each node has been marked as abnormal. Since the hazardous waste management system has the characteristics of multi-node collaborative work, once a data anomaly occurs in a node, it is very likely to have a collateral impact on the data information of other nodes. How to quickly identify the key nodes that pose the greatest threat to the overall data consistency in the complex node relationships has become another key link to ensure the safety and stability of the system. To this end, step S3 proposes a technical method based on multi-node data association network and topological analysis, using high-order correlation and betweenness centrality indicators to quantify the influence and criticality of each node in abnormal events from a global perspective, thereby forming a list of key nodes, laying a solid data foundation for subsequent root cause analysis and fault handling.

[0072] Step S3 includes the following contents:

[0073] After completing the abnormal marking of each node data in step S2, the system passes these marks along with the original time series data to the multi-node data association analysis module to build a multi-node data association network and determine the key nodes that are most susceptible to overall data consistency. The specific processing logic is as follows:

[0074] S3.1, first, the system constructs a multi-dimensional cross-analysis matrix based on the historical observation values and anomaly marks of each node in the same period. In order to avoid misleading influences on the network topology, the partial mutual information method is used to deduplicate the mutual information between each pair of nodes and eliminate the mixed interference caused by third-party nodes. Partial mutual information measures the degree of feature sharing between two nodes after controlling other potential correlation factors. The higher the value, the stronger the direct correlation between the two nodes. This method can effectively reduce false correlations caused by link redundancy or multicollinearity. The specific logic is as follows:

[0075] Set the time window of abnormal events as T, extract all abnormal nodes N={n1, n2, ..., n k} data set D = {d1, d2, ..., d k}, where each data point d i Contains node n i The values of various parameters at time t.

[0076] For any two abnormal nodes n i and n j , given all other node sets C=N\{n i , n j}, the partial mutual information is defined as: Among them, P(d i , dj , d C ) is the joint probability distribution, P(d i , d j |d C ) is the conditional probability distribution.

[0077] By calculating partial mutual information, the indirect correlation caused by other nodes can be effectively eliminated, and only the direct information dependency between the two nodes is retained.

[0078] Set the partial mutual information threshold θ PMI , only when PMI(n i ;n j |C)≥θ PMI When i Establish edge connections between nodes and form an adjacency matrix A, that is:

[0079] S3.2, next, according to the value obtained from the partial mutual information, an undirected edge is assigned to each pair of nodes and an adjacency matrix is constructed, that is, a multi-node data association network. Whenever the partial mutual information value exceeds the partial mutual information threshold, it is considered that there may be coupling in the abnormal event. After obtaining the complete multi-node data association network, the network calculation betweenness centrality index is calculated to measure the importance of the node to the information flow in the network. Specifically, the betweenness centrality is calculated by counting the frequency of the shortest path passing between each node. The higher the value, the more influential the node is in the process of abnormal information propagation, and the more likely it is to affect the overall data consistency. The specific logic is as follows:

[0080] Based on the partial mutual information calculation results, a multi-node data association network G = (N, E) is constructed, where the node set N is the abnormal node set, and the edge set E consists of node pairs (n i , n j This network structure reflects the direct correlation between abnormal nodes and forms a graph model for subsequent network analysis.

[0081] In a constructed multi-node data association network, the betweenness centrality of each node is calculated to measure the importance of the node as an information bridge in the network. The betweenness centrality is defined as: Among them, σ sd represents the total number of shortest paths from node s to node d, σ sd (n i ) indicates that the node n i A high betweenness centrality value indicates that the node plays a key bridge role in the network and has a greater influence on information flow.

[0082] S3.3, then, based on the calculated betweenness centrality values, sort all abnormal nodes from high to low according to betweenness centrality. Set the betweenness centrality threshold θ BC , filter out the nodes whose betweenness centrality value exceeds the betweenness centrality threshold, and form a key node list K = {k1, k2, ..., k m}, where m ≤ k. The nodes in this list are considered the key nodes with the greatest impact on overall data consistency during the current anomaly event. The betweenness centrality ranking reflects the role of nodes in bridging anomaly propagation within the network, while the partial mutual information value demonstrates the strength of direct correlations between nodes. Combining these two, we can quickly identify the dominant node with the greatest destructive power over overall data consistency among multiple anomaly nodes.

[0083] In step S3.4, the generated list of key nodes and their topological information within the multi-node data association network are passed to step S4 for further causal analysis and root cause identification. The network structure and key node information of the current abnormal event are also recorded as reference data for subsequent model optimization and algorithm adjustments, ensuring that the system can more efficiently handle similar abnormal events in the future.

[0084] Step S3 constructs an efficient and accurate multi-node data association network by combining partial mutual information and betweenness centrality indicators, which can quickly identify the key nodes that have the greatest impact on the overall data consistency in complex and changeable abnormal events. The partial mutual information method is used to eliminate high-order correlation interference to ensure the authenticity of direct correlations between nodes; the betweenness centrality indicator is used to quantify the importance of nodes in information flow and accurately lock key nodes. This technical logic not only improves the accuracy and efficiency of key node identification, but also provides a clear goal for subsequent root cause analysis and fault handling, significantly enhancing the intelligence and responsiveness of the solid hazardous waste intelligent supervision system based on the Internet of Things, and ensuring the efficient and stable operation of the system in the face of multi-node anomalies.

[0085] Step S3 successfully identifies the key nodes that have the greatest impact on the overall data consistency by constructing a multi-node data association network and calculating the correlation and betweenness centrality indicators. However, after identifying the key nodes, it is still necessary to deeply analyze the causal relationship between these nodes to clarify the specific root cause of the data mutation. To this end, step S4 proposes an innovative frequency domain causal analysis method based on wavelet transform and partial directed coherence, combined with a historical fault case library for matching, to achieve accurate identification of the root cause of abnormal events. This method improves the accuracy and sensitivity of causal relationship identification through frequency domain analysis, provides a scientific basis for system fault diagnosis and prevention, and significantly enhances the intelligence level and response capability of the hazardous waste management system.

[0086] Step S4 includes the following contents:

[0087] After generating a list of key nodes and their associated network topology information in step S3, step S4 aims to accurately identify the root cause of data mutations by combining wavelet transform and partial directed coherence method with a historical fault case library. The specific technical processing process is as follows:

[0088] S4.1, first, perform continuous wavelet transform on the time series data of each key node in the key node list to capture the time-frequency characteristics of the data.

[0089] Subsequently, the frequency range is divided into several frequency band sets according to the scale parameters of the wavelet coefficients, and each frequency band corresponds to a specific subset of wavelet coefficients, so as to analyze the causal relationship between nodes at different frequencies.

[0090] S4.2, for each frequency band ω v , construct a multivariate model of the key node list and calculate partial directed coherence to quantify the causal relationship between nodes. The specific process is as follows:

[0091] S4.2.1, for each frequency band, assume that the time series data of the key nodes in the corresponding frequency band satisfy the multivariate autoregressive model, expressed as: X(ω v , t) = A(ω v )X(ω v ,t-1)+∈(ω v , t); where X(ω v , t) is the key node in the frequency band ω v The state vector under v ) is the coefficient matrix, ∈(ω v , t) is the noise vector.

[0092] S4.2.2, by solving the multivariate autoregressive model, obtain the frequency band ω v The transfer function matrix H(ω v ): H(ω v )=[IA(ω v )] -1 ; where I is the identity matrix.

[0093] S4.2.3, for key node k o With key node k v In the frequency band ω v The following partially directed coherent WPDC is defined as: Among them, H ov (ω v ) is the element in the oth row and vth column of the transfer function matrix. The higher the partial directional coherence value, the more critical the node k is. v In the frequency band ω v On the key node ko The stronger the causal influence.

[0094] S4.2.4, set the partial directed coherence threshold, and consider the key node k as a node only when the partial directed coherence is or is equal to the partial directed coherence threshold. v In the frequency band ω v The upper Granger leads to the key node k o At this time, construct the directed edge (k v →k o ) indicates a causal relationship.

[0095] S4.3, for each frequency band, construct a causal relationship matrix based on the partial directed coherence calculation results in: The causal relationship matrix of all frequency bands Perform logical OR operations to generate a comprehensive cause-effect relationship matrix Among them, V represents the logical OR operation, which means that as long as there is a causal relationship in any frequency band, the corresponding element in the comprehensive matrix is 1.

[0096] S4.4, the fault case library contains a comprehensive causal relationship matrix template for various abnormal events that have been identified in history and their corresponding fault categories Where C is the number of historical failure cases.

[0097] For the comprehensive causal relationship matrix of the current abnormal event, calculate its Jaccard similarity index with each historical fault case. The Jaccard similarity index value range is [0, 1]. The closer the value is to 1, the more similar the two causal relationship matrices are.

[0098] Traverse all historical fault cases and calculate the Jaccard similarity index between the current comprehensive causal relationship matrix and each historical case. Select the fault case with the highest Jaccard similarity index value that exceeds the set threshold and use its corresponding fault category as the root cause of the current abnormal event. If all Jaccard similarity index values do not exceed the set threshold, mark the current abnormal event as a new fault category and add it to the fault case library later.

[0099] S4.5, based on the matching result of step S4.4, classify the current abnormal event into one of the following specific root causes:

[0100] Sensor hardware failure: data anomalies caused by sensor damage and signal distortion.

[0101] Communication link interruption: data anomalies caused by network failure or data transmission errors.

[0102] System configuration errors: data anomalies caused by software setting errors and parameter configuration errors.

[0103] The root cause classification results, the current comprehensive causal relationship matrix, and the matched historical fault cases and their fault categories are all passed to step S5 for automated isolation and repair. At the same time, the detailed information of this root cause analysis is recorded in the fault case library as reference data for similar abnormal events in the future, supporting the continuous optimization of the long-short-term memory network model and multi-node data association network analysis method.

[0104] Step S4 introduces wavelet transform and partial directed coherence method to conduct fine-grained analysis of the causal relationship between key nodes in the frequency domain, and realizes the capture of complex dynamic coupling characteristics between nodes in different time scales and frequency ranges. Compared with traditional time domain methods, this method can simultaneously analyze short-term mutation characteristics and long-term trend associations, and is particularly suitable for the multi-dimensional characteristics of sensor data in hazardous waste management scenarios, such as weight, position and environmental parameters, which may present nonlinear dependencies in different time and frequency bands. Combined with the scenario of the present invention, this method can effectively filter out false correlations caused by short-term noise or accidental interference, accurately identify the source of deep-level data mutations across links, and provide a highly reliable and accurate analysis basis for subsequent root cause identification and repair, thereby significantly improving the overall system's abnormality handling capabilities and decision-making efficiency.

[0105] In the IoT-based intelligent supervision system for solid hazardous waste, timely response and handling of abnormal events at key nodes are crucial to ensuring the overall stability and security of the system. Step S4 uses Granger causality analysis to accurately identify the root causes of data mutations, including sensor hardware failures, human intervention, and communication link interruptions. To further ensure the continuity and reliability of the system, step S5 designs and implements a set of automated isolation and repair mechanisms based on the root cause identification results of key nodes, and at the same time calls backup monitoring devices to ensure that the system monitoring function is not affected during the isolation or repair of abnormal nodes. This innovative processing flow can not only quickly restore normal operating conditions, but also effectively prevent potential safety hazards and enhance the system's self-healing and risk resistance capabilities.

[0106] Step S5 includes the following contents:

[0107] S5.1, based on the key node list and fault root cause output from step S4, first load the corresponding handling strategy script. The script predefines the automated operation process corresponding to different fault categories. Taking sensor hardware failure as an example, the system will perform the following mathematical model selection and action instruction issuance:

[0108] The system logically isolates abnormal sensors so that the data they report do not enter the normal data flow of the central database temporarily. o ∈{0, 1} indicates whether the key node o is isolated:

[0109]

[0110] Once a key node k0 is diagnosed as a hardware failure, Notify upper-layer applications to no longer use the key node data for real-time decision-making.

[0111] If redundant or backup sensors are configured in the deployment plan, the pre-written switching rules f switch (·) to activate the corresponding backup device. Define the backup relationship g(k o ) is mapped to a faulty node → a standby sensor, and execute: This ensures that the system still has basic monitoring capabilities after the faulty node stops uploading data.

[0112] If the diagnosis result shows that the communication link is interrupted, the system will automatically restart and self-check the corresponding network interface, call the communication scheduling module and execute the link rerouting algorithm to transfer the node traffic that originally relied on the faulty link to the backup link. This operation maintains a fault link matrix C in the network management layer. fault (k o ), if k o If the failure rate of the communication path exceeds the threshold, it will automatically switch to other available links.

[0113] Step S5 realizes immediate response and processing of abnormal events at key nodes by automatically executing isolation and repair operations, combined with the call of backup monitoring devices. Specifically, the system automatically isolates the faulty node based on the root cause identified in step S4 to prevent abnormal data from further affecting the overall operation of the system. At the same time, the backup monitoring device is started to ensure the continuity of the system monitoring function and the integrity of the data during the period when the main monitoring device is affected. This step reduces the need for human intervention through preset automated processes, shortens fault handling time, and improves the system's response speed and processing efficiency. Step S5 significantly enhances the stability and security of the solid hazardous waste intelligent supervision system based on the Internet of Things, ensuring that the system can quickly resume normal operation and protect environmental and public safety when faced with complex and changing abnormal situations.

[0114] Example 2: Figure 2 The present invention provides a hazardous waste intelligent supervision system based on the Internet of Things, including: a data acquisition module, an anomaly detection module, a key identification module, a root cause analysis module and an automatic response module:

[0115] Data collection module: In each link of hazardous waste disposal, various sensors are deployed to collect data from corresponding nodes, and the data is synchronously stored in the central database according to the timestamp. The collected data is then transmitted to the anomaly detection module;

[0116] Anomaly Detection Module: Based on the data set in the central database, the long-short-term memory network model is used to build the normal operation baseline of each node. The module predicts and compares the data uploaded by the sensor in real time. When the actual value deviates from the baseline prediction value by more than the preset threshold, the corresponding node data is marked as abnormal and passed to the key identification module.

[0117] Key Identification Module: For nodes marked as abnormal, by constructing a multi-node data association network and calculating the correlation and betweenness centrality indicators between each node, it determines the key nodes that have the greatest impact on the overall data consistency in the corresponding abnormal event, forms a key node list, and passes the key node list to the root cause analysis module;

[0118] Root cause analysis module: This module performs Granger causality analysis on the nodes in the key node list and matches them with the node anomaly information in the historical fault case library to identify the root cause of the data mutation and transmits the root cause identification results to the automatic response module.

[0119] Automatic response module: Based on the root cause identification results of key nodes, it automatically performs isolation and repair operations and calls backup monitoring devices.

[0120] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters in the formulas are set by technicians in this field according to actual conditions.

[0121] The above description is merely illustrative of certain exemplary embodiments of the present invention. It goes without saying that those skilled in the art will be able to modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and description are illustrative in nature and should not be construed as limiting the scope of protection of the claims.

[0122] It should be noted that, in this document, if there are relational terms such as first and second, etc., they are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprises", "includes" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device that includes a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article or device. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or device that includes the element.

[0123] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. The intelligent management method for hazardous waste based on the Internet of Things is characterized by: Including steps: S1: In each link of hazardous waste disposal, various sensors are deployed to collect data from corresponding nodes and store them synchronously in the central database according to timestamps; S2: Based on the data set in the central database, a long short-term memory network model is used to build a normal operating baseline for each node. The data uploaded by the sensor in real time is predicted and compared. When the actual value deviates from the baseline prediction value by more than the preset threshold, the corresponding node data is marked as abnormal. S3: For nodes marked as abnormal, a multi-node data association network is constructed and the correlation and betweenness centrality index between each node is calculated to determine the key nodes that have the greatest impact on the overall data consistency in the corresponding abnormal event, and form a key node list; S4: Perform Granger causality analysis on the nodes in the key node list and match them with the node anomaly information in the historical fault case library to identify the root cause of the data mutation; S5: Based on the root cause identification results of key nodes, isolation and repair operations are automatically performed and backup monitoring devices are called.

2. The method for intelligent management of hazardous waste based on the Internet of Things according to claim 1 is characterized in that: Step S2 includes the following contents: First, for each node's dataset, a sliding window method is used to split the time series data into fixed-length subsequences as the input of the LSTM model; data of different scales are converted to a unified range through normalization. Secondly, we build multivariate LSTM models, each of which is trained specifically on the dataset of one node. Subsequently, in the real-time data monitoring phase, new data from each node is continuously received and input into the trained LSTM model to generate real-time prediction values; The residual between the predicted value and the actual observed value is used to calculate the anomaly score of each node by applying a dynamic threshold setting method based on information entropy.

3. The method for intelligent management of hazardous waste based on the Internet of Things according to claim 2 is characterized in that: Step S3 includes the following contents: S3.

1. First, a multi-dimensional cross-analysis matrix is constructed based on the historical observation values and anomaly marks of each node in the same period. The partial mutual information method is used to remove redundancy in the mutual information between each pair of nodes and eliminate the confounding interference caused by third-party nodes. The specific logic is as follows: Set the time window of abnormal events as T, extract all abnormal nodes N={n1, n2, ..., n k } data set D = {d1, d2, ..., d k }, where each data point d i Contains node n i The values of various parameters at time t; For any two abnormal nodes n i and n j , given all other node sets C=N\{n i , n j }, the partial mutual information is defined as: Among them, P(d i , d j , d C ) is the joint probability distribution, P(d i , d j |d C ) is the conditional probability distribution; set the partial mutual information threshold θ PMI , only when PMI(n i ;n j |C)≥θ PMI When i and node n j Edge connections are established between them to form an adjacency matrix.

4. The method for intelligent management of hazardous waste based on the Internet of Things according to claim 3 is characterized in that: Step S3 also Includes the following: S3.

2. Next, based on the values obtained from the partial mutual information, an undirected edge is assigned to each pair of nodes and an adjacency matrix is constructed, i.e., a multi-node data association network. The network betweenness centrality index is then calculated to measure the importance of the node to information flow in the network. The specific logic is as follows: Based on the partial mutual information calculation results, a multi-node data association network G = (N, E) is constructed, where the node set N is the abnormal node set, and the edge set E consists of node pairs (n i , n j )composition; In the constructed multi-node data association network, the betweenness centrality of each node is calculated to measure the importance of the node as an information bridge in the network; S3.

3. Then, based on the calculated betweenness centrality values, all abnormal nodes are sorted from high to low according to betweenness centrality; nodes whose betweenness centrality values exceed the betweenness centrality threshold are screened out to form a key node list.

5. The method for intelligent management of hazardous waste based on the Internet of Things according to claim 4 is characterized in that: Step S4 includes the following contents: S4.1, first, perform continuous wavelet transform on the time series data of each key node in the key node list to capture the time-frequency characteristics of the data; Then, according to the scale parameter of the wavelet coefficients, the frequency range is divided into several frequency band sets, each of which corresponds to a specific subset of wavelet coefficients, so as to analyze the causal relationship between nodes at different frequencies; S4.2, for each frequency band ω v , construct a multivariate model of the key node list and calculate partial directed coherence to quantify the causal relationship between nodes. The specific process is as follows: S4.2.1, for each frequency band, assume that the time series data of the key nodes in the corresponding frequency band satisfy the multivariate autoregressive model, expressed as: X(ω v , t) = A(ω v )X(ω v ,t-1)+∈(ω v , t); where X(ω v , t) is the key node in the frequency band ω v The state vector under v ) is the coefficient matrix, ∈(ω v , t) is the noise vector; S4.2.2, by solving the multivariate autoregressive model, obtain the frequency band ω v The transfer function matrix H(ω v ): H(ω v )=[IA(ω v )] -1 ; Where I is the identity matrix; S4.2.3, for key node k o With key node k v In the frequency band ω v The following partially directed coherent WPDC is defined as: Among them, H ov (ω v ) is the element in the oth row and vth column of the transfer function matrix; S4.2.4, a node k is considered critical only if it is partially directed coherent or equal to the partially directed coherent threshold. v In the frequency band ω v The upper Granger leads to the key node k o ; At this time, construct the directed edge (k v →k o ) indicates a causal relationship; S4.3, for each frequency band, construct a causal relationship matrix based on the partial directed coherence calculation results; perform a logical OR operation on the causal relationship matrices of all frequency bands to generate a comprehensive causal relationship matrix.

6. The method for intelligent management of hazardous waste based on the Internet of Things according to claim 5, characterized in that: Step S4 also includes the following: S4.4, the fault case library contains comprehensive causal relationship matrix templates of various abnormal events identified in history and their corresponding fault categories; For the comprehensive causal relationship matrix of the current abnormal event, calculate its Jaccard similarity index with each historical fault case; Traverse all historical fault cases and calculate the Jaccard similarity index value between the current comprehensive causal relationship matrix and each historical case; The fault case with the highest Jaccard similarity index value that exceeds the set threshold is selected, and its corresponding fault category is used as the root cause of the current abnormal event; if all Jaccard similarity index values do not exceed the set threshold, the current abnormal event is marked as a new fault category and will be added to the fault case library later.

7. The method for intelligent management of hazardous waste based on the Internet of Things according to claim 6, characterized in that: Step S5 includes the following contents: Based on the list of key nodes and the root cause of the fault, the corresponding handling strategy script is first loaded. The script pre-defines the automated operation process corresponding to different fault categories.

8. An intelligent hazardous waste supervision system based on the Internet of Things, used to implement the intelligent hazardous waste supervision method based on the Internet of Things according to any one of claims 1 to 7, characterized in that: include: Data acquisition module, anomaly detection module, key identification module, root cause analysis module and automatic response module; Data collection module: In each link of hazardous waste disposal, various sensors are deployed to collect data from corresponding nodes, and the data is synchronously stored in the central database according to the timestamp. The collected data is then transmitted to the anomaly detection module; Anomaly Detection Module: Based on the data set in the central database, the long-short-term memory network model is used to build the normal operation baseline of each node. The module predicts and compares the data uploaded by the sensor in real time. When the actual value deviates from the baseline prediction value by more than the preset threshold, the corresponding node data is marked as abnormal and passed to the key identification module. Key Identification Module: For nodes marked as abnormal, by constructing a multi-node data association network and calculating the correlation and betweenness centrality indicators between each node, it determines the key nodes that have the greatest impact on the overall data consistency in the corresponding abnormal event, forms a key node list, and passes the key node list to the root cause analysis module; Root cause analysis module: This module performs Granger causality analysis on the nodes in the key node list and matches them with the node anomaly information in the historical fault case library to identify the root cause of the data mutation and transmits the root cause identification results to the automatic response module. Automatic response module: Based on the root cause identification results of key nodes, it automatically performs isolation and repair operations and calls backup monitoring devices.

Images