Sensor network monitoring method, system and device based on industrial internet of things

Abstract

This application discloses a sensor network monitoring method, system, and device based on the Industrial Internet of Things (IIoT), relating to the technical field of sensors. The method includes: determining at least one operating condition cluster based on performance time-series data and energy consumption time-series data; constructing a first operating energy consumption correlation tensor based on the performance time-series data and energy consumption time-series data corresponding to the sensor node, and determining a first operating energy consumption pattern vector for the sensor node based on the first operating energy consumption correlation tensor; for the same operating condition cluster, constructing a second operating energy consumption correlation matrix for the operating condition cluster based on the corresponding first operating energy consumption pattern vector, and determining a second operating energy consumption pattern vector for the operating condition cluster based on the second operating energy consumption correlation matrix; and determining an evaluation index for the sensor node based on the first operating energy consumption pattern vector and the second operating energy consumption pattern vector. This application improves the accuracy of sensor network monitoring.

Description

Technical Field

[0001] This application relates to the field of sensor technology, and in particular to sensor network monitoring methods, systems and devices based on the Industrial Internet of Things. Background Technology

[0002] The rapid development of the Industrial Internet of Things (IIoT) has made large-scale, densely deployed sensor networks a core infrastructure in industrial environments such as smart manufacturing and smart factories. These sensor nodes are responsible for collecting various industrial data in real time, such as temperature, pressure, vibration, and flow rate, providing crucial information for process monitoring, fault early warning, and optimized control. Ensuring the continuous, stable, and efficient operation of the entire sensor network is an important guarantee for the safety and efficiency of industrial production.

[0003] Traditional sensor network monitoring methods mostly focus on analyzing single performance indicators, such as data packet loss rate, acquisition accuracy, or response latency. However, in complex industrial environments, sensor performance is closely related to its energy consumption. A sensor with seemingly normal performance parameters may be operating with abnormally high energy consumption, which is often a precursor to early failure or performance degradation. Existing methods lack mechanisms to correlate performance with energy consumption, making it difficult to gain a holistic understanding of the sensor's true operational health.

[0004] Furthermore, sensor nodes in industrial settings are often located in different physical locations and environments, undertaking diverse production monitoring tasks. This means they actually operate under different conditions. For example, sensors monitoring high-temperature furnaces and those monitoring ambient-temperature warehouses should have different normal energy consumption and performance baselines. Existing monitoring solutions typically use uniform evaluation thresholds and strategies for all nodes. This "one-size-fits-all" approach fails to accurately reflect the actual operating status of nodes under different conditions, easily leading to false alarms or missed alarms, resulting in extensive and inefficient operation and maintenance management.

[0005] Meanwhile, as the network scale expands, it becomes impractical to conduct isolated, manual evaluations of each sensor. Although there are some anomaly detection methods based on historical data, most of them remain at the level of individual nodes and fail to effectively explore and utilize the collaborative operation rules and patterns inherent in the sensor network. This makes it difficult for the monitoring system to grasp the overall network operation status from a higher dimension and to provide in-depth decision support for predictive maintenance and energy efficiency optimization.

[0006] Therefore, there is an urgent need in this field for a comprehensive evaluation method to achieve more accurate and intelligent monitoring of the operational status of industrial IoT sensor networks. Summary of the Invention

[0007] To improve the accuracy of sensor network monitoring, this application provides a sensor network monitoring method, system, and device based on the Industrial Internet of Things.

[0008] Firstly, this application provides a sensor network monitoring method based on the Industrial Internet of Things, employing the following technical solution:

[0009] A sensor network monitoring method based on the Industrial Internet of Things (IIoT) is applied to an IIoT system, which includes a management platform, a sensor network platform, and an object platform connected in sequence. The method is executed by the management platform and includes:

[0010] Acquire the performance time-series data and energy consumption time-series data of each sensor node in the sensor network, and cluster the sensor nodes according to the performance time-series data and the energy consumption time-series data to obtain at least one operating condition cluster;

[0011] For each sensor node, based on the performance time-series data and energy consumption time-series data corresponding to the sensor node, a first operating energy consumption correlation tensor for the sensor node is constructed, and a first operating energy consumption mode vector for the sensor node is determined according to the first operating energy consumption correlation tensor.

[0012] For the same operating condition cluster, a second operating energy consumption correlation matrix is ​​constructed based on the corresponding first operating energy consumption mode vector, and the second operating energy consumption mode vector of the operating condition cluster is determined based on the second operating energy consumption correlation matrix.

[0013] Based on the first operating energy consumption mode vector and the second operating energy consumption mode vector, the evaluation index of the sensor node is determined to evaluate the operating status of each sensor node in different clusters.

[0014] By adopting the above technical solution, performance time-series data and energy consumption time-series data of each sensor node in the sensor network are first acquired. Then, the sensor nodes are clustered based on these data to obtain at least one operating condition cluster. For each sensor node, a first operating energy consumption correlation tensor is constructed based on its corresponding performance and energy consumption time-series data. A first operating energy consumption pattern vector is then determined based on this tensor. For the same operating condition cluster, a second operating energy consumption correlation matrix is ​​constructed based on the corresponding first operating energy consumption pattern vector. A second operating energy consumption pattern vector is then determined based on this matrix. Finally, evaluation indicators for the sensor nodes are determined based on both the first and second operating energy consumption pattern vectors to assess the operating status of each sensor node within different clusters. This method overcomes the limitations of traditional monitoring methods, achieving accurate assessment and early warning of sensor node operating status, thereby improving the intelligence level of the entire sensor network monitoring and the efficiency of operation and maintenance management.

[0015] Optionally, the step of clustering the sensor nodes based on the performance time-series data and the energy consumption time-series data to obtain at least one operating condition cluster includes:

[0016] The performance time-series data and the energy consumption time-series data are preprocessed to obtain preprocessed data;

[0017] Based on the preprocessed data, statistical features are extracted from the sensor nodes to obtain feature vectors;

[0018] Based on the K-means algorithm, cluster analysis is performed on the sensor nodes according to the feature vectors to obtain at least one working condition cluster.

[0019] By adopting the above technical solution, in order to achieve the division of operating condition clusters, the performance time series data and energy consumption time series data are preprocessed to obtain preprocessed data. Then, based on the preprocessed data, statistical features of sensor nodes are extracted to obtain feature vectors. Then, based on the K-means algorithm, cluster analysis is performed on the sensor nodes according to the feature vectors to obtain at least one operating condition cluster.

[0020] Optionally, the step of constructing a first operating energy consumption correlation tensor for the sensor node based on the performance time-series data and the energy consumption time-series data corresponding to the sensor node includes:

[0021] Based on the performance time-series data and the energy consumption time-series data, a multi-parameter time-series matrix is ​​constructed, wherein the rows or columns of the multi-parameter time-series matrix are used to represent the performance parameters and energy consumption parameters at different time points;

[0022] Based on a preset fixed-length sliding time window, the multi-parameter time series matrix is ​​sampled by sliding to obtain multiple time series sub-matrices;

[0023] For each time series submatrix, high-level semantic features are extracted from the time series submatrix to obtain the corresponding correlation matrix, wherein the correlation matrix is ​​used to represent the inherent correlation pattern between each parameter channel after deep mining;

[0024] Based on the temporal order of the time-series submatrix, the correlation matrix is ​​stacked to form the first operating energy consumption correlation tensor of the sensor node.

[0025] By adopting the above technical solution, in order to construct the first operating energy consumption correlation tensor, a multi-parameter time series matrix is ​​constructed based on performance time series data and energy consumption time series data. The rows or columns of the multi-parameter time series matrix are used to represent the performance parameters and energy consumption parameters at different time points. Then, based on a preset fixed-length sliding time window, the multi-parameter time series matrix is ​​slidably acquired to obtain multiple time series sub-matrices. For each time series sub-matrix, high-level semantic features are extracted to obtain the corresponding correlation matrix. The correlation matrix is ​​used to represent the inherent correlation pattern between the parameter channels after deep mining. Then, based on the time order of the time series sub-matrices, the correlation matrices are stacked to form the first operating energy consumption correlation tensor of the sensor node.

[0026] Optionally, the step of extracting high-level semantic features from the temporal submatrix to obtain the corresponding association matrix includes:

[0027] Semantic embedding is performed on the time-series submatrix to obtain semantic embedding features, wherein the semantic embedding features include multiple feature channels, and the feature channels correspond to the parameters of the sensor;

[0028] In-channel depth mining is performed on each of the feature channels to form in-channel enhanced features corresponding to each feature channel. In the process of in-channel depth mining, after performing depth mining at different levels on the feature channels, correlation mining is performed on the mined depth features.

[0029] For each of the in-channel enhancement features, perform out-of-channel correlation mining to form out-of-channel correlation features corresponding to each of the in-channel enhancement features;

[0030] The correlation matrix is ​​obtained by concatenating the external correlation features of the channel.

[0031] By adopting the above technical solution, in order to obtain the correlation matrix, semantic embedding is performed on the time-series submatrix to obtain semantic embedding features. The semantic embedding features include multiple feature channels, and the feature channels correspond to the parameters of the sensor. Then, in-channel depth mining is performed on each feature channel to form in-channel enhancement features corresponding to each feature channel. In the process of in-channel depth mining, after performing depth mining at different levels on the feature channels, correlation mining is performed on the mined depth features. Then, out-of-channel correlation mining is performed on each in-channel enhancement feature to form out-of-channel correlation features corresponding to each in-channel enhancement feature. Finally, the out-of-channel correlation features are concatenated to obtain the correlation matrix.

[0032] Optionally, the step of performing out-of-channel correlation mining on each of the intra-channel enhancement features to form out-of-channel correlation features corresponding to each of the intra-channel enhancement features includes:

[0033] For each channel-in-channel enhancement feature, the channel-in-channel enhancement feature is taken as the target enhancement feature. The correlation parameters between the target enhancement feature and each non-target enhancement feature are determined. Based on the correlation parameters, the target enhancement feature is adjusted and the target enhancement adjustment feature is output.

[0034] The enhancement and adjustment features of each target are fused to obtain the out-of-channel correlation features.

[0035] By adopting the above technical solution, in order to generate out-of-channel correlation features, for each in-channel enhancement feature, the in-channel enhancement feature is taken as the target enhancement feature, the correlation parameters between the target enhancement feature and each non-target enhancement feature are determined, and the target enhancement feature is adjusted based on the correlation parameters to output the target enhancement adjusted feature. Then, the target enhancement adjusted features are fused to obtain the out-of-channel correlation features.

[0036] Optionally, the step of determining the first operating energy consumption mode vector of the sensor node based on the first operating energy consumption correlation tensor includes:

[0037] The first operating energy consumption correlation tensor is decomposed by Tucker to obtain the core tensor, the first factor matrix, the second factor matrix and the third factor matrix, wherein the first factor matrix, the second factor matrix and the third factor matrix correspond to the row dimension, the column dimension and the time dimension, respectively.

[0038] The element with the largest value is determined from the core tensor, and the corresponding eigenvector in the first factor matrix is ​​determined based on the index of this element in the row dimension of the core tensor.

[0039] The feature vector is normalized to obtain the first operating energy consumption mode vector of the sensor node.

[0040] By adopting the above technical solution, in order to determine the first operating energy consumption mode vector, the first operating energy consumption correlation tensor is decomposed by Tucker to obtain the core tensor, the first factor matrix, the second factor matrix, and the third factor matrix. The first factor matrix, the second factor matrix, and the third factor matrix correspond to the row dimension, the column dimension, and the time dimension, respectively. Then, the element with the largest value is determined from the core tensor, and the corresponding feature vector in the first factor matrix is ​​determined according to the index of the element in the row dimension of the core tensor. Then, the feature vector is normalized to obtain the first operating energy consumption mode vector of the sensor node.

[0041] Optionally, the step of determining the second operating energy consumption mode vector of the operating condition cluster based on the second operating energy consumption correlation matrix includes:

[0042] Initialize a random, non-zero initial vector v. 0 and for the initial vector v 0 Normalization is performed so that ||v 0 ||2=1;

[0043] Based on a predefined iterative formula, the second operating energy consumption correlation matrix is ​​iteratively calculated, wherein the calculation formula for the (k+1)th iteration is: C is the second operating energy consumption correlation matrix, v k This is the vector obtained from the k-th iteration;

[0044] For each iteration, calculate v k+1 With v k If the L2 norm of the difference between the two values ​​is less than a preset convergence threshold, then the iteration stops.

[0045] The final iteration result v final This serves as the second operating energy consumption mode vector for this operating condition cluster.

[0046] By adopting the above technical solution, in order to determine the second operating energy consumption mode vector, a random non-zero initial vector v is initialized. 0 and the initial vector v 0 Normalization is performed so that ||v 0 ||2=1, and then based on the predefined iterative formula, the second operating energy consumption correlation matrix is ​​iteratively calculated, where the calculation formula for the (k+1)th iteration is: C is the second operating energy consumption correlation matrix, v k Let v be the vector obtained in the k-th iteration. Then, for each iteration, calculate v. k+1 With vk If the L2 norm of the difference between the values ​​is less than a preset convergence threshold, the iteration stops, and the final iteration result v is then calculated. final This serves as the second operating energy consumption mode vector for this operating condition cluster.

[0047] Optionally, the step of determining the evaluation index of the sensor node based on the first operating energy consumption pattern vector and the second operating energy consumption pattern vector includes:

[0048] The first operating energy consumption mode vector and the second operating energy consumption mode vector are respectively transformed into probability distributions to obtain a first probability distribution vector and a second probability distribution vector.

[0049] Calculate the JS divergence between the first probability distribution vector and the second probability distribution vector, and determine the evaluation index of the sensor node based on the JS divergence.

[0050] By adopting the above technical solution, in order to determine the evaluation index of the sensor node, the first operating energy consumption mode vector and the second operating energy consumption mode vector are transformed into probability distributions to obtain the first probability distribution vector and the second probability distribution vector. Then, the JS divergence between the first probability distribution vector and the second probability distribution vector is calculated, and the evaluation index of the sensor node is determined based on the JS divergence.

[0051] Secondly, this application also provides a sensor network monitoring system based on the Industrial Internet of Things, which adopts the following technical solution:

[0052] A sensor network monitoring system based on the Industrial Internet of Things (IIoT) includes a management platform, a sensor network platform, and an object platform that are sequentially connected in communication. The management platform is configured with:

[0053] The node partitioning module is used to acquire the performance time-series data and energy consumption time-series data of each sensor node in the sensor network, and to cluster the sensor nodes according to the performance time-series data and the energy consumption time-series data to obtain at least one operating condition cluster.

[0054] The node mode determination module is used to construct a first operating energy consumption correlation tensor for each sensor node based on the performance time-series data and energy consumption time-series data corresponding to the sensor node, and determine the first operating energy consumption mode vector of the sensor node based on the first operating energy consumption correlation tensor.

[0055] The cluster mode determination module is used to construct a second operating energy consumption correlation matrix for the same operating condition cluster based on the corresponding first operating energy consumption mode vector, and determine the second operating energy consumption mode vector for the operating condition cluster based on the second operating energy consumption correlation matrix.

[0056] The evaluation module is used to determine the evaluation index of the sensor node based on the first operating energy consumption mode vector and the second operating energy consumption mode vector, so as to evaluate the operating status of each sensor node in different clusters.

[0057] Thirdly, this application also provides a computer device, which adopts the following technical solution:

[0058] A computer device includes a memory and a processor, the memory storing a computer program executable on the processor, the processor executing the computer program to implement the method described in the first aspect.

[0059] In summary, this application includes at least the following beneficial technical effects: First, it acquires the performance time-series data and energy consumption time-series data of each sensor node in the sensor network, and clusters the sensor nodes based on the performance time-series data and energy consumption time-series data to obtain at least one operating condition cluster. Then, for each sensor node, based on the corresponding performance time-series data and energy consumption time-series data, it constructs a first operating energy consumption correlation tensor for that sensor node, and determines a first operating energy consumption pattern vector for that sensor node based on the first operating energy consumption correlation tensor. Then, for the same operating condition cluster, it constructs a second operating energy consumption correlation matrix for that operating condition cluster based on the corresponding first operating energy consumption pattern vector, and determines a second operating energy consumption pattern vector for that operating condition cluster based on the second operating energy consumption correlation matrix. Finally, it determines the evaluation index of the sensor node based on the first and second operating energy consumption pattern vectors to evaluate the operating status of each sensor node in different clusters. Through the above method, the extensiveness of traditional monitoring methods is overcome; accurate evaluation and early warning of the operating status of sensor nodes are achieved, thereby improving the intelligence level of the entire sensor network monitoring and the efficiency of operation and maintenance management. Attached Figure Description

[0060] Figure 1 This is a schematic diagram of the overall process of an embodiment of this application.

[0061] Figure 2 This is a structural diagram of one application scenario of the system according to an embodiment of this application.

[0062] Figure 3 This is a structural diagram of another application scenario of the system according to an embodiment of this application.

[0063] Figure 4 This is a structural block diagram of the computer device described in this application. Detailed Implementation

[0064] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0065] This application discloses a sensor network monitoring method based on the Industrial Internet of Things (IIoT).

[0066] Reference Figure 1 A sensor network monitoring method based on the Industrial Internet of Things (IIoT), characterized in that it is applied to an IIoT system, which includes a management platform, a sensor network platform, and an object platform connected in sequence. The method is executed by the management platform and includes:

[0067] Step S11: Obtain the performance time-series data and energy consumption time-series data of each sensor node in the sensor network, and cluster the sensor nodes according to the performance time-series data and energy consumption time-series data to obtain at least one operating condition cluster.

[0068] It's important to note that industrial environments are complex, and sensor nodes may be distributed across different devices, in different physical environments, or performing different monitoring tasks, leading to inherent differences in their operating load and status. The core objective of this step is to abandon the traditional "one-size-fits-all" monitoring strategy. By performing cluster analysis on two key time-series data points reflecting node operating status—performance (such as accuracy and response speed) and energy consumption—the entire sensor network is automatically divided into several "condition clusters" with similar operating characteristics. This is equivalent to establishing refined management groups for the network, enabling subsequent analysis to be based on a unified benchmark under the same operating condition, laying the foundation for accurate assessment.

[0069] Step S12: For each sensor node, based on the performance time-series data and energy consumption time-series data corresponding to the sensor node, construct the first operating energy consumption correlation tensor of the sensor node, and determine the first operating energy consumption mode vector of the sensor node according to the first operating energy consumption correlation tensor.

[0070] It should be noted that this step aims to deeply explore the operational characteristics of each individual sensor node. Performance and energy consumption are not isolated; there is a dynamic and complex correlation between them. The first operational energy consumption correlation tensor is a high-order data structure that can transcend simple scalars or matrices, fusing and characterizing the deep coupling patterns between performance index fluctuations and energy consumption changes across multiple spatiotemporal scales. By decomposing or extracting features from this tensor, a simplified first operational energy consumption pattern vector can be condensed. This vector is a unique and highly generalized operational fingerprint of the node, accurately representing its inherent performance-energy consumption correlation pattern under normal conditions.

[0071] Step S13: For the same operating condition cluster, construct the second operating energy consumption correlation matrix of the operating condition cluster based on the corresponding first operating energy consumption mode vector, and determine the second operating energy consumption mode vector of the operating condition cluster based on the second operating energy consumption correlation matrix.

[0072] It should be noted that after obtaining the individual characteristics of all nodes, this step aims to extract the common characteristics of the entire group. By integrating the first operating energy consumption pattern vectors of all nodes within the same operating condition cluster, a second operating energy consumption correlation matrix is ​​constructed. This matrix reveals the optimal or typical operating paradigm that the sensor nodes in this group should follow under the same or similar operating conditions. By performing principal component analysis or feature extraction on this matrix, a second operating energy consumption pattern vector can be further obtained. This vector represents the group standard pattern of the operating condition cluster and serves as a benchmark for measuring whether the operating status of any node within the cluster is healthy or deviates from the group's.

[0073] Step S14: Determine the evaluation index of the sensor node based on the first operating energy consumption mode vector and the second operating energy consumption mode vector, so as to evaluate the operating status of each sensor node in different clusters.

[0074] It should be noted that the evaluation is not conducted in isolation, but rather by calculating the similarity or analyzing the difference between the individual operating fingerprint of each node (the first operating energy consumption pattern vector) and the group standard pattern of its cluster (the second operating energy consumption pattern vector). The resulting evaluation index is a relatively quantitative value that considers both the node's own state changes and its relative performance among similar nodes. This method enables operations and maintenance personnel to clearly identify which nodes are operating robustly in different operating condition clusters, and which nodes, although their individual indicators have not exceeded the absolute threshold, have begun to deviate from the group's normal operating pattern. This allows for early and accurate detection and warning of potential faults, performance degradation, or energy efficiency anomalies.

[0075] In the above implementation, firstly, performance time-series data and energy consumption time-series data of each sensor node in the sensor network are acquired. Then, the sensor nodes are clustered based on these data to obtain at least one operating condition cluster. For each sensor node, a first operating energy consumption correlation tensor is constructed based on its corresponding performance and energy consumption time-series data. A first operating energy consumption pattern vector is then determined based on this tensor. For the same operating condition cluster, a second operating energy consumption correlation matrix is ​​constructed based on the corresponding first operating energy consumption pattern vector. A second operating energy consumption pattern vector is then determined based on this matrix. Finally, evaluation indicators for the sensor nodes are determined based on both the first and second operating energy consumption pattern vectors to assess the operating status of each sensor node within different clusters. This method overcomes the limitations of traditional monitoring methods, achieving accurate assessment and early warning of sensor node operating status, thereby improving the intelligence level of the entire sensor network monitoring and the efficiency of operation and maintenance management.

[0076] As a further implementation of the method, the step of clustering sensor nodes based on performance time-series data and energy consumption time-series data to obtain at least one operating condition cluster includes:

[0077] Step S21: Preprocess the performance time-series data and energy consumption time-series data to obtain preprocessed data.

[0078] Step S22: Based on the preprocessed data, perform statistical feature extraction on the sensor nodes to obtain feature vectors.

[0079] Step S23: Based on the K-means algorithm, perform cluster analysis on the sensor nodes according to the feature vectors to obtain at least one working condition cluster.

[0080] It should be noted that steps S21 to S23 together constitute a complete and automated data processing flow for clustering sensor nodes based on their operating conditions. Specifically, the original performance time-series data and energy consumption time-series data are first preprocessed, including cleaning, denoising, and standardization, to improve data quality. Next, in step S22, the system extracts key statistical features that characterize the node's operating status from the preprocessed clean data, thus condensing the lengthy time-series data into feature vectors containing information density, achieving data dimensionality reduction and feature representation. Finally, based on these feature vectors, the classic and efficient K-means clustering algorithm is used to automatically divide the nodes in the entire sensor network into different operating condition clusters, thereby achieving the goal of scientifically grouping nodes based on their actual operating behavior.

[0081] In the above implementation, in order to divide the operating condition clusters, the performance time series data and energy consumption time series data are preprocessed to obtain preprocessed data. Then, based on the preprocessed data, statistical features of the sensor nodes are extracted to obtain feature vectors. Then, based on the K-means algorithm, cluster analysis is performed on the sensor nodes according to the feature vectors to obtain at least one operating condition cluster.

[0082] As a further implementation of the method, the step of constructing a first operational energy consumption correlation tensor for the sensor node based on the performance time-series data and energy consumption time-series data corresponding to the sensor node includes:

[0083] Step S31: Construct a multi-parameter time series matrix based on performance time series data and energy consumption time series data, wherein the rows or columns of the multi-parameter time series matrix are used to represent performance parameters and energy consumption parameters at different time points.

[0084] Step S32: Based on a preset fixed-length sliding time window, slide acquisition is performed on the multi-parameter time series matrix to obtain multiple time series sub-matrices.

[0085] Step S33: For each time series submatrix, perform high-level semantic feature extraction on the time series submatrix to obtain the corresponding correlation matrix, where the correlation matrix is ​​used to represent the inherent correlation pattern between each parameter channel after deep mining.

[0086] Step S34: Based on the temporal order of the temporal submatrix, stack the correlation matrix to form the first operating energy consumption correlation tensor of the sensor node.

[0087] It should be noted that steps S31 to S34 together construct a complete processing flow from raw time-series data to high-order feature representation. Specifically, firstly, by integrating the time-series data of the two key parameters, performance and energy consumption, into a unified multi-parameter time-series matrix, a multi-dimensional synchronous description of the sensor node's operating state is achieved. Based on this, step S32 uses a sliding time window of a preset length to truncate the matrix, thereby dividing the continuous operation process into a series of continuous time-series segments, effectively capturing the dynamic local features of node operation. Subsequently, step S33 performs deep feature learning and mining on each time-series sub-matrix to generate a semantic feature matrix that can profoundly reveal the complex nonlinear correlation pattern between performance parameters and energy consumption parameters. Finally, step S34 stacks the above series of correlation matrices arranged in chronological order in the time dimension to form a first operating energy consumption correlation tensor that can comprehensively characterize the dynamic correlation characteristics of performance and energy consumption of the sensor node throughout the entire operating cycle, thereby transforming the raw, planar time-series data into a high-order data structure rich in spatiotemporal correlation information.

[0088] In the above implementation, in order to construct the first operating energy consumption correlation tensor, a multi-parameter time series matrix is ​​constructed based on performance time series data and energy consumption time series data. The rows or columns of the multi-parameter time series matrix are used to represent the performance parameters and energy consumption parameters at different time points. Then, based on a preset fixed-length sliding time window, the multi-parameter time series matrix is ​​slidably acquired to obtain multiple time series sub-matrices. For each time series sub-matrix, high-level semantic features are extracted to obtain the corresponding correlation matrix. The correlation matrix is ​​used to represent the inherent correlation pattern between the parameter channels after deep mining. Then, based on the time order of the time series sub-matrices, the correlation matrices are stacked to form the first operating energy consumption correlation tensor of the sensor node.

[0089] As a further implementation of the method, the step of extracting high-level semantic features from the temporal submatrix to obtain the corresponding association matrix includes:

[0090] Step S41: Semantic embedding is performed on the time-series submatrix to obtain semantic embedding features. The semantic embedding features include multiple feature channels, and the feature channels correspond to the parameters of the sensor.

[0091] Step S42: Perform in-channel depth mining on each feature channel to form in-channel enhanced features corresponding to each feature channel. In the process of in-channel depth mining, after performing depth mining at different levels on the feature channels, perform correlation mining on the mined depth features.

[0092] Step S43: Perform out-of-channel correlation mining on each channel-intra-channel enhanced feature to form the out-of-channel correlation feature corresponding to each channel-intra-channel enhanced feature.

[0093] Step S44: Concatenate the external correlation features to obtain the correlation matrix.

[0094] It should be noted that step S41 uses semantic embedding technology to transform the original parameters in the temporal submatrix into a feature space containing richer semantic information, forming a multi-feature channel representation with clear physical meaning; step S42 focuses on the internal structure of each feature channel, and through multi-level, in-depth feature mining and internal correlation analysis, it achieves enhanced expression of the potential laws within a single channel, forming enhanced features within the channel; step S43 expands the analysis perspective from within the channel to between channels, systematically mining the mutual correlation and synergistic patterns between different feature channels, thereby capturing the deep coupling relationship between parameters, forming external correlation features; step S44 effectively integrates and splices all external correlation features to generate a correlation matrix that can comprehensively and profoundly represent the complex intrinsic correlation patterns between multiple parameters, thus realizing the progression from preliminary feature representation to in-depth correlation mining, providing key input for constructing high-order feature tensors.

[0095] In the above implementation, in order to obtain the correlation matrix, semantic embedding is performed on the time-series submatrix to obtain semantic embedding features. The semantic embedding features include multiple feature channels, and the feature channels correspond to the parameters of the sensor. Then, in-channel depth mining is performed on each feature channel to form in-channel enhancement features corresponding to each feature channel. During the in-channel depth mining process, after performing depth mining at different levels on the feature channels, correlation mining is performed on the mined depth features. Then, out-of-channel correlation mining is performed on each in-channel enhancement feature to form out-of-channel correlation features corresponding to each in-channel enhancement feature. Finally, the out-of-channel correlation features are concatenated to obtain the correlation matrix.

[0096] As a further implementation of the method, the step of performing out-of-channel correlation mining on each channel-intra-channel enhancement feature to form the out-of-channel correlation feature corresponding to each channel-intra-channel enhancement feature includes:

[0097] Step S51: For each channel-in-channel enhancement feature, the channel-in-channel enhancement feature is taken as the target enhancement feature. The correlation parameters between the target enhancement feature and each non-target enhancement feature are determined respectively. Based on the correlation parameters, the target enhancement feature is adjusted and the target enhancement adjustment feature is output.

[0098] Step S52: The enhancement and adjustment features of each target are fused to obtain the external correlation features.

[0099] It should be noted that step S51 processes the enhanced features within each channel: it takes turns using them as target enhanced features and calculates the correlation parameters between them and all other non-target enhanced features. These correlation parameters quantify the degree of interdependence between different feature channels. The system then dynamically adjusts the target enhanced features based on these parameters to include correlation information from other channels, thereby outputting more context-aware target enhanced adjusted features. Step S52 unifies and merges the target enhanced adjusted features obtained after the above interactive adjustment of all channels, and finally generates out-of-channel correlation features that can comprehensively and accurately represent the complex correlation patterns between multi-parameter channels. This establishes an explicit correlation model between channels and performs feature adjustment and fusion, effectively improving the richness and accuracy of feature expression.

[0100] In the above implementation, in order to generate out-of-channel correlation features, for each in-channel enhancement feature, the in-channel enhancement feature is taken as the target enhancement feature, the correlation parameters between the target enhancement feature and each non-target enhancement feature are determined, and the target enhancement feature is adjusted based on the correlation parameters to output the target enhancement adjustment feature. Then, the target enhancement adjustment features are fused to obtain the out-of-channel correlation features.

[0101] As a further implementation of the method, the step of determining the first operating energy consumption mode vector of the sensor node based on the first operating energy consumption correlation tensor includes:

[0102] Step S61: Perform Tucker decomposition on the first operating energy consumption correlation tensor to obtain the core tensor, the first factor matrix, the second factor matrix, and the third factor matrix, where the first factor matrix, the second factor matrix, and the third factor matrix correspond to the row dimension, the column dimension, and the time dimension, respectively.

[0103] Step S62: Determine the element with the largest value from the core tensor, and determine the corresponding eigenvector in the first factor matrix based on the index of this element in the row dimension of the core tensor.

[0104] Step S63: Normalize the feature vector to obtain the first operating energy consumption mode vector of the sensor node.

[0105] It should be noted that step S61 decomposes the constructed first operational energy consumption correlation tensor into a core tensor and three factor matrices corresponding to the row, column, and time dimensions, respectively. This decomposition operation can significantly reduce the complexity of the data while preserving the essential correlation structure between the dimensions of the tensor. Step S62 identifies the element with the largest value in the core tensor and uses its index in the row dimension as a guide to accurately locate the potential feature vector that best represents the core correlation pattern between the sensor node's performance and energy consumption from the first factor matrix. Step S63 normalizes the feature vector to eliminate the influence of dimensions and transforms it into a standardized first operational energy consumption pattern vector that can be used for subsequent comparison and calculation. This achieves the extraction of a highly representative and easily processed low-dimensional feature vector from complex high-dimensional tensor data.

[0106] In the above implementation, in order to determine the first operating energy consumption mode vector, the first operating energy consumption correlation tensor is decomposed by Tucker to obtain a core tensor, a first factor matrix, a second factor matrix, and a third factor matrix. The first factor matrix, the second factor matrix, and the third factor matrix correspond to the row dimension, the column dimension, and the time dimension, respectively. Then, the element with the largest value is determined from the core tensor, and the corresponding feature vector in the first factor matrix is ​​determined according to the index of the element in the row dimension of the core tensor. Then, the feature vector is normalized to obtain the first operating energy consumption mode vector of the sensor node.

[0107] As a further implementation of the method, the step of determining the second operating energy consumption mode vector of the operating condition cluster based on the second operating energy consumption correlation matrix includes:

[0108] Step S71: Initialize a random non-zero initial vector v 0 and the initial vector v 0 Normalization is performed so that ||v 0 ||2=1.

[0109] Step S72: Based on a predefined iterative formula, the second operating energy consumption correlation matrix is ​​iteratively calculated, wherein the calculation formula for the (k+1)th iteration is: C is the second operating energy consumption correlation matrix, v k This is the vector obtained in the k-th iteration.

[0110] Step S73, for each iteration, calculate v k+1 With v k If the L2 norm of the difference between the two values ​​is less than the preset convergence threshold, the iteration stops.

[0111] Step S74, the final iteration result v final This serves as the second operating energy consumption mode vector for this operating condition cluster.

[0112] It should be noted that step S71 initializes a random non-zero vector and performs normalization, setting a numerically stable starting point for the iterative calculation; in step S72, the system repeatedly calculates the second operating energy consumption correlation matrix based on a predefined iterative formula. This formula, by multiplying the matrix with the current vector and then normalizing it again, ensures that the vector continuously approaches the main characteristic direction of the matrix in each iteration; step S73 monitors the difference between adjacent iteration results to ensure that the algorithm automatically terminates when the calculation result is stable, thereby achieving a balance between calculation accuracy and efficiency; in step S74, the final iteration result that meets the convergence condition is confirmed as the second operating energy consumption mode vector of the operating condition cluster, thereby extracting a benchmark mode that can represent the common operating characteristics of the entire operating condition cluster from the group operating data.

[0113] In the above implementation, in order to determine the second operating energy consumption mode vector, a random non-zero initial vector v is initialized. 0 and the initial vector v 0 Normalization is performed so that ||v 0 ||2=1, and then based on the predefined iterative formula, the second operating energy consumption correlation matrix is ​​iteratively calculated, where the calculation formula for the (k+1)th iteration is: C is the second operating energy consumption correlation matrix, v k Let v be the vector obtained in the k-th iteration. Then, for each iteration, calculate v. k+1 With v k If the L2 norm of the difference between the values ​​is less than a preset convergence threshold, the iteration stops, and the final iteration result v is then calculated.final This serves as the second operating energy consumption mode vector for this operating condition cluster.

[0114] As a further implementation of the method, the step of determining the evaluation index of the sensor node based on the first operating energy consumption mode vector and the second operating energy consumption mode vector includes:

[0115] Step S81: Perform probability distribution transformation on the first operating energy consumption mode vector and the second operating energy consumption mode vector respectively to obtain the first probability distribution vector and the second probability distribution vector.

[0116] Step S82: Calculate the JS divergence between the first probability distribution vector and the second probability distribution vector, and determine the evaluation index of the sensor node based on the JS divergence.

[0117] It should be noted that in step S81, the first operating energy consumption mode vector and the second operating energy consumption mode vector are respectively transformed into probability distribution vectors. This transformation process maps the abstract vector features into a comparable distribution form that conforms to the probability axiom. Subsequently, step S82 calculates the JS divergence between these two probability distribution vectors to accurately measure their distribution differences. As a symmetric and smooth information measurement tool, JS divergence can stably reflect the degree of deviation between individual operating modes and group standard modes. Finally, the evaluation index of the sensor node is determined based on this divergence value, which has been rigorously derived mathematically, so that the evaluation results can intuitively reflect the relative health status of the node operation, thereby realizing a reliable conversion from feature space to evaluation index.

[0118] In the above implementation, in order to determine the evaluation index of the sensor node, the first operating energy consumption mode vector and the second operating energy consumption mode vector are respectively transformed into probability distributions to obtain the first probability distribution vector and the second probability distribution vector. Then, the JS divergence between the first probability distribution vector and the second probability distribution vector is calculated, and the evaluation index of the sensor node is determined based on the JS divergence.

[0119] This application also discloses a sensor network monitoring system based on the Industrial Internet of Things.

[0120] refer to Figure 2 A sensor network monitoring system based on the Industrial Internet of Things (IIoT) includes a management platform, a sensor network platform, and an object platform that are sequentially connected in communication. The management platform is configured with:

[0121] The node partitioning module is used to acquire the performance time-series data and energy consumption time-series data of each sensor node in the sensor network, and to cluster the sensor nodes based on the performance time-series data and energy consumption time-series data to obtain at least one operating condition cluster.

[0122] The node mode determination module is used to construct a first operating energy consumption correlation tensor for each sensor node based on the performance time-series data and energy consumption time-series data corresponding to the sensor node, and determine the first operating energy consumption mode vector of the sensor node based on the first operating energy consumption correlation tensor.

[0123] The cluster mode determination module is used to construct a second operating energy consumption correlation matrix for the same operating condition cluster based on the corresponding first operating energy consumption mode vector, and determine the second operating energy consumption mode vector for the operating condition cluster based on the second operating energy consumption correlation matrix.

[0124] The evaluation module is used to determine the evaluation index of the sensor node based on the first operating energy consumption mode vector and the second operating energy consumption mode vector, so as to evaluate the operating status of each sensor node in different clusters.

[0125] The overall framework of another application scenario of the sensor network monitoring system based on the Industrial Internet of Things of this invention is as follows: Figure 3 As shown, the system can include a user platform, a service platform, a management platform, a sensor network platform, and an object platform that interact sequentially, forming a five-platform architecture based on the Industrial Internet of Things (IIoT). The service platform consists of a main service database, multiple service sub-platforms, and multiple service sub-databases. The management platform includes a feature value generation module, a feature construction module, a feature decomposition module, an inspection quality index generation module, and an inspection personnel allocation module. The management platform can interact with the sensor network platform and the service platform. The sensor network platform can include a main sensor database, multiple sensor network sub-platforms, and multiple sensor sub-databases. In this embodiment, there are n sensor network sub-platforms and n sensor sub-databases. Each sensor network sub-platform has a corresponding sensor sub-database. The sensor network platform can interact with the object platform.

[0126] By leveraging the interaction between various functional platforms of the industrial IoT-based sensor network monitoring system, which is based on the aforementioned three or five platforms, a complete closed-loop information operation logic is established, ensuring the orderly operation of sensing and control information and realizing intelligent equipment management.

[0127] Specifically, the sensor network monitoring system based on the Industrial Internet of Things in this embodiment includes a management platform. The management platform is configured to: acquire performance time-series data and energy consumption time-series data of each sensor node in the sensor network, and cluster the sensor nodes according to the performance time-series data and energy consumption time-series data to obtain at least one operating condition cluster; for each sensor node, construct a first operating energy consumption correlation tensor for the sensor node based on the corresponding performance time-series data and energy consumption time-series data, and determine a first operating energy consumption pattern vector for the sensor node according to the first operating energy consumption correlation tensor; for the same operating condition cluster, construct a second operating energy consumption correlation matrix for the operating condition cluster according to the corresponding first operating energy consumption pattern vector, and determine a second operating energy consumption pattern vector for the operating condition cluster according to the second operating energy consumption correlation matrix; and determine the evaluation index of the sensor node according to the first operating energy consumption pattern vector and the second operating energy consumption pattern vector to evaluate the operating status of each sensor node in different clusters.

[0128] The sensor network monitoring system based on the Industrial Internet of Things (IIoT) of the present invention can implement any of the sensor network monitoring methods based on the Industrial Internet of Things, and the specific working process of the sensor network monitoring system based on the Industrial Internet of Things of the present invention can refer to the corresponding process in the above-mentioned sensor network monitoring methods based on the Industrial Internet of Things.

[0129] This application also discloses a computer device.

[0130] refer to Figure 4 A computer device includes a memory and a processor, wherein the memory stores a computer program that can run on the processor, and the processor executes the computer program to implement any of the above-described sensor network monitoring methods based on the Industrial Internet of Things.

[0131] The above are all preferred embodiments of this application and are not intended to limit the scope of protection of this application. Any feature disclosed in this specification (including the abstract and drawings) may be replaced by other equivalent or similar features unless specifically stated otherwise. That is, unless specifically stated otherwise, each feature is only one example of a series of equivalent or similar features.

Claims

1. A sensor network monitoring method based on industrial internet of things, characterized in that, Applied to an industrial Internet of Things (IIoT) system, the IIoT system includes a management platform, a sensor network platform, and an object platform that are sequentially connected in communication. The method is executed by the management platform and includes: Acquire the performance time-series data and energy consumption time-series data of each sensor node in the sensor network, and cluster the sensor nodes according to the performance time-series data and the energy consumption time-series data to obtain at least one operating condition cluster; For each sensor node, based on the performance time-series data and energy consumption time-series data corresponding to the sensor node, a first operating energy consumption correlation tensor for the sensor node is constructed, and a first operating energy consumption mode vector for the sensor node is determined according to the first operating energy consumption correlation tensor. For the same operating condition cluster, a second operating energy consumption correlation matrix is ​​constructed based on the corresponding first operating energy consumption mode vector, and a second operating energy consumption mode vector for the operating condition cluster is determined based on the second operating energy consumption correlation matrix. Based on the first operating energy consumption mode vector and the second operating energy consumption mode vector, the evaluation index of the sensor node is determined to evaluate the operating status of each sensor node in different clusters. The step of constructing a first operating energy consumption correlation tensor for the sensor node based on the performance time-series data and energy consumption time-series data corresponding to the sensor node includes: Based on the performance time-series data and the energy consumption time-series data, a multi-parameter time-series matrix is ​​constructed, wherein the rows or columns of the multi-parameter time-series matrix are used to represent the performance parameters and energy consumption parameters at different time points; Based on a preset fixed-length sliding time window, the multi-parameter time series matrix is ​​sampled by sliding to obtain multiple time series sub-matrices; For each time series submatrix, high-level semantic features are extracted from the time series submatrix to obtain the corresponding correlation matrix, wherein the correlation matrix is ​​used to represent the inherent correlation pattern between each parameter channel after deep mining; Based on the temporal order of the temporal submatrix, the correlation matrix is ​​stacked to form the first operating energy consumption correlation tensor of the sensor node; The step of determining the first operating energy consumption mode vector of the sensor node based on the first operating energy consumption correlation tensor includes: The first operating energy consumption correlation tensor is decomposed by Tucker to obtain the core tensor, the first factor matrix, the second factor matrix and the third factor matrix, wherein the first factor matrix, the second factor matrix and the third factor matrix correspond to the row dimension, the column dimension and the time dimension, respectively. The element with the largest value is determined from the core tensor, and the corresponding eigenvector in the first factor matrix is ​​determined based on the index of this element in the row dimension of the core tensor. The feature vector is normalized to obtain the first operating energy consumption mode vector of the sensor node; The step of determining the second operating energy consumption mode vector of the operating condition cluster based on the second operating energy consumption correlation matrix includes: Initialize a random, non-zero initial vector v. 0 and for the initial vector v 0 Normalization is performed so that ||v 0 ||2=1; Based on a predefined iterative formula, the second operating energy consumption correlation matrix is ​​iteratively calculated, wherein the calculation formula for the (k+1)th iteration is: C is the second operating energy consumption correlation matrix, v k This is the vector obtained from the k-th iteration; For each iteration, calculate v k+1 With v k If the L2 norm of the difference between the two values ​​is less than a preset convergence threshold, then the iteration stops. The final iteration result v final as the second operation energy consumption mode vector of the working condition cluster. 2.The industrial Internet of things based sensor network monitoring method according to claim 1, characterized in that, The step of clustering the sensor nodes based on the performance time-series data and the energy consumption time-series data to obtain at least one operating condition cluster includes: The performance time-series data and the energy consumption time-series data are preprocessed to obtain preprocessed data; Based on the preprocessed data, statistical features are extracted from the sensor nodes to obtain feature vectors; Based on the K-means algorithm, cluster analysis is performed on the sensor nodes according to the feature vectors to obtain at least one working condition cluster. 3.The industrial Internet of things based sensor network monitoring method according to claim 1, characterized in that, The step of extracting high-level semantic features from the temporal submatrix to obtain the corresponding association matrix includes: Semantic embedding is performed on the time-series submatrix to obtain semantic embedding features, wherein the semantic embedding features include multiple feature channels, and the feature channels correspond to the parameters of the sensor; In-channel depth mining is performed on each of the feature channels to form in-channel enhanced features corresponding to each feature channel. In the process of in-channel depth mining, after performing depth mining at different levels on the feature channels, correlation mining is performed on the mined depth features. For each of the in-channel enhancement features, perform out-of-channel correlation mining to form out-of-channel correlation features corresponding to each of the in-channel enhancement features; The correlation matrix is ​​obtained by concatenating the external correlation features of the channel. 4.The industrial Internet of things based sensor network monitoring method according to claim 3, characterized in that, The step of performing out-of-channel correlation mining on each of the intra-channel enhancement features to form the out-of-channel correlation features corresponding to each of the intra-channel enhancement features includes: For each channel-in-channel enhancement feature, the channel-in-channel enhancement feature is taken as the target enhancement feature. The correlation parameters between the target enhancement feature and each non-target enhancement feature are determined. Based on the correlation parameters, the target enhancement feature is adjusted and the target enhancement adjustment feature is output. The enhancement and adjustment features of each target are fused to obtain the out-of-channel correlation features.

5. The sensor network monitoring method based on the Industrial Internet of Things according to claim 1, characterized in that, The step of determining the evaluation index of the sensor node based on the first operating energy consumption mode vector and the second operating energy consumption mode vector includes: The first operating energy consumption mode vector and the second operating energy consumption mode vector are respectively transformed into probability distributions to obtain a first probability distribution vector and a second probability distribution vector. Calculate the JS divergence between the first probability distribution vector and the second probability distribution vector, and determine the evaluation index of the sensor node based on the JS divergence.

6. A sensor network monitoring system based on the Industrial Internet of Things, characterized in that, It includes a management platform, a sensor network platform, and an object platform that are connected in sequence. The management platform is configured with: The node partitioning module is used to acquire the performance time-series data and energy consumption time-series data of each sensor node in the sensor network, and to cluster the sensor nodes according to the performance time-series data and the energy consumption time-series data to obtain at least one operating condition cluster. The node mode determination module is used to construct a first operating energy consumption correlation tensor for each sensor node based on the performance time-series data and energy consumption time-series data corresponding to the sensor node, and determine the first operating energy consumption mode vector of the sensor node based on the first operating energy consumption correlation tensor. The cluster mode determination module is used to construct a second operating energy consumption correlation matrix for the same operating condition cluster based on the corresponding first operating energy consumption mode vector, and determine the second operating energy consumption mode vector for the operating condition cluster based on the second operating energy consumption correlation matrix. The evaluation module is used to determine the evaluation index of the sensor node based on the first operating energy consumption mode vector and the second operating energy consumption mode vector, so as to evaluate the operating status of each sensor node in different clusters. The step of constructing a first operating energy consumption correlation tensor for the sensor node based on the performance time-series data and energy consumption time-series data corresponding to the sensor node includes: Based on the performance time-series data and the energy consumption time-series data, a multi-parameter time-series matrix is ​​constructed, wherein the rows or columns of the multi-parameter time-series matrix are used to represent the performance parameters and energy consumption parameters at different time points; Based on a preset fixed-length sliding time window, the multi-parameter time series matrix is ​​sampled by sliding to obtain multiple time series sub-matrices; For each time series submatrix, high-level semantic features are extracted from the time series submatrix to obtain the corresponding correlation matrix, wherein the correlation matrix is ​​used to represent the inherent correlation pattern between each parameter channel after deep mining; Based on the temporal order of the temporal submatrix, the correlation matrix is ​​stacked to form the first operating energy consumption correlation tensor of the sensor node; The step of determining the first operating energy consumption mode vector of the sensor node based on the first operating energy consumption correlation tensor includes: The first operating energy consumption correlation tensor is decomposed by Tucker to obtain the core tensor, the first factor matrix, the second factor matrix and the third factor matrix, wherein the first factor matrix, the second factor matrix and the third factor matrix correspond to the row dimension, the column dimension and the time dimension, respectively. The element with the largest value is determined from the core tensor, and the corresponding eigenvector in the first factor matrix is ​​determined based on the index of this element in the row dimension of the core tensor. The feature vector is normalized to obtain the first operating energy consumption mode vector of the sensor node; The step of determining the second operating energy consumption mode vector of the operating condition cluster based on the second operating energy consumption correlation matrix includes: Initialize a random, non-zero initial vector v. 0 and for the initial vector v 0 Normalization is performed so that ||v 0 ||2=1; Based on a predefined iterative formula, the second operating energy consumption correlation matrix is ​​iteratively calculated, wherein the calculation formula for the (k+1)th iteration is: C is the second operating energy consumption correlation matrix, v k This is the vector obtained from the k-th iteration; For each iteration, calculate v k+1 With v k If the L2 norm of the difference between the two values ​​is less than a preset convergence threshold, then the iteration stops. The final iteration result v final This serves as the second operating energy consumption mode vector for this operating condition cluster.

7. A computer device, characterized by The method includes a memory and a processor, wherein the memory stores a computer program that can run on the processor, and the processor executes the computer program to implement the method of any one of claims 1 to 5.

Images