Data joint debugging method and system based on heterogeneous monitoring system

Abstract

The invention relates to the technical field of heterogeneous data analysis, and discloses a data joint debugging method and system based on a heterogeneous monitoring system, and the method comprises the steps: obtaining the original data of each data source in the heterogeneous monitoring system; each data source is used as a node in the graph, edges among the nodes represent association relationships among different data sources, and a knowledge graph of the heterogeneous monitoring system is constructed; by analyzing the dependency relationship between the edges of the nodes and adopting a graph time sequence conjoint analysis method, carrying out delay correlation analysis on the feature data of each node, so that the original data of each node is correlated; and triggering a joint debugging mechanism according to the attention weight and the data feature of the node to complete multi-source data collaborative joint debugging of the heterogeneous monitoring system. The time sequence consistency of multi-source monitoring data, the timeliness of abnormal response and the intelligent level of a joint debugging strategy are remarkably improved, and the method has good adaptability and expandability and can be widely applied to complex monitoring scenes such as industrial Internet of Things and smart cities.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of heterogeneous data analysis, in particular to a data joint debugging method and system based on a heterogeneous monitoring system. BACKGROUND

[0002] With the development of Internet of Things, intelligent manufacturing and industrial automation, the scale and complexity of monitoring systems are continuously improving. A large number of data sources from different devices, different manufacturers and different protocols are simultaneously accessed to the monitoring network, forming a typical heterogeneous monitoring environment. There are great differences in data format, sampling frequency, communication protocol and data precision of each monitoring data source, which leads to time sequence offset of collected data, difficulty in aligning state information, and affects data consistency and accuracy of subsequent analysis.

[0003] The prior art usually relies on fixed rules or manually set thresholds to perform single-point monitoring and simple linkage on multi-source data, but in the face of complex dependency relationships between nodes, dynamic scene changes and potential abnormal states, the traditional method cannot realize efficient correction and adaptive collaborative adjustment of multi-source monitoring data, which easily causes data conflicts, abnormal omission and response lag, bringing uncertainty and potential risks to system operation. SUMMARY

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

[0005] Therefore, the technical problem solved by the present application is that the existing data joint debugging method of the heterogeneous monitoring system has problems such as large data format difference, weak time sequence correlation and inability to adaptively trigger efficient joint debugging.

[0006] To solve the above technical problems, the present application provides the following technical scheme: a data joint debugging method based on a heterogeneous monitoring system, comprising:

[0007] In the heterogeneous monitoring system, the original data of each data source is obtained, and the original data collected by each data source is aggregated to form a structured multi-source original data pool;

[0008] Each data source is taken as a node in the graph, and the edges between the nodes represent the association relationship between different data sources, and a knowledge graph of the heterogeneous monitoring system is constructed;

[0009] By analyzing the dependency relationship between the edges of the nodes, a graph time sequence joint analysis method is adopted to perform delay correlation analysis on the feature data of each node, so that the original data of each node is correlated; wherein, in the graph time sequence joint analysis, the feature data of each node is analyzed and reconstructed in time sequence to obtain the feature data of each node;

[0010] Based on the feature data of the reconstructed nodes, the attention weight of each node is calculated; according to the attention weight of the node and the data feature, the joint debugging mechanism is triggered to complete the multi-source data collaborative joint debugging of the heterogeneous monitoring system.

[0011] As a preferred scheme of the data joint debugging method based on the heterogeneous monitoring system, the heterogeneous monitoring system comprises a monitoring system for data acquisition at both homogeneous and heterogeneous data sources; the homogeneous data source is a data source with consistent data attributes, and comprises a monitoring device deployed locally for collecting the same kind of data; the heterogeneous data source is a data source with inconsistent data attributes, and comprises a monitoring device deployed locally for collecting different kinds of data.

[0012] The structured multi-source raw data pool comprises a data set obtained by performing format standardization on the raw data according to field specifications.

[0013] The field specifications include but are not limited to data source type, device ID, timestamp, data content, data source location, and data state identifier.

[0014] As a preferred scheme of the data joint debugging method based on the heterogeneous monitoring system, in the knowledge graph, the correlation between the nodes is established according to the linkage between different nodes during actual operation of each node; and the correlation between the nodes regarding the response state is generated according to the Bayesian network.

[0015] As a preferred scheme of the data joint debugging method based on the heterogeneous monitoring system, the graph time sequence joint analysis comprises setting the delay of the ith node for the current time as , and constructing a delay set T of all nodes ; wherein L represents the number of nodes. The initial value of the delay is the delay calculation result of the last calibration period.

[0016] At the time corresponding to the maximum value of the delay in the delay set T, the fitness between the nodes is evaluated, the improved annealing algorithm is used to iterate the delay of each node, and the delay set T+ with the highest fitness is obtained as the updated delay set.

[0017] The improved annealing algorithm comprises presetting an initial temperature, a maximum iteration number and a temperature decay parameter, and performing iteration; in each iteration process, a random disturbance is added to each element in T with a preset probability triggering the disturbance; if any node i triggers the disturbance with the probability , the delay of the node i is changed by a minimum step size in any direction. Perturb; if any node i has probability If no disturbance is triggered, the delay for node i is retained. ;

[0018] After traveling through all nodes, calculate the global fitness. If the fitness is higher than the value before the iteration, keep the delay set after the current update; if the fitness is not higher than the value before the iteration, keep the delay set before the iteration.

[0019] Assume that the disturbance direction is positive and negative. After each iteration, the iterative process of each node is optimized for the disturbance direction: when the fitness is higher than the value before the iteration, the pheromone is added to the disturbance direction of each node, and the additional value is 1; in the positive and negative directions of the disturbance of each node, the pheromone is accumulated respectively; the disturbance probability in the positive direction is expressed as: ; The perturbation probability in the negative direction is expressed as: ;

[0020] in, Indicates the initial value of the direction disturbance preset, Indicates the cumulative value of pheromone in the positive direction, Indicates the cumulative value of negative direction pheromone; in step length Delay for i When performing disturbance, the disturbance direction is determined by the disturbance probabilities in positive and negative directions;

[0021] The fitness includes the maximum value of the delay in the updated delay set during the current iteration. ;Will The corresponding timestamp is used as the measurement moment of node j; represents the current moment, j represents the node index corresponding to the maximum delay value; for the i-th node, As the measurement time of node i; obtain feature data of all nodes according to their respective measurement times and sliding windows; travel through each node and use the Bayesian network relationship of the edges in the knowledge graph to generate prediction results of the node data sequence: fix the feature data of nodes other than the prediction node and generate prediction results of the data sequence of fixed length in the prediction node;

[0022] For node i, the deviation between the predicted result and the actual data sequence is calculated and recorded as: ; Sum the deviations of each node to get the fitness of this round;

[0023] The association analysis includes associating the original data of each node at the actual occurrence time according to the updated delay set: obtaining the measurement time of each node using the updated delay set, and establishing associations between the measurement times of each node;

[0024] The calibration period is a preset period for updating the delay set.

[0025] As a preferred scheme of the data joint debugging method based on the heterogeneous monitoring system, the timing feature analysis and reconstruction comprises, at a metering time of each node, configuring the sliding window, the head of the window sliding forward from the metering time, and the tail of the window sliding backward from the metering time; analyzing the strong correlation between the data in the current time window and the data at the previous time or the data at the next time through mutual information; if the mutual information with the data at the previous time is greater than a preset value, the head of the current time window is slid forward by a time length; if the mutual information with the data at the next time is greater than a preset value, the tail of the current time window is slid backward by a time length.

[0026] Through the time window, the data sequence in the time window is obtained as the feature data of each node at the metering time.

[0027] As a preferred scheme of the data joint debugging method based on the heterogeneous monitoring system, the calculation process of the attention weight comprises, clustering all nodes according to the edge relationship, regarding the nodes directly or indirectly associated with the edge as a family, and obtaining N families.

[0028] In each family, the feature data of all nodes in the family at the metering time is subjected to principal component analysis: after mean centering, the covariance matrix is calculated and subjected to eigenvalue decomposition, and the eigenvector corresponding to the maximum eigenvalue is extracted as the first principal component direction; the feature data of each node is projected onto the first principal component direction to obtain the projection value of the node in the first principal component direction; for all nodes in the family, the projection value is normalized according to the absolute value as the first attention weight of the node in the family.

[0029] For each node in each family, the minimum Euclidean distance between the feature data at the metering time and the node standard data is calculated; the node standard data corresponding to the minimum Euclidean distance is taken as the denominator, and the minimum Euclidean distance is taken as the numerator to calculate the contribution score of the node; after the contribution scores of all nodes in the family are normalized, the second attention weight of the node in the family is obtained.

[0030] The first attention weight and the second attention weight of each node are added and then divided by two to obtain the attention weight of each node in the family.

[0031] The node standard data comprises a preselected standard data sequence of each node when working normally; according to the features of the node, one node contains at least one node standard data.

[0032] ​As a preferred scheme of the data joint debugging method based on the heterogeneous monitoring system, wherein: the joint debugging mechanism comprises: when the minimum Euclidean distance between the feature data of any node i at the metering moment and the node standard data exceeds the preset value of the node i, it is judged that the node i is abnormal at the metering moment, and the joint debugging is triggered.

[0033] The monitored device corresponding to the abnormal node sends a joint debugging request to other monitored devices corresponding to nodes in the same family; the pre-trained neural network is input: the feature data of all nodes in the family where the abnormal node is located, and the attention weight of each node in the family; and output: the parameter control scheme for the monitored devices corresponding to all nodes in the family;

[0034] After completing a joint debugging, after re-evaluate whether the node is abnormal and perform the next joint debugging; wherein, indicates the maximum value of the delay in the updated delay set T+.

[0035] A data joint debugging system based on a heterogeneous monitoring system using the method described in the application, wherein: the acquisition unit obtains the raw data of each data source in the heterogeneous monitoring system, aggregates the raw data collected by each data source, and forms a structured multi-source raw data pool;

[0036] The association unit constructs a knowledge graph of the heterogeneous monitoring system by taking each data source as a node in the graph and representing the association relationship between different data sources as the edges between nodes.

[0037] The analysis unit performs delay correlation analysis on the feature data of each node by analyzing the dependency relationship between the edges of the nodes using a graph time sequence joint analysis method, and associates the raw data of each node; wherein, during the graph time sequence joint analysis, the feature data of each node is obtained by performing time sequence feature analysis and reconstruction on the node data.

[0038] The adjustment unit calculates the attention weight of each node based on the reconstructed feature data of the node, and triggers the joint debugging mechanism according to the attention weight and data characteristics of the node to complete the multi-source data collaborative joint debugging of the heterogeneous monitoring system.

[0039] A computer device, comprising: a memory and a processor; the memory stores a computer program, wherein: the processor implements the steps of the method of any one of the application when executing the computer program.

[0040] A computer readable storage medium having a computer program stored thereon, wherein: the computer program is executed by a processor to implement the steps of the method of any one of the application.

[0041] The application has the beneficial effects that the data joint debugging method based on the heterogeneous monitoring system provided by the application models the heterogeneous data source as a node and mines the dependency relationship thereof by using a Bayesian network to realize the correlation and dynamic time sequence alignment of multi-source data and overcome the defect that the traditional method cannot accurately correct the time sequence. The improved annealing algorithm combined with the pheromone mechanism is used to effectively optimize the node delay and ensure the consistency and predictability of the data. The node attention weight is calculated by combining local clustering, principal component analysis and minimum distance contribution to improve the identification accuracy of the key driving node. Further, the node anomaly detection and the regulation and control scheme generated by the pre-trained neural network are combined to realize the adaptive collaborative joint debugging of multiple nodes. Compared with the prior art, the time sequence consistency of multi-source monitoring data, the timeliness of abnormal response and the intelligent level of joint debugging strategy are significantly improved, and the method has good adaptability and scalability and can be widely applied to complex monitoring scenes such as industrial Internet of Things and smart city. BRIEF DESCRIPTION OF DRAWINGS

[0042] In order to more clearly illustrate the technical solutions of the embodiments of the application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the application, and for those skilled in the art, other drawings can be obtained without creative labor on the basis of these drawings.

[0043] Figure 1 The application provides a data joint debugging method based on a heterogeneous monitoring system. DETAILED DESCRIPTION

[0044] In order to make the above-mentioned purposes, features and advantages of the application more obvious and easy to understand, the specific embodiments of the application will be described in detail below with reference to the drawings. Obviously, the described embodiments are only a part of the embodiments of the application, not all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor should be within the protection scope of the application.

[0045] Embodiment 1, refer to Figure 1 The application provides a data joint debugging method based on a heterogeneous monitoring system, which comprises the following steps.

[0046] S1: In the heterogeneous monitoring system, the original data of each data source is acquired, and the original data collected by each data source is converged to form a structured multi-source original data pool.

[0047] The heterogeneous monitoring system comprises a monitoring system for data collection at both homogeneous and heterogeneous data sources. The homogeneous data source refers to a data source with consistent data attributes, including a monitoring device deployed locally for collecting the same type of data. The heterogeneous data source refers to a data source with inconsistent data attributes, including a monitoring device deployed locally for collecting different types of data.

[0048] Specifically, in the present embodiment, the homogeneous data source refers to a data source from a monitoring device or sensor with consistent attributes, the same structure or unified collection logic, and usually collects the same type of physical quantity (such as temperature, humidity, pressure, flow, etc.), with consistent data format, sampling period and data structure, facilitating direct comparison and analysis under the same standard. For example, multiple temperature sensors of the same type deployed in the same factory workshop constitute a typical homogeneous data source. The heterogeneous data source refers to a data source from a monitoring device or sensor with different attributes, inconsistent structure, different collection content or sampling method. Such data sources usually collect different types of physical quantities (such as temperature and vibration, image and current, etc.), or use different protocols, sampling frequencies and data formats, making it difficult to directly align and compare the data. For example, the data collected by a pressure sensor, a vibration accelerometer and an industrial camera deployed simultaneously on the same production line constitute a typical heterogeneous data source. Therefore, the homogeneous data source is convenient for direct horizontal comparison and consistency verification, while the heterogeneous data source needs to be processed through feature extraction, time alignment and dependency modeling to realize effective cross-source data fusion and joint analysis.

[0049] The structured multi-source raw data pool comprises a data set of the raw data after format standardization according to field specifications. The field specifications include but are not limited to data source type, device ID, timestamp, data content, data source location and data status identifier.

[0050] It is to be known that in a heterogeneous monitoring system, different data sources often have differences in data format, naming rules, timestamp accuracy, source description, etc., which directly lead to the increase in complexity of subsequent time alignment and dependency modeling. To realize efficient aggregation, unified management and cross-source analysis of multi-source monitoring data, field specifications including data source type, device ID, timestamp, data content, data source location, data state identifier, etc. are set as important basis for standardization of raw data. By preprocessing and format conversion of raw data from different sources according to the field specifications, the scattered, heterogeneous and lack of unified structure data can be quickly converted into a structured data set that can be uniformly indexed, aligned, searched and analyzed. The structured data pool provides standardized, unambiguous and easily expandable underlying data support for subsequent knowledge graph construction, graph-time sequence joint analysis and attention weight calculation, significantly improving the usability of data and the real-time performance of cross-source debugging, and ensuring that the multi-source monitoring system can still run stably and accurately in large-scale and complex scenarios.

[0051] S2: Each data source is taken as a node in the graph, and the edges between the nodes represent the association relationship between different data sources, and a knowledge graph of the heterogeneous monitoring system is constructed.

[0052] Further, in the knowledge graph, during the actual operation of each node, the association relationship is established according to the linkage between different nodes; and according to the Bayesian network, the association relationship between the nodes about the response state is generated.

[0053] In a multi-source heterogeneous monitoring system, there are often coupling relationships between different data sources at the physical layer, logical layer and business process layer, such as causal or linkage effects between temperature sensors, vibration sensors and motor operating states. However, these relationships are usually represented by artificial rules or isolated thresholds in traditional monitoring data management, which are difficult to dynamically update, automatically reason or adaptively identify. Therefore, the present application abstracts each data source as a node in the knowledge graph, and represents the dependency or influence path between different nodes by edges, so that the system can explicitly represent and store the internal relationship between multi-source monitoring data in the form of a graph structure.

[0054] Further, to enhance the reasonability and dynamics of the relationship, the Bayesian network is used to learn and generate the response state association relationship between nodes based on actual operation data. The Bayesian network can automatically derive conditional probability and causal relationship based on observation data, thereby reflecting the mutual influence and linkage degree of each node in different states. Through the combination of this structured graph and probability inference, the dependency relationship between nodes can be updated in real time during operation, supporting time correction, anomaly detection and linkage decision, ensuring the intelligentization and adaptivity of multi-source data collaborative processing and abnormal response.

[0055] S3: By analyzing the dependency relationship between the edges of the nodes, the graph-time joint analysis method is used to perform delayed correlation analysis on the characteristic data of each node, so as to associate the original data of each node. In the graph-time joint analysis, the characteristic analysis and reconstruction of the time series of each node data are performed to obtain the characteristic data of each node.

[0056] Specifically, the graph timing joint analysis: Let the delay of the i-th node at the current moment be , build the delay set of all nodes ; Where L represents the number of nodes; The initial value of is the delay calculation result of the last calibration cycle. In the delay set T, At the moment corresponding to the maximum value of , the fitness of each node is evaluated, and the improved annealing algorithm is used to iterate the delay of each node to obtain the delay set T+ with the highest fitness, which is used as the updated delay set.

[0057] In heterogeneous monitoring systems, different data sources often have inherent time deviations due to sampling cycles, data upload delays, or network transmission. If the timing of each node data is not dynamically corrected, the reflection time of the same event in different data sources may be inconsistent, thereby affecting the accuracy of association analysis, anomaly identification, and joint debugging strategies. The present invention sets the delay parameters of each node at the current moment and constructs a delay set for all nodes, using the delay results of the previous calibration cycle as the initial reference to ensure the continuity and historical consistency of timing analysis.

[0058] To further dynamically optimize globally, an improved annealing algorithm was introduced to iteratively optimize node delays. A fitness evaluation function was used to measure the impact of each node delay on overall prediction accuracy in real time, gradually converging to an optimal delay set that maximizes the temporal consistency of multi-node data. This approach not only avoids errors caused by manually configured time deviations but also adaptively adjusts the temporal alignment between nodes by integrating dependencies within the knowledge graph. This significantly improves the alignment accuracy of multi-source data and the reliability of subsequent feature extraction, anomaly detection, and joint debugging.

[0059] The improved annealing algorithm includes: presetting the initial temperature, the maximum number of iterations and the temperature decay parameter, and performing iterations; adding random perturbations to each element in T during each iteration, with a preset probability Trigger disturbance; if any node i has probability If the disturbance is triggered, the minimum step size in any direction is Delay to node i Perturb; if any node i has probability If no disturbance is triggered, the delay for node i is retained. .

[0060] After visiting all nodes, the global fitness is calculated, if the fitness is higher than the value before iteration, the delay set after this round of update is reserved, if the fitness is not higher than the value before iteration, the delay set before iteration is reserved.

[0061] The disturbance direction is positive and negative, after each iteration, the disturbance direction of each node is optimized: when the fitness is higher than the value before iteration, the pheromone of each node is added, and the added value is 1; the pheromone is accumulated in the positive and negative directions of the disturbance of each node; the disturbance probability of the positive direction is represented as: ; the disturbance probability of the negative direction is represented as: .

[0062] Wherein, represents the initial value of the disturbance direction, represents the cumulative value of the positive direction pheromone, represents the cumulative value of the negative direction pheromone; in the step , the delay of i is disturbed, and the disturbance direction is determined according to the disturbance probability of the positive and negative directions.

[0063] It is to be said that in the time sequence alignment of multi-source heterogeneous monitoring data, the adjustment of the delay parameter of each node may have problems such as slow convergence speed, falling into local optimum, unstable disturbance direction, etc., which affects the accuracy and stability of global delay optimization. Therefore, the present application improves the classical annealing algorithm, introduces a random disturbance and probability triggering mechanism for each node, ensures the sufficiency and diversity of the search space, and enhances the ability of the delay parameter to jump out of the local optimal solution in the global range.

[0064] At the same time, the disturbance direction is dynamically optimized combined with the pheromone mechanism: after each iteration, if the global fitness is improved, the pheromone reward is added to the current disturbance direction and the historical disturbance effect is accumulated, so that the selection probability of the positive and negative directions is adaptively adjusted, and the more effective disturbance direction is preferentially reserved. This pheromone-driven disturbance direction adjustment not only speeds up the convergence speed, but also improves the global search ability and local fine-tuning flexibility of the delay optimization process, ensures that the delay set obtained finally maximizes the multi-node time sequence consistency, and provides a stable and reliable time sequence basis for subsequent feature extraction, anomaly identification and joint debugging execution.

[0065] The fitness includes, in this round of iteration, the maximum value of the delay in the updated delay set ; the corresponding time stamp of is taken as the measurement time of node j; represents the current time, j represents the node index corresponding to the maximum delay value; for the i-th node, the measurement time of the node i is denoted as the measurement time of each node; the feature data of all nodes are obtained according to the respective measurement time and the sliding window; each node is traversed, and the prediction result of the data sequence of the node is generated by using the Bayesian network relationship of the edges in the knowledge graph; the feature data of the nodes except the prediction node is fixed (in fact, for the Bayesian network, this is equal to the input), and the prediction result of the fixed-length data sequence in the prediction node is generated. In the heterogeneous monitoring system, in order to ensure the global consistency and prediction accuracy of the data of each node under different delay correction, the delay optimization and the inter-node dependency reasoning are combined by constructing an adaptive function. Specifically, after each iteration, the time corresponding to the maximum delay in the updated delay set is selected as the reference measurement time, so as to unify the analysis benchmark of different nodes. Each node obtains the feature data of the same length by using the sliding window according to the respective corrected measurement time, so as to ensure the time sequence consistency and comparability of the input.

[0066] For the i-th node, the deviation (such as mean square error (MSE), mean absolute error (MAE), etc.) between the prediction result and the actual data sequence is calculated, denoted as: The deviations of each node are summed to obtain the adaptive degree of this round. The adaptive value obtained by summing the residuals of each node can truly reflect the improvement degree of the state explanation ability of the nodes by the delay setting of this round, and is used as the basis for iteration direction and good-bad discrimination of the annealing algorithm. This method significantly improves the adaptive coupling effect of delay tuning and state prediction, and ensures the high robustness of multi-source data alignment and the interpretability of the prediction result.

[0067] Further, the correlation analysis includes correlating the original data of each node at the actual occurrence time according to the updated delay set; the measurement time of each node is obtained by using the updated delay set, and the correlation between the measurement times of each node is established.

[0068] It is also known that the correction period is a preset period for updating the delay set .

[0069] In the embodiment, the feature analysis and reconstruction of the time sequence includes, at the metering time of each node, configuring the sliding window, the head of the window sliding forward from the metering time, and the tail of the window sliding backward from the metering time (at the tail of the sliding window, the constraint is that the position of the tail is less than the current time. However, due to the calculation of mutual information, this constraint is covered, and therefore no additional constraint is applied); analyzing the strong correlation of the data in the current time window with the data at the previous time or the data at the next time through mutual information; if the mutual information with the data at the previous time is greater than a preset value, the head of the current time window is slid forward by a time length; if the mutual information with the data at the next time is greater than a preset value, the tail of the current time window is slid backward by a time length.

[0070] Through the time window, the data sequence in the time window is obtained as the feature data of each node at the metering time. By introducing the adaptive window adjustment based on mutual information, the extracted time sequence data not only retains the main information of the node state, but also can flexibly expand or shrink the time range according to the correlation degree before and after, which improves the perception ability of sudden abnormalities, short-term fluctuations and gradual trends. The node features are guaranteed to be representative and time sequence complete, providing more real and unbiased input basis for subsequent anomaly detection, state reasoning and joint debugging decision, fundamentally enhancing the dynamic adaptability and robustness of time sequence features, and avoiding information loss and local distortion caused by artificial setting of fixed time window.

[0071] S4: based on the feature data of the reconstructed nodes, calculating the attention weight of each node; according to the attention weight of the node and the data feature, triggering the joint debugging mechanism to complete the multi-source data collaborative joint debugging of the heterogeneous monitoring system.

[0072] The calculation process of the attention weight: all nodes are clustered according to the relationship of the edges, and nodes with direct or indirect edge association are regarded as a family, obtaining N families. In the heterogeneous monitoring system, the correlation between different nodes often has the characteristics of localization, hierarchical and non-homogeneity: nodes with direct or indirect edge relationship in the same group usually correspond to physically interacting or business logic tightly coupled subsystems, such as the same production line, the same device cluster or the same process section; while the physical quantities collected by different families are quite different, the data distribution is obviously different, and the change mode is also inconsistent.

[0073] If the analysis is directly performed on all nodes mixed in the global range, the principal component analysis is easily dominated by strong signals with high variance, resulting in that the weak signal nodes or local mode information are submerged, and the key change factors under each local dependent structure cannot be effectively captured; at the same time, the normal deviation baselines of different nodes are not consistent, and a single global model will reduce the sensitivity and discrimination accuracy of anomaly detection.

[0074] Therefore, nodes with direct or indirect edge correlations will be clustered into several groups, ensuring that nodes within a group share similar dependency and coupling patterns. Independent principal component analysis is performed within each group to accurately extract the main variation direction in the local area, avoiding global interference; at the same time, the deviation of nodes from the normal state of the group is calculated within the group, which is more in line with the operation characteristics of the local subsystem. By analyzing and generating weights independently for each group, the weight distribution is both locally adaptive and reasonably superimposed from a global perspective, significantly improving the accuracy and interpretability of overall integration and testing.

[0075] Within each group, principal component analysis is performed on the feature data of all nodes at the moment: after mean centering, the covariance matrix is calculated and eigenvalue decomposition is performed, and the eigenvector corresponding to the largest eigenvalue is extracted as the first principal component direction; project the feature data of each node onto the first principal component direction to obtain the projection value of the node in the first principal component direction; for all nodes in the group, normalize the absolute value of the projection value as the first attention weight of the node in the group.

[0076] Specifically: let the group have nodes, each node has features, denoted by . denotes the th node in the group. denotes the th feature dimension of the node.

[0077] Take the mean of the th feature of all nodes in the group:

[0078]

[0079] where is the original feature value, is the mean value of the dimension in the group.

[0080] Mean center the original data:

[0081]

[0082] where is the centered value.

[0083] Construct the covariance matrix after centering:

[0084]

[0085] where is the index of the feature dimension. is the covariance matrix element.

[0086] Solve the covariance matrix:

[0087] The eigenvalue decomposition of the first intra-group covariance matrix is as follows:

[0088]

[0089] wherein, λi represents the first eigenvalue; vi represents the eigenvector corresponding to .

[0090] Take the eigenvector corresponding to the maximum eigenvalue :

[0091]

[0092] wherein, λmax represents the maximum eigenvalue. v1 represents the first principal component vector.

[0093] The projection value of each node in the cluster in the principal component direction is , and the projection value of each node in the cluster is:

[0094]

[0095] wherein, vi represents the first component of the eigenvector . vi,p represents the projection value of the first node in the first principal component direction in the first cluster.

[0096] After taking the absolute value of the projection value and normalizing, the first attention weight of the first node in the kth cluster is obtained:

[0097]

[0098] wherein, s represents the traversal index when accumulating normalization.

[0099] For each node in each cluster, the minimum Euclidean distance between the feature data at the measurement time and the node standard data (which can be multiple, and the minimum value is selected after calculating the Euclidean distance) is calculated; the node standard data corresponding to the minimum Euclidean distance is taken as the denominator, and the minimum Euclidean distance is taken as the numerator to calculate the contribution score of the node; after normalizing the contribution scores of all nodes in the cluster, the second attention weight of the node in the cluster is obtained.

[0100] The first attention weight and the second attention weight of each node are added and then divided by two to obtain the attention weight of each node within the family.

[0101] The node standard data includes a preselected standard data sequence of each node in a normal working state; according to the characteristics of the node, a node contains at least one node standard data. Specifically, a node in a normal state does not have only one average value, but can have multiple typical "standard values", which can be "representative comparison values" of the node in a typical normal mode, so as to subsequently compare one by one in the minimum distance to find the closest "standard state". For example, a certain temperature node can record its typical temperature values under low load, rated load and high load as multiple standard points; a certain vibration node can establish corresponding standard values according to the typical vibration amplitudes under different speeds. During operation, when the node characteristic data needs to be matched by the minimum Euclidean distance, the system will calculate the Euclidean distance between the current data of the node and all its standard comparison values one by one, select the standard value corresponding to the minimum distance as the "optimal comparison reference" at the current time, and use it for contribution scoring and abnormality judgment.

[0102] It is also worth mentioning that the joint debugging mechanism includes that when the minimum Euclidean distance between the characteristic data of any node i at the metering time and the node standard data exceeds the preset value of the node i, it is judged that the node i is abnormal at the metering time, triggering joint debugging.

[0103] The abnormal node corresponds to a monitored device, which sends a joint debugging request to other monitored devices corresponding to nodes in the same family; through a pre-trained neural network, input: the characteristic data of all nodes in the family to which the abnormal node belongs, and the attention weight of each node in the family; output: a parameter control scheme for the monitored devices corresponding to all nodes in the family.

[0104] After completing a joint debugging, the system re-evaluates whether the nodes are abnormal and performs the next joint debugging. After completing a joint debugging, the system re-evaluates whether the nodes are abnormal and performs the next joint debugging. The maximum value of the updated delay set T+ is represented. It is ensured that the parameters adjusted in the last round of joint debugging are rechecked after the actual physical delays of all nodes are fully transmitted and take effect, to avoid misjudgment of new abnormalities due to incomplete effectiveness, so as to ensure that the next round of joint debugging is based on the real state that has fully played its role, and to improve the accuracy of adjustment and closed-loop stability.

[0105] After determining the single-node anomaly, to avoid information inconsistency or multi-source state imbalance caused by isolated treatment, the design automatically sends a joint debugging request to other associated nodes in the same family through the intra-family edge relationship of the knowledge graph. A pre-trained neural network is used to dynamically output the optimal parameter adjustment scheme for multiple nodes by combining the current feature data and attention weights of all nodes, ensuring that the adjustment process considers both local state and overall operation goals. This mechanism establishes a closed-loop process of "automatic anomaly detection - intra-family joint adjustment - neural network assisted reasoning - post-joint evaluation", with online learning and adaptive capabilities, significantly improving the response speed, adjustment effect, and system stability of complex heterogeneous monitoring systems in abnormal situations.

[0106] In this embodiment, the neural network refers to a multi-layer feedforward neural network constructed based on a deep learning framework such as TensorFlow or PyTorch. Its structure includes an input layer, at least one hidden layer, and an output layer, which is used to automatically output the optimal parameter adjustment scheme for all monitored devices in the family based on the input of the current feature data and corresponding attention weights of each node in the family, through nonlinear mapping and parameter training. The neural network can be pre-trained offline using historical monitoring data, with the training goal being to minimize the deviation and energy consumption of the adjustment effect loss function, and the weights can be updated regularly based on actual operation data to improve the accuracy of the adjustment.

[0107] In an optional embodiment, the neural network can include but is not limited to the following alternative structures: Recurrent Neural Network (RNN) or Long Short-Term Memory Network (LSTM): suitable for capturing the time dependence of node feature data, enhancing the learning ability of time series dynamic characteristics. Convolutional Neural Network (CNN): can be used to process local spatial relationship patterns between multiple nodes, suitable for scenarios with high feature dimension or fixed node topology. Graph Neural Network (GNN): when there are complex edge relationships between nodes in the family, a graph-based neural network can be used to directly utilize node and edge information in the knowledge graph for message passing and aggregation, improving the structural adaptability of the adjustment scheme. Ensemble learning network: can be based on multiple neural networks or combined with decision trees, support vector machines, and other machine learning models to form a hybrid model, further optimizing the robustness and generalization performance of the adjustment scheme.

[0108] Embodiment 2 also provides a data joint debugging system based on a heterogeneous monitoring system, which includes:

[0109] The acquisition unit acquires the original data of each data source in the heterogeneous monitoring system, aggregates the original data collected by each data source, and forms a structured multi-source original data pool.

[0110] The association unit takes each data source as a node in a graph, and an edge between nodes represents an association relationship between different data sources, and constructs a knowledge graph of the heterogeneous monitoring system.

[0111] The analysis unit performs time-delayed correlation analysis on the feature data of each node by analyzing the dependency relationship of the edges between nodes using a graph timing joint analysis method, and correlates the original data of each node; wherein, during the graph timing joint analysis, the feature analysis and reconstruction of the timing of each node data are performed to obtain the feature data of each node.

[0112] The adjustment unit calculates the attention weight of each node based on the feature data of the reconstructed node; and triggers the joint debugging mechanism according to the attention weight of the node and the data feature, and completes the multi-source data collaborative joint debugging of the heterogeneous monitoring system.

[0113] If the above functions are realized in the form of software function units and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application or the part of the prior art that essentially contributes or the part of the technical solutions can be embodied in the form of a software product, which is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device) to execute all or part of the steps of the method described in the embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk and various program code storage media.

[0114] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered a list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus, such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute them, or in conjunction with these instruction execution systems, devices or apparatus. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate or transport programs for use by an instruction execution system, device or apparatus, or in conjunction with these instruction execution systems, devices or apparatus.

[0115] More specific examples (a non-exhaustive list) of the computer-readable medium include the following: an electrical connection (electronic) having one or more wires, a portable computer diskette (magnetic), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). Additionally, the computer readable medium can be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, for example via an optical scanner, then compiled, interpreted or otherwise processed in a suitable manner, if necessary, and then stored in a computer memory in a suitable manner.

[0116] It should be understood that aspects of the application can be implemented in hardware, software, firmware or combinations thereof. In the embodiments described above, various steps or methods can be implemented, for example in software or firmware, stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, and in another embodiment, any of the following technologies, known in the art, or a combination thereof, can be used: discrete logic circuitry having logic gates for implementing logic functions upon an application of data signals, application specific integrated circuits having appropriate combinational logic gates, programmable gate arrays (PGA), field programmable gate arrays (FPGA), and the like.

[0117] It should be noted that the above examples are merely intended to illustrate the technical solutions of the present application and not to limit the same. Although the present application has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the present application, and all such modifications or replacements should be included in the scope of the claims of the present application.

Claims

1. A data joint debugging method based on heterogeneous monitoring systems, characterized in that: include: In a heterogeneous monitoring system, the raw data from each data source is obtained and aggregated to form a structured multi-source raw data pool. Each data source is treated as a node in the graph, and the edges between nodes represent the association between different data sources, thus constructing a knowledge graph for heterogeneous monitoring systems. By analyzing the dependency relationship between the edges of the nodes, the method of graph-time joint analysis is used to perform delayed correlation analysis on the characteristic data of each node, so as to associate the original data of each node. In the graph-time joint analysis, the characteristic analysis and reconstruction of the time series of each node data are performed to obtain the characteristic data of each node. Based on the feature data after node reconstruction, the attention weight of each node is calculated; according to the attention weight and data characteristics of the node, the joint debugging mechanism is triggered to complete the multi-source data collaborative debugging of the heterogeneous monitoring system.

2. The data joint debugging method based on heterogeneous monitoring systems according to claim 1, characterized in that: The heterogeneous monitoring system includes a monitoring system that collects data from both homogeneous and heterogeneous data sources; wherein a homogeneous data source is a data source with consistent data attributes, including monitoring devices deployed locally for collecting the same type of data; a heterogeneous data source is a data source with inconsistent data attributes, including monitoring devices deployed locally for collecting different types of data; The structured multi-source raw data pool includes a data set obtained by formatting the raw data in accordance with field specifications; The field specifications include, but are not limited to, data source type, device ID, timestamp, data content, data source location, and data status identifier.

3. The data joint debugging method based on heterogeneous monitoring systems according to claim 2, characterized in that: In the knowledge graph, during the actual operation of each node, an association relationship is established based on the linkage between different nodes; based on the Bayesian network, an association relationship between nodes regarding the response status is generated.

4. The data joint debugging method based on heterogeneous monitoring systems according to claim 3 is characterized in that: The graph timing joint analysis includes: assuming that the delay of the i-th node at the current moment is , build the delay set of all nodes ; Where L represents the number of nodes; The initial value of is the delay calculation result of the last calibration cycle; In the delay set T, At the moment corresponding to the maximum value of , the fitness of each node is evaluated, and the improved annealing algorithm is used to iterate the delay of each node to obtain the delay set T+ with the highest fitness, which is used as the updated delay set; The improved annealing algorithm includes presetting the initial temperature, the maximum number of iterations and the temperature decay parameter, and performing iterations; For each element in T, random perturbation is added in each iteration with a preset probability Trigger disturbance; if any node i has probability If the disturbance is triggered, the minimum step size in any direction is Delay to node i Perturb; if any node i has probability If no disturbance is triggered, the delay for node i is retained. ; After traveling through all nodes, calculate the global fitness. If the fitness is higher than the value before the iteration, keep the delay set after the current update; if the fitness is not higher than the value before the iteration, keep the delay set before the iteration. Assume that the disturbance direction is positive and negative. After each iteration, the iterative process of each node is optimized for the disturbance direction: when the fitness is higher than the value before the iteration, the pheromone is added to the disturbance direction of each node, and the additional value is 1; in the positive and negative directions of the disturbance of each node, the pheromone is accumulated respectively; the disturbance probability in the positive direction is expressed as: ; The perturbation probability in the negative direction is expressed as: ; in, Indicates the initial value of the direction disturbance preset, Indicates the cumulative value of pheromone in the positive direction, Indicates the cumulative value of negative direction pheromone; in step length Delay for i When performing disturbance, the disturbance direction is determined by the disturbance probabilities in positive and negative directions; The fitness includes the maximum value of the delay in the updated delay set during the current iteration. ;Will The corresponding timestamp is used as the measurement moment of node j; represents the current moment, j represents the node index corresponding to the maximum delay value; for the i-th node, As the measurement time of node i; obtain feature data of all nodes according to their respective measurement times and sliding windows; travel through each node and use the Bayesian network relationship of the edges in the knowledge graph to generate prediction results of the node data sequence: fix the feature data of nodes other than the prediction node and generate prediction results of the data sequence of fixed length in the prediction node; For node i, the deviation between the predicted result and the actual data sequence is calculated and recorded as: ; Sum the deviations of each node to get the fitness of this round; The association analysis includes associating the original data of each node at the actual occurrence time according to the updated delay set: obtaining the measurement time of each node using the updated delay set, and establishing associations between the measurement times of each node; The calibration period is a delay set The preset period for updates.

5. The data joint debugging method based on heterogeneous monitoring systems according to claim 4 is characterized in that: The characteristic analysis and reconstruction of the time series includes configuring the sliding window at the metering time of each node, with the head of the window sliding forward from the metering time and the tail of the window sliding backward from the metering time; analyzing the strong correlation between the data in the current time window and the data at the previous time or the data at the next time through mutual information; if the mutual information with the data at the previous time is greater than a preset value, sliding the head of the current time window forward by a time length; if the mutual information with the data at the next time is greater than a preset value, sliding the tail of the current time window backward by a time length; Through the time window, the data sequence within the time window is obtained as the characteristic data of each node at the measurement moment.

6. The data joint debugging method based on heterogeneous monitoring systems according to claim 5, characterized in that: The attention weight calculation process includes clustering all nodes according to edge relationships, treating nodes that are directly or indirectly edge-related as a cluster, and obtaining N clusters; Within each cluster, principal component analysis is performed on the characteristic data of all nodes in the cluster at the measurement moment: after mean centering, the covariance matrix is ​​calculated and eigenvalue decomposition is performed, and the eigenvector corresponding to the maximum eigenvalue is extracted as the first principal component direction; the characteristic data of each node is projected onto the first principal component direction to obtain the projection value of the node on the first principal component direction; for all nodes in the cluster, the absolute value of the projection value is normalized and used as the first attention weight of the node in the cluster; For each node in each cluster, the minimum Euclidean distance between the feature data and the node standard data at the measurement moment is calculated. The node standard data corresponding to the minimum Euclidean distance is used as the denominator, and the minimum Euclidean distance is used as the numerator to calculate the node contribution score. After normalizing the contribution scores of all nodes in the cluster, the second attention weight of the node in the cluster is obtained. Add the first attention weight and the second attention weight of each node and divide it by two to get the attention weight of each node in the family; The node standard data includes a standard data sequence preselected when each node is operating normally; according to the characteristics of the node, a node contains at least one of the node standard data.

7. The data joint debugging method based on heterogeneous monitoring systems according to claim 6 is characterized in that: The joint debugging mechanism includes: when the minimum Euclidean distance between the characteristic data of any node i at the metering time and the node standard data exceeds the preset value for node i, it is determined that node i has an abnormality at the metering time and the joint debugging is triggered; The monitored device corresponding to the abnormal node sends a joint debugging request to other monitored devices corresponding to nodes in the same cluster. The pre-trained neural network inputs: the feature data of all nodes in the cluster where the abnormal node is located, as well as the attention weight of each node in the cluster. Output: parameter control scheme for all nodes in the cluster corresponding to the monitored devices; After completing a joint commissioning, After that, re-evaluate whether the node is abnormal and perform the next joint debugging; Indicates the maximum delay value in the updated delay set T+.

8. A data joint debugging system based on heterogeneous monitoring systems using the method according to any one of claims 1 to 7, characterized in that: The acquisition unit obtains the raw data from each data source in the heterogeneous monitoring system, aggregates the raw data collected from each data source, and forms a structured multi-source raw data pool; The association unit treats each data source as a node in the graph, and the edges between nodes represent the association relationship between different data sources, thus constructing a knowledge graph for heterogeneous monitoring systems. The analysis unit analyzes the dependency relationship between the edges of the nodes and uses the graph-time joint analysis method to perform delayed correlation analysis on the feature data of each node, so as to associate the original data of each node. In the graph-time joint analysis, the feature analysis and reconstruction of the time series of each node data are performed to obtain the feature data of each node. The adjustment unit calculates the attention weight of each node based on the reconstructed feature data of the node; according to the attention weight and data characteristics of the node, the joint debugging mechanism is triggered to complete the multi-source data collaborative debugging of the heterogeneous monitoring system.

9. A computer device comprising: A memory and a processor; the memory stores a computer program, wherein the processor implements the steps of the method according to any one of claims 1 to 7 when executing the computer program.

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

Images