A monitoring method for single-phase smart electric energy meter automatic verification assembly line

Abstract

The application discloses a kind of monitoring methods for single-phase intelligent electric energy meter automatic verification assembly line, it is related to electric energy measurement automation detection field, including, the operation data set of each station of single-phase intelligent electric energy meter automatic verification assembly line is collected, environmental operation data set and electric energy meter image data, form initial multi-source data set with space-time mark;Initial multi-source data set with space-time mark is carried out data cleaning and feature extraction processing, generate pure neat multi-dimensional feature vector set;Pure neat multi-dimensional feature vector set is input into space-time graph neural network model and is carried out abnormal propagation mode identification output abnormal propagation probability graph;Based on abnormal propagation probability graph, trigger verification assembly line knowledge graph to carry out cross-modal root cause reasoning, generate explainable root cause hypothesis set.The application verifies hypothesis and generates optimization instruction by digital twin, forms monitoring-diagnosis-optimization-self-learning control, effectively solves the problem of cross work of traditional method.

Description

Technical Field

[0001] This invention relates to the field of automated testing of electricity metering, and in particular to a monitoring method for automated calibration lines of single-phase smart meters. Background Technology

[0002] With the continuous development of smart grid construction, single-phase smart meters, as key terminals for electricity metering, directly affect fair trade and grid security in terms of metering accuracy. To meet the needs of large-scale verification, automated verification lines have become the core equipment for electricity meter verification. Existing technologies generally adopt a monitoring architecture based on programmable logic controllers and distributed data acquisition systems. Through sensor networks deployed at each workstation, operating parameters such as voltage and current are collected in real time, and status monitoring and alarms are performed based on preset thresholds.

[0003] Existing monitoring methods have limitations in dealing with latent anomalies caused by multi-factor coupling and cross-workstation propagation. Because there are close connections between units in the production line, small deviations at a single workstation may trigger cascading effects as materials flow, ultimately leading to systematic verification deviations. Existing technologies focus on independent monitoring of single-point data and lack the ability to model the propagation path of anomalies in the spatiotemporal dimension. This makes it difficult to identify the propagation trend and locate the root cause in the early stages of a fault, resulting in delayed operation and maintenance intervention. Summary of the Invention

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

[0005] Therefore, this invention provides a monitoring method for automatic verification pipelines of single-phase smart energy meters to solve the problem of delayed operation and maintenance intervention caused by the lack of modeling ability of anomalies propagation paths in the spatiotemporal dimension in existing technologies.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a monitoring method for an automatic calibration production line of single-phase smart meters, which includes collecting operation datasets, environmental operation datasets and meter image data of each station of the automatic calibration production line of single-phase smart meters to form an initial multi-source data set with spatiotemporal markers. Data cleaning and feature extraction are performed on the initial multi-source dataset with spatiotemporal labels to generate a clean and well-organized set of multidimensional feature vectors. A clean and well-ordered set of multidimensional feature vectors is input into a spatiotemporal graph neural network model to perform anomaly propagation pattern recognition and output an anomaly propagation probability map. Based on the anomaly propagation probability graph, the knowledge graph of the triggering verification pipeline is used to perform cross-modal root cause reasoning and generate an interpretable set of root cause hypotheses. In the digital twin of the automatic calibration pipeline for single-phase smart energy meters, the set of interpretable root cause hypotheses is verified and the corresponding set of optimized control instructions is generated. The system executes an optimized control instruction set to adjust the operating parameters of the automatic calibration pipeline for single-phase smart energy meters, and collects a set of feedback data after the strategy execution. The system then uses this set of feedback data to incrementally learn and update the spatiotemporal neural network model and the calibration pipeline knowledge graph.

[0007] As a preferred embodiment of the monitoring method for the automatic calibration production line of single-phase smart meters described in this invention, the method involves: collecting operational data sets, environmental operational data sets, and meter image data from each station of the automatic calibration production line to form an initial multi-source data set with spatiotemporal markers, including the following steps: Collect operational data sets, environmental operation data sets, and meter image data from each workstation of the automatic calibration production line for single-phase smart meters; Add timestamps and workstation space identifiers to the runtime dataset, environmental runtime dataset, and electricity meter image data; The runtime dataset with timestamps and workstation space identifiers, the environmental runtime dataset, and the electricity meter image data are aggregated to form an initial multi-source dataset with spatiotemporal tags.

[0008] As a preferred embodiment of the monitoring method for the automatic verification pipeline of single-phase smart energy meters described in this invention, the method includes the following steps: Data cleaning and feature extraction processing are performed on an initial multi-source data set with spatiotemporal markers to generate a clean and well-organized multi-dimensional feature vector set: Data alignment and outlier removal are performed on the initial multi-source dataset with spatiotemporal tags to generate a time-aligned dataset; Z-score normalization is performed on the runtime dataset and environment runtime dataset in the dataset with time-series alignment anomalies to generate a normalized runtime environment dataset; The energy meter image data in the time-aligned dataset is processed by grayscale conversion, Gaussian filtering for noise reduction, and Canny edge detection to generate a visual feature matrix. The standardized operating environment dataset and the visual feature matrix are spliced ​​together and subjected to principal component analysis for dimensionality reduction, generating a clean and well-organized set of multidimensional feature vectors.

[0009] As a preferred embodiment of the monitoring method for the automatic verification production line of single-phase smart energy meters described in this invention, the method includes the following steps: inputting a clean and regular set of multidimensional feature vectors into a spatiotemporal graph neural network model to perform anomaly propagation pattern recognition and output anomaly propagation probability map. A set of clean and well-ordered multidimensional feature vectors is mapped onto a graph structure with assembly line workstations as nodes and inter-workstation connections as edges to generate spatiotemporal graph structure data. The spatiotemporal graph structure data is input into the graph convolutional layer of the spatiotemporal graph neural network model to extract the spatial dependency features between nodes and generate spatial feature representations. Spatial feature representations are input into the gated recurrent unit layer of the spatiotemporal graph neural network model to learn dynamic change patterns on the time series and generate spatiotemporal fusion features; Based on the spatiotemporal fusion characteristics, the probability of abnormal states of each node and edge in the graph is calculated to generate an anomaly propagation probability graph.

[0010] As a preferred embodiment of the monitoring method for the automatic verification pipeline of single-phase smart energy meters described in this invention, the method includes the following steps: based on the anomaly propagation probability graph, triggering cross-modal root cause reasoning using the knowledge graph of the verification pipeline to generate an interpretable set of root cause hypotheses. Based on the anomaly propagation probability map, paths with a probability exceeding the probability distribution of successful early warnings in the historical anomaly propagation probability map are identified, and anomaly propagation path thresholds are set by comprehensively balancing the false alarm rate and the false negative rate, thereby generating a set of high-risk anomaly paths. The set of high-risk abnormal paths is matched and mapped with entities and relationships in the knowledge graph of the verification pipeline to generate a set of candidate failure modes. Calculate the confidence level of the candidate failure mode set based on graph relationships to generate a set of interpretable root cause hypotheses with confidence levels.

[0011] As a preferred embodiment of the monitoring method for the automatic calibration pipeline of single-phase smart energy meters described in this invention, the method includes the following steps: verifying an interpretable set of root cause hypotheses in the digital twin of the automatic calibration pipeline of single-phase smart energy meters and generating a corresponding set of optimized control instructions. In the digital twin of the automatic calibration pipeline for single-phase smart energy meters, the performance parameters of the corresponding virtual components are adjusted according to the root cause assumptions in the set of interpretable root cause assumptions to generate a digital twin with adjusted parameters. Run the simulated verification process in the digital twin with adjusted parameters, record the simulated verification result data, and generate simulated verification result data. The simulated verification results data are compared and analyzed with the actual abnormal data collected from the physical production line to verify the validity of the root cause hypothesis in the explanatory root cause hypothesis set and generate the verified valid root cause set. Based on the verified set of valid root causes, specific adjustment commands are generated for the physical pipeline equipment parameters or scheduling logic, forming an optimized control instruction set.

[0012] As a preferred embodiment of the monitoring method for the automatic calibration pipeline of single-phase smart meters described in this invention, the method includes the following steps: executing an optimized control instruction set to adjust the operating parameters of the automatic calibration pipeline of single-phase smart meters, and collecting a set of feedback data after the strategy execution. The optimized control instruction set is sent to the programmable logic controller and power source of the single-phase smart energy meter automatic verification production line through the control interface for execution. After the optimized control instruction set is executed, a new round of operation dataset, environmental operation dataset, and meter image data are collected from the automatic verification pipeline of single-phase smart energy meters under the new operating parameters. Add timestamps and workstation space identifiers to the new round of operation dataset, environmental operation dataset, and electricity meter image data to form a set of feedback data after strategy execution.

[0013] As a preferred embodiment of the monitoring method for the automatic calibration pipeline of single-phase smart energy meters described in this invention, the method includes the following steps: Incremental learning and updating of the spatiotemporal graph neural network model and the calibration pipeline knowledge graph using the feedback data set after strategy execution; The feedback data set after the strategy execution is cleaned and feature extracted to generate a new, clean and well-organized set of multidimensional feature vectors. The new, clean, and well-organized set of multidimensional feature vectors and their corresponding anomaly labels are used as incremental training samples to fine-tune the parameters of the spatiotemporal graph neural network model, generating an updated spatiotemporal graph neural network model. Based on the correctness of the interpretable root cause hypothesis set verified by the feedback data set after strategy execution, the confidence weights of the corresponding entity relationships in the verification pipeline knowledge graph are updated, and the updated verification pipeline knowledge graph is generated.

[0014] In a second aspect, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the monitoring method for an automatic calibration pipeline of single-phase smart energy meters as described in the first aspect of the present invention.

[0015] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the monitoring method for an automatic calibration pipeline of a single-phase smart energy meter as described in the first aspect of the present invention.

[0016] The beneficial effects of this invention are as follows: By deeply integrating spatiotemporal graph neural networks and knowledge graphs, dynamic identification and root cause reasoning of anomaly propagation patterns in the inspection process are realized. The spatiotemporal correlation characteristics between multi-source data are analyzed using spatiotemporal graph neural networks to accurately identify anomaly propagation paths and generate probability graphs. Cross-modal reasoning is performed based on knowledge graphs, integrating data-driven anomaly patterns with equipment knowledge to generate interpretable root cause hypotheses. Hypotheses are verified through digital twins and optimization instructions are generated, forming a monitoring-diagnosis-optimization-self-learning control system, which effectively solves the problem of cross-modal processing in traditional methods. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of a monitoring method for an automated calibration production line for single-phase smart energy meters.

[0019] Figure 2 A flowchart for generating a pure and well-ordered set of multidimensional feature vectors. Detailed Implementation

[0020] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0021] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0022] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0023] Reference Figure 1 and Figure 2 This is one embodiment of the present invention, which provides a monitoring method for an automatic calibration production line of single-phase smart energy meters, including the following steps: S1. Collect the operation dataset, environmental operation dataset, and meter image data of each station in the automatic calibration production line for single-phase smart energy meters to form an initial multi-source data set with spatiotemporal markers.

[0024] S1.1 Collect the operation data set, environmental operation data set, and meter image data of each station in the automatic calibration production line for single-phase smart energy meters.

[0025] Furthermore, in the automated calibration production line for single-phase smart energy meters, the high-precision standard energy meter at the withstand voltage test station collects voltage and current waveform data, the power source at the error calibration station records power output parameters, and the programmable logic controller reads the positioning coordinates of the robotic arm and the values ​​of the station status register, together forming an operational dataset. Temperature and humidity sensors distributed in the calibration chamber periodically record ambient temperature and relative humidity data, forming an environmental operational dataset. A high-resolution industrial camera is triggered to capture images of the nameplates and terminals of the single-phase smart energy meters flowing through each station, generating energy meter image data. S1.2. Add timestamps and workstation space identifiers to the running dataset, environmental running dataset, and electricity meter image data.

[0026] Furthermore, the operating dataset consisting of voltage and current waveform data collected at the withstand voltage test station, power output parameters recorded at the error verification station, robot positioning coordinates read by the programmable logic controller, and station status register values, as well as the environmental operating dataset consisting of ambient temperature and relative humidity data recorded by the temperature and humidity sensor, are each appended with millisecond-precision timestamps synchronized by the network time protocol and marked with a unique identifier of the source station. The energy meter image data consisting of images of the nameplate and terminal blocks of a single-phase smart energy meter captured by a high-resolution industrial camera is appended with a timestamp of the image capture time and an image acquisition station identifier.

[0027] S1.3 The running dataset with timestamps and workstation space identifiers, the environmental running dataset, and the electricity meter image data are aggregated to form an initial multi-source data set with spatiotemporal tags.

[0028] Furthermore, the operational dataset, environmental operational dataset, and energy meter image data, all with timestamps and workstation space identifiers, are transmitted to the central processing unit via a data bus. They are then sorted and stored according to the timestamp sequence and workstation space identifiers, and integrated to form an initial multi-source data set with spatiotemporal markers. Each data record in the initial multi-source data set contains a timestamp, a workstation space identifier, and the corresponding measurement value or image feature.

[0029] S2. Perform data cleaning and feature extraction on the initial multi-source dataset with spatiotemporal labels to generate a clean and well-organized set of multidimensional feature vectors.

[0030] S2.1 Perform data alignment and outlier removal on the initial multi-source dataset with spatiotemporal markers to generate a time-aligned dataset.

[0031] Furthermore, based on the timestamp and workstation space identifier attached to each record in the initial multi-source dataset with spatiotemporal markers, the operation dataset records from different workstations but with timestamp differences within a preset tolerance range are aligned with the environmental operation dataset records. The withstand voltage test workstation voltage data and the error verification workstation power data are aligned to the same time point. Isolated data points that fail to align due to communication delays are removed. For the data that has been time-aligned, a sliding window method based on the three-standard-deviation criterion is used to identify and remove outliers that significantly deviate from the normal range.

[0032] S2.2 Perform Z-score standardization on the runtime dataset and environment runtime dataset in the dataset with time-series alignment anomalies to generate a standardized runtime environment dataset.

[0033] Furthermore, all numerical fields of the running dataset and the environment running dataset are extracted from the time-aligned dataset, and the arithmetic mean and standard deviation of each field are calculated. For the value of each data point, the arithmetic mean of the field to which it belongs is subtracted and then divided by the standard deviation of the field to which it belongs, thus completing the Z-score standardization process.

[0034] S2.3. Perform grayscale conversion, Gaussian filtering for noise reduction, and Canny edge detection on the energy meter image data in the time-aligned dataset to generate a visual feature matrix.

[0035] Furthermore, the energy meter image data is extracted from the time-aligned dataset, the color energy meter image data is converted into grayscale images, and then a Gaussian filter with a standard deviation of 1.5 is applied to the grayscale image for convolution to smooth the noise. The Canny edge detection operator is then used to process the denoised grayscale image, extract the edge contour information of the nameplate characters and wiring terminals in the image, and convert the processed image data into a numerical matrix representation to generate a visual feature matrix.

[0036] S2.4. The standardized operating environment data set and the visual feature matrix are spliced ​​together and subjected to principal component analysis for dimensionality reduction to generate a clean and well-organized set of multidimensional feature vectors.

[0037] Furthermore, the standardized values ​​in the standardized operating environment dataset and the matrix values ​​in the visual feature matrix are aligned and spliced ​​according to the timestamps of the data records to form a high-dimensional feature vector. Principal component analysis is then applied to this high-dimensional feature vector to project the original high-dimensional feature vector onto a new space composed of principal components, generating a pure and regular set of multi-dimensional feature vectors with lower dimensions and no correlation between dimensions.

[0038] S3. Input the pure and regular set of multidimensional feature vectors into the spatiotemporal graph neural network model to perform anomaly propagation pattern recognition and output the anomaly propagation probability map.

[0039] S3.1 Map the pure and regular set of multidimensional feature vectors onto a graph structure with assembly line workstations as nodes and the connection relationships between workstations as edges to generate spatiotemporal graph structure data.

[0040] Furthermore, each feature vector in the pure and regular multidimensional feature vector set is assigned to a corresponding node in the graph structure with the production line workstations as nodes, based on its source workstation space identifier; according to the physical layout of the production line and the material flow direction, the connection relationship between workstation nodes is defined as the edge of the graph, for example, there is a directed edge from the pressure test workstation node to the error verification workstation node; each node is attached with a corresponding pure and regular multidimensional feature vector at each time, generating spatiotemporal graph structure data containing node features, edge connection relationships and time series.

[0041] S3.2 Input the spatiotemporal graph structure data into the graph convolutional layer of the spatiotemporal graph neural network model, extract the spatial dependency features between nodes, and generate spatial feature representations.

[0042] Furthermore, the spatiotemporal graph structure data is input into the graph convolutional layer of the spatiotemporal graph neural network model. Based on the adjacency relationship of the graph, the graph convolutional layer updates the node representation by aggregating the feature information of each node and its neighboring nodes. The features of the pressure test station node will be fused with the features of its neighboring error verification station node. After multiple layers of graph convolution operations, the spatial dependency features between each station node in the graph are extracted and enhanced to generate spatial feature representation.

[0043] It should be noted that the initial multi-source dataset with spatiotemporal labels from historical normal and abnormal detection periods, along with their labeled real anomaly propagation paths, are used to minimize the cross-entropy loss between the anomaly propagation probability map predicted by the spatiotemporal graph neural network model and the real path. The parameters of the spatiotemporal graph neural network model are then iteratively updated using the Adam optimizer to obtain the trained spatiotemporal graph neural network model.

[0044] S3.3 Input the spatial feature representation into the gated recurrent unit layer of the spatiotemporal graph neural network model to learn the dynamic change pattern on the time series and generate spatiotemporal fusion features.

[0045] Furthermore, the spatial feature representations are input sequentially into the gated recurrent unit layer of the spatiotemporal graph neural network model in time series. The gated recurrent unit layer learns the dynamic evolution of the features of each workstation node in the time dimension through its internal gating mechanism, such as capturing the trend changes of the features of the error verification workstation node over time. The output of the gated recurrent unit layer integrates spatially dependent features and time dynamic change patterns to generate spatiotemporally fused features.

[0046] S3.4 Based on the spatiotemporal fusion features, calculate the probability of abnormal states of each node and edge in the graph, and generate an anomaly propagation probability graph.

[0047] Furthermore, the spatiotemporal fusion features are input into an output layer consisting of a fully connected layer and a Sigmoid activation function. The output layer calculates an abnormal state probability value between 0 and 1 for each node and edge in the spatiotemporal graph structure data. The abnormal state probability values ​​of all nodes and edges together form an abnormal propagation probability graph, which quantitatively indicates the risk level of abnormality or propagation of abnormality at a specific time in each workstation and the connection path between workstations in the pipeline.

[0048] S4. Based on the anomaly propagation probability graph, trigger the verification pipeline knowledge graph to perform cross-modal root cause reasoning and generate an interpretable set of root cause hypotheses.

[0049] S4.1 Based on the anomaly propagation probability map, identify paths with a probability exceeding the probability distribution of successful early warnings in the historical anomaly propagation probability map, and set anomaly propagation path threshold by comprehensively considering the balance between false alarm rate and false negative rate, thereby generating a set of high-risk anomaly paths.

[0050] Furthermore, the probability distribution of historical anomaly propagation probability maps that have been verified as real anomaly propagation paths is analyzed, and it is determined that the probability values ​​are mainly concentrated above 0.7. In order to balance the risks of false alarms and false negatives, the threshold for anomaly propagation paths is set to 0.7. All paths in the current anomaly propagation probability map are traversed, and anomaly propagation paths with a probability value greater than or equal to the threshold of 0.7 are selected to generate a set of high-risk anomaly paths.

[0051] S4.2 Match and map the set of high-risk abnormal paths with the entities and relationships in the knowledge graph of the verification pipeline to generate a set of candidate fault modes.

[0052] Furthermore, each path in the set of high-risk abnormal paths is matched with an entity in the knowledge graph of the verification pipeline; a high-risk path from the withstand voltage test station to the error verification station is mapped to the potential consequences of the existence of the power source entity in the knowledge graph, the relationship between voltage fluctuation and the impact of the voltage fluctuation entity, and the error calculation relationship; by traversing the fault mode class entities associated with the high-risk path nodes in the knowledge graph, a set of candidate fault modes is generated.

[0053] S4.3 Perform confidence inference calculation based on graph relationships on the candidate failure mode set to generate an interpretable root cause hypothesis set with confidence levels.

[0054] The confidence expression for the root cause is: ; in, Candidate Failure Modes Confidence level, abnormal propagation path The risk probability value, Fault mode Prior weights, This is a high-risk, abnormal transmission route. Candidate failure modes This is a set of high-risk abnormal paths. The strength of the associations defined in the knowledge graph.

[0055] Furthermore, for each failure mode in the candidate failure mode set, a confidence calculation formula is applied. The formula is that the confidence of a failure mode is equal to the sum of the products of the risk probability values ​​of all high-risk abnormal paths and the correlation strength of the failure mode, divided by the sum of the risk probability values ​​of all high-risk abnormal paths, and then multiplied by the prior weight of the failure mode. After the calculation is completed, an interpretable root cause hypothesis with confidence is generated for each failure mode, and all hypotheses constitute a set of interpretable root cause hypotheses with confidence.

[0056] It should be noted that the knowledge graph of the calibration pipeline is constructed based on equipment manuals, historical maintenance records, and domain expert knowledge. The knowledge graph includes entity types such as equipment components, failure modes, and calibration parameters, as well as relationship types such as potential causes, impacts, and location at workstations. The power source entity is associated with the output instability failure mode entity through potential causes, and the output instability failure mode entity is associated with the error value calibration parameter entity through impacts. The strength weight of the relationship between entities is assigned based on historical co-occurrence frequency or expert experience, and the prior weight of the failure mode is set based on its historical occurrence frequency.

[0057] S5. Verify the set of explainable root cause hypotheses in the digital twin of the automatic calibration pipeline for single-phase smart energy meters, and generate the corresponding set of optimized control instructions.

[0058] S5.1 In the digital twin of the automatic calibration pipeline for single-phase smart energy meters, the performance parameters of the corresponding virtual components are adjusted according to the root cause assumptions in the set of interpretable root cause assumptions, and the adjusted digital twin is generated.

[0059] Furthermore, in the digital twin of the automatic calibration pipeline for single-phase smart energy meters, the virtual component pointed to by the root cause hypothesis in the set of explainable root cause hypotheses is located. For the root cause hypothesis of power source module aging, the corresponding virtual power source component in the digital twin is located. According to the fault mode described by the root cause hypothesis, the performance parameters of the virtual component are adjusted. After the parameter adjustment is completed, the parameter-adjusted digital twin is generated.

[0060] S5.2 Run the simulated verification process in the digital twin after parameter adjustment, record the simulated output verification result data, and generate simulated verification result data.

[0061] Furthermore, in the digital twin with adjusted parameters, the virtual energy meter is driven through the complete process from online testing, withstand voltage testing, error verification to offline testing, following the standard verification procedure of the automatic verification line for single-phase smart energy meters. During this simulated operation, data such as voltage, current, power, and error values ​​output by the virtual standard energy meter and the virtual power source are recorded to generate simulated verification result data.

[0062] S5.3 Compare and analyze the simulated verification results with the actual abnormal data collected from the physical production line to verify the validity of the root cause hypothesis in the explanatory root cause hypothesis set and generate the verified valid root cause set.

[0063] Furthermore, key indicators in the simulated verification results data, such as the error value sequence of the error verification station, are compared with the error value sequence of the same station and the same time period in the actual abnormal data collected from the physical production line; similarity indicators of the two sets of data sequences, such as the Pearson correlation coefficient; if the similarity index exceeds a preset threshold (e.g., 0.8), the root cause hypothesis is determined to be valid; all root cause hypotheses in the explainable root cause hypothesis set are traversed, and the root cause hypotheses that pass the verification are selected to generate a set of valid root causes after verification.

[0064] S5.4 Based on the verified set of valid root causes, generate specific adjustment commands for physical pipeline equipment parameters or scheduling logic to form an optimized control instruction set.

[0065] Furthermore, based on each valid root cause hypothesis in the verified valid root cause set, corresponding specific equipment parameter adjustment commands or pipeline scheduling logic adjustment commands are generated. For the root cause hypothesis of power source module aging in the verified valid root cause set, a scheduling logic adjustment command is generated to route subsequent verification tasks to the backup power source device. The command content is to switch the power source of workstation ST01 to backup source B. For the root cause hypothesis of ambient temperature sensor reading drift in the verified valid root cause set, a parameter adjustment command is generated to perform software calibration of the physical ambient temperature sensor. The command content is to apply an offset of -0.5℃ to the calibration of sensor T001. All specific adjustment commands for physical pipeline equipment parameters or scheduling logic are summarized and encapsulated to form an optimized control instruction set.

[0066] S6. Execute the optimized control instruction set to adjust the operating parameters of the automatic calibration pipeline of single-phase smart energy meters, and collect the feedback data set after the strategy is executed.

[0067] S6.1. The optimized control instruction set is sent to the programmable logic controller and power source of the single-phase smart energy meter automatic verification production line through the control interface for execution.

[0068] Furthermore, through industry-standard control interfaces such as the OPC UA protocol or Modbus TCP protocol, the specific adjustment commands encapsulated in the optimized control instruction set are sent to the programmable logic controller and power source equipment of the single-phase smart energy meter automatic verification production line; the optimized control instruction set, which includes the command to switch the power source of station ST01 to the backup source B, is sent to the programmable logic controller, which executes the command and drives the corresponding electrical switch to complete the power source switching operation.

[0069] S6.2 After the optimized control instruction set is executed, the new round of operation data set, environmental operation data set, and meter image data generated by the single-phase smart energy meter automatic verification pipeline under the new operating parameters are collected.

[0070] Furthermore, after the optimized control instruction set has been executed and the production line is running stably, a new round of data acquisition is initiated. The voltage and current parameters output by the switched backup power source, the error values ​​measured by the high-precision standard energy meter, and the workstation status information read by the programmable logic controller are collected to form a new round of operational dataset. Simultaneously, data recorded by the temperature and humidity sensors in the calibration chamber are collected to form a new round of environmental operational dataset. The industrial camera is triggered to capture images of the single-phase smart energy meters flowing through each workstation, forming a new round of energy meter image data.

[0071] S6.3. Add timestamps and workstation space identifiers to the new round of operation dataset, environmental operation dataset, and electricity meter image data to form a set of feedback data after strategy execution.

[0072] Furthermore, for the voltage, current, error values, and other data in the new round of operation dataset, the temperature and humidity data in the new round of environmental operation dataset, and the new round of electricity meter image data, respectively, a precise timestamp of the data acquisition time and a unique spatial identifier of the source workstation are added. All the new round of data with the added timestamp and workstation spatial identifier are aggregated and stored in time series to form a set of feedback data after strategy execution.

[0073] S7. Use the feedback data set after strategy execution to incrementally learn and update the spatiotemporal graph neural network model and the verification pipeline knowledge graph.

[0074] S7.1 Perform data cleaning and feature extraction on the feedback data set after the strategy is executed to generate a new, clean, and well-organized set of multidimensional feature vectors.

[0075] Furthermore, the feedback data set after strategy execution is cleaned and its features are extracted; data alignment and outlier removal are performed on the feedback data set after strategy execution to generate a new time-aligned data set; Z-score standardization is performed on the running dataset and environment running dataset in the new time-aligned data set to generate a new standardized running environment dataset; grayscale conversion, Gaussian filtering for noise reduction, and Canny edge detection are performed on the electricity meter image data in the new time-aligned data set to generate a new visual feature matrix; the new standardized running environment dataset and the new visual feature matrix are concatenated and subjected to principal component analysis for dimensionality reduction to finally generate a new, clean, and well-organized set of multidimensional feature vectors.

[0076] S7.2. Using the new, clean, and well-organized set of multidimensional feature vectors and the corresponding anomaly labels as incremental training samples, the parameters of the spatiotemporal graph neural network model are fine-tuned to generate an updated spatiotemporal graph neural network model.

[0077] Furthermore, the new, clean, and well-organized set of multidimensional feature vectors is mapped to new spatiotemporal graph structure data. Based on the actual verification results corresponding to the feedback data set after policy execution, the new spatiotemporal graph structure data is labeled to indicate whether there are abnormal propagation paths. Using the newly labeled spatiotemporal graph structure data as incremental training samples, a smaller learning rate than the initial training is used. The Adam optimizer minimizes the cross-entropy loss between the predicted output of the spatiotemporal graph neural network model and the true label, and the existing parameters in the spatiotemporal graph neural network model are fine-tuned to generate an updated spatiotemporal graph neural network model.

[0078] S7.3. Based on the correctness of the set of interpretable root cause hypotheses verified by the feedback data set after strategy execution, update the confidence weights of the corresponding entity relationships in the verification pipeline knowledge graph, and generate the updated verification pipeline knowledge graph.

[0079] Furthermore, based on the correctness of each root cause hypothesis in the interpretable root cause hypothesis set verified by the feedback data set after strategy execution, the confidence weights of the corresponding entity relationships in the verification pipeline knowledge graph are adjusted. If the root cause hypothesis of power source module aging is verified as correct, the weight value of the possible relationship between the power source entity and the output instability fault mode entity in the verification pipeline knowledge graph is increased; if a root cause hypothesis is proven to be incorrect, the weight of its association relationship is reduced accordingly. After all weight adjustments are completed, an updated verification pipeline knowledge graph is generated.

[0080] This embodiment also provides a computer device applicable to the monitoring method for an automatic verification line of single-phase smart energy meters, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the monitoring method for an automatic verification line of single-phase smart energy meters as proposed in the above embodiment.

[0081] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0082] This embodiment also provides a storage medium storing a computer program. When executed by a processor, the program implements the monitoring method for an automatic verification pipeline for single-phase smart meters as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0083] In summary, this invention achieves dynamic identification and root cause reasoning of anomaly propagation patterns in the inspection pipeline through the deep integration of spatiotemporal graph neural networks and knowledge graphs. It utilizes spatiotemporal graph neural networks to analyze the spatiotemporal correlation characteristics between multi-source data, accurately identifies anomaly propagation paths and generates probability graphs, performs cross-modal reasoning based on knowledge graphs, integrates data-driven anomaly patterns with equipment knowledge, generates interpretable root cause hypotheses, verifies hypotheses through digital twins and generates optimization instructions, forming a monitoring-diagnosis-optimization-self-learning control system, effectively solving the problem of cross-modal issues encountered by traditional methods.

[0084] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A monitoring method for an automated calibration production line of single-phase smart energy meters, characterized in that: This includes collecting operational datasets, environmental operational datasets, and meter image data from each workstation of the automatic calibration production line for single-phase smart meters, forming an initial multi-source dataset with spatiotemporal markers; Data cleaning and feature extraction are performed on the initial multi-source dataset with spatiotemporal labels to generate a clean and well-organized set of multidimensional feature vectors. A clean and well-ordered set of multidimensional feature vectors is input into a spatiotemporal graph neural network model to perform anomaly propagation pattern recognition and output an anomaly propagation probability map. Based on the anomaly propagation probability graph, the knowledge graph of the triggering verification pipeline is used to perform cross-modal root cause reasoning and generate an interpretable set of root cause hypotheses. In the digital twin of the automatic calibration pipeline for single-phase smart energy meters, the set of interpretable root cause hypotheses is verified and the corresponding set of optimized control instructions is generated. The system executes an optimized control instruction set to adjust the operating parameters of the automatic calibration pipeline for single-phase smart energy meters, and collects a set of feedback data after the strategy execution. The system then uses this set of feedback data to incrementally learn and update the spatiotemporal neural network model and the calibration pipeline knowledge graph.

2. The monitoring method for an automatic calibration production line of single-phase smart energy meters as described in claim 1, characterized in that: The process involves collecting operational datasets, environmental operation datasets, and meter image data from each workstation of the automatic calibration line for single-phase smart meters to form an initial multi-source dataset with spatiotemporal tags, including the following steps: Collect operational data sets, environmental operation data sets, and meter image data from each workstation of the automatic calibration production line for single-phase smart meters; Add timestamps and workstation space identifiers to the runtime dataset, environmental runtime dataset, and electricity meter image data; The runtime dataset with timestamps and workstation space identifiers, the environmental runtime dataset, and the electricity meter image data are aggregated to form an initial multi-source dataset with spatiotemporal tags.

3. The monitoring method for an automatic calibration production line of single-phase smart energy meters as described in claim 2, characterized in that: The initial multi-source dataset with spatiotemporal labels is cleaned and its features are extracted to generate a clean and well-organized set of multidimensional feature vectors. This process includes the following steps: Data alignment and outlier removal are performed on the initial multi-source dataset with spatiotemporal tags to generate a time-aligned dataset; Z-score normalization is performed on the runtime dataset and environment runtime dataset in the dataset with time-series alignment anomalies to generate a normalized runtime environment dataset; The energy meter image data in the time-aligned dataset is processed by grayscale conversion, Gaussian filtering for noise reduction, and Canny edge detection to generate a visual feature matrix. The standardized operating environment dataset and the visual feature matrix are spliced ​​together and subjected to principal component analysis for dimensionality reduction, generating a clean and well-organized set of multidimensional feature vectors.

4. The monitoring method for an automatic calibration production line of single-phase smart energy meters as described in claim 3, characterized in that: The pure and well-ordered set of multidimensional feature vectors is input into a spatiotemporal graph neural network model for anomaly propagation pattern recognition, outputting an anomaly propagation probability map. This includes the following steps: A set of clean and well-ordered multidimensional feature vectors is mapped onto a graph structure with assembly line workstations as nodes and inter-workstation connections as edges to generate spatiotemporal graph structure data. The initial multi-source dataset with spatiotemporal labels for historical normal and abnormal detection periods, along with their labeled real anomaly propagation paths, are used to minimize the cross-entropy loss between the anomaly propagation probability map predicted by the spatiotemporal graph neural network model and the real path. The parameters of the spatiotemporal graph neural network model are then iteratively updated using the Adam optimizer to obtain the trained spatiotemporal graph neural network model. The spatiotemporal graph structure data is input into the graph convolutional layer of the spatiotemporal graph neural network model to extract the spatial dependency features between nodes and generate spatial feature representations. Spatial feature representations are input into the gated recurrent unit layer of the spatiotemporal graph neural network model to learn dynamic change patterns on the time series and generate spatiotemporal fusion features; Based on the spatiotemporal fusion characteristics, the probability of abnormal states of each node and edge in the graph is calculated to generate an anomaly propagation probability graph.

5. The monitoring method for an automatic calibration production line of single-phase smart energy meters as described in claim 4, characterized in that: Based on the anomaly propagation probability graph, a cross-modal root cause reasoning is performed using the knowledge graph of the triggering pipeline to generate an interpretable set of root cause hypotheses, including the following steps: Based on the anomaly propagation probability map, paths with a probability exceeding the probability distribution of successful early warnings in the historical anomaly propagation probability map are identified, and anomaly propagation path thresholds are set by comprehensively balancing the false alarm rate and the false negative rate, thereby generating a set of high-risk anomaly paths. The set of high-risk abnormal paths is matched and mapped with entities and relationships in the knowledge graph of the verification pipeline to generate a set of candidate failure modes. Calculate confidence inference based on graph relationships for the candidate failure mode set to generate interpretable root cause hypotheses with confidence levels.

6. The monitoring method for an automatic calibration production line of single-phase smart energy meters as described in claim 5, characterized in that: In the digital twin of the automatic calibration pipeline for single-phase smart energy meters, the explainable set of root cause hypotheses is verified, and the corresponding optimized control instruction set is generated, including the following steps: In the digital twin of the automatic calibration pipeline for single-phase smart energy meters, the performance parameters of the corresponding virtual components are adjusted according to the root cause assumptions in the set of interpretable root cause assumptions to generate a digital twin with adjusted parameters. Run the simulated verification process in the digital twin with adjusted parameters, record the simulated verification result data, and generate simulated verification result data. The simulated verification results data are compared and analyzed with the actual abnormal data collected from the physical production line to verify the validity of the root cause hypothesis in the explanatory root cause hypothesis set and generate the verified valid root cause set. Based on the verified set of valid root causes, specific adjustment commands are generated for the physical pipeline equipment parameters or scheduling logic, forming an optimized control instruction set.

7. The monitoring method for an automatic calibration production line of single-phase smart energy meters as described in claim 6, characterized in that: Execute an optimized control instruction set to adjust the operating parameters of the automatic calibration pipeline for single-phase smart energy meters, and collect a set of feedback data after the strategy execution, including the following steps: The optimized control instruction set is sent to the programmable logic controller and power source of the single-phase smart energy meter automatic verification production line through the control interface for execution. After the optimized control instruction set is executed, a new round of operation dataset, environmental operation dataset, and meter image data are collected from the automatic verification pipeline of single-phase smart energy meters under the new operating parameters. Add timestamps and workstation space identifiers to the new round of operation dataset, environmental operation dataset, and electricity meter image data to form a set of feedback data after strategy execution.

8. The monitoring method for an automatic calibration production line of single-phase smart energy meters as described in claim 7, characterized in that, The spatiotemporal graph neural network model and the knowledge graph of the verification pipeline are incrementally updated using the feedback data set after policy execution, including the following steps: The feedback data set after the strategy execution is cleaned and feature extracted to generate a new, clean and well-organized set of multidimensional feature vectors. The new, clean, and well-organized set of multidimensional feature vectors and their corresponding anomaly labels are used as incremental training samples to fine-tune the parameters of the spatiotemporal graph neural network model, generating an updated spatiotemporal graph neural network model. Based on the correctness of the interpretable root cause hypothesis set verified by the feedback data set after strategy execution, the confidence weights of the corresponding entity relationships in the verification pipeline knowledge graph are updated, and the updated verification pipeline knowledge graph is generated.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the monitoring method for the automatic calibration pipeline of single-phase smart energy meters as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the monitoring method for the automatic verification pipeline of single-phase smart energy meters as described in any one of claims 1 to 8.

Images