Artificial intelligence-based wiring calibrator output data processing method and system

Abstract

The application discloses a wiring calibration instrument output data processing method and system based on artificial intelligence, a wiring topology structure is modeled into a graph structure based on a GNN graph neural network, nodes represent electric meter terminals, and edges represent wiring relationships, a hybrid model is constructed by combining GCN graph convolution and GRU neural network processing dynamic data, the hyperparameters of the hybrid model are optimized through an improved IWOA whale optimization algorithm to obtain a target G hybrid model, the standard multi-source data are input into the target hybrid model for training, and a wiring calibration instrument error index is output, a nonlinear error correction model is constructed based on SVR support vector regression, the wiring calibration instrument error index is input, and a correction coefficient is output. Compared with a traditional calibration process, the method is shorter in time consumption, higher in efficiency, supports batch calibration, and can simultaneously process calibration and correction tasks of multiple electric meters through a wireless communication module.

Description

Technical Field

[0001] This invention relates to the field of power metering equipment calibration technology, and in particular to a method and system for processing output data of a wiring calibration instrument based on artificial intelligence. Background Technology

[0002] In the field of power system wiring calibration, wiring calibrators are critical equipment, and the accuracy of their output data directly affects power metering, equipment operation safety, and power system stability. Traditional wiring calibrators often use linear correction models for output data processing, which are insufficient to handle the complex characteristics of nonlinear errors in power systems, resulting in limited calibration accuracy. Furthermore, they typically rely on manual experience and lack scientific optimization mechanisms. When wiring topology or operating conditions change, their adaptability to the power system is poor, requiring frequent manual adjustments, which increases maintenance costs and operational complexity. Summary of the Invention

[0003] The purpose of this invention is to solve the above problems by designing an artificial intelligence-based method and system for processing output data of a wiring calibrator.

[0004] To achieve the above objectives, the technical solution of the present invention further includes the following steps in the above-mentioned artificial intelligence-based wiring calibrator output data processing method:

[0005] Voltage, current and power data are collected and wiring topology information is recorded. After noise interference is eliminated by using a sliding window filtering algorithm, abnormal data points are removed by using an isolated forest algorithm to obtain standard multi-source data.

[0006] The wiring topology is modeled as a graph structure based on the GNN graph neural network, where nodes represent meter terminals and edges represent wiring relationships. The GCN graph convolution and GRU neural network are combined to process dynamic data and construct a GNN-GRU hybrid model.

[0007] The hyperparameters of the GNN-GRU hybrid model are optimized using the improved IWOA whale optimization algorithm to obtain the target GNN-GRU hybrid model.

[0008] The standard multi-source data is input into the target GNN-GRU hybrid model for training, and the error index of the wiring calibrator is output.

[0009] A nonlinear error correction model is constructed based on SVR support vector regression, with the error index of the wiring calibrator as input and the correction coefficient as output.

[0010] Furthermore, in the aforementioned AI-based wiring calibration instrument output data processing method, the acquisition of voltage, current, and power data, recording of wiring topology information, elimination of noise interference using a sliding window filtering algorithm, and removal of outlier data points using an isolated forest algorithm yield standard multi-source data, including:

[0011] The calibrator collects voltage, current, and power data using its built-in sensors and microprocessor, and records wiring topology information.

[0012] Set the window size of the sliding window filtering algorithm to 10, and arrange the collected voltage, current and power data in chronological order;

[0013] Starting with the first data point, take 10 consecutive data points as the first window, calculate the average value of the data within the window, and use the average value as the filtered data at the center of the window. Repeat this calculation to obtain the processed target filtered data.

[0014] Furthermore, in the aforementioned AI-based wiring calibration instrument output data processing method, the steps of collecting voltage, current, and power data, recording wiring topology information, eliminating noise interference using a sliding window filtering algorithm, and removing outlier data points using an isolated forest algorithm to obtain standard multi-source data also include:

[0015] The number of trees in the isolated forest algorithm is set to 100, the sample subset size is 256, the processed target filtered data is input into the isolated forest algorithm, the distribution of abnormal scores of normal data is calculated, and the 95th percentile of abnormal scores is used as the threshold.

[0016] When the outlier score of the data exceeds the threshold, it is identified as an outlier data point; outlier data points are removed to obtain standard multi-source data.

[0017] Furthermore, in the aforementioned AI-based wiring calibration instrument output data processing method, the wiring topology is modeled as a graph structure based on a GNN graph neural network, where nodes represent meter terminals and edges represent wiring relationships. A GNN-GRU hybrid model is constructed by combining GCN graph convolution and GRU neural network to process dynamic data, including:

[0018] Construct an initial graph structure matrix, and generate an adjacency matrix based on the node feature set and the edge connection attributes to quantify the strength of the association between nodes;

[0019] Spatial feature learning is achieved through multi-layer GCN graph convolution. The first layer graph convolution focuses on local connectivity to extract feature interaction patterns between adjacent terminals. The deep network aggregates global topological information to capture indirect associations between terminals across layers.

[0020] After GCN processing, the features of each node are fused with its own state and spatial association information in the whole topology network to generate a spatial feature vector with a global perspective.

[0021] Furthermore, in the aforementioned AI-based wiring calibrator output data processing method, the step of optimizing the hyperparameters of the GNN-GRU hybrid model using the improved IWOA whale optimization algorithm to obtain the target GNN-GRU hybrid model includes:

[0022] Initialize the whale population, with each individual corresponding to a set of hyperparameters, and generate the initial population from a set search range through random sampling;

[0023] The model's performance on the validation set is used as the fitness function to evaluate the model performance for each set of hyperparameters. The position of each individual is updated according to the improved IWOA search rule based on the fitness value, simulating whale encirclement, bubble net attack and random search behavior.

[0024] When the change in the optimal fitness value is less than a set threshold in multiple consecutive iterations, the optimal parameter combination is output to obtain the target GNN-GRU hybrid model.

[0025] Furthermore, in the above-mentioned AI-based wiring calibrator output data processing method, the step of inputting the standard multi-source data into the target GNN-GRU hybrid model for training and outputting the wiring calibrator error index includes:

[0026] The standard multi-source data is input into the target GNN-GRU hybrid model, and the corresponding wiring calibrator error index is output. The error index must include at least the overall system error mean, the individual error value of each meter terminal, and the error distribution range under different wiring relationships.

[0027] Furthermore, in the aforementioned AI-based wiring calibrator output data processing method, the step of constructing a nonlinear error correction model based on SVR support vector regression, inputting the wiring calibrator error index, and outputting correction coefficients includes:

[0028] Input the error index of the wiring calibrator into the nonlinear error correction model constructed based on SVR support vector regression to output the correction coefficient;

[0029] Establish a mapping relationship between correction coefficients and actual calibration operations, and convert the generated correction coefficients into executable calibration instructions, including at least adjusting the measurement threshold and compensation coefficient of the calibrator.

[0030] Furthermore, in the AI-based wiring calibrator output data processing system, the wiring calibrator output data processing system includes the following modules:

[0031] The multi-source data acquisition module is used to acquire voltage, current and power data, and record wiring topology information. After eliminating noise interference using a sliding window filtering algorithm, it removes abnormal data points using an isolated forest algorithm to obtain standard multi-source data.

[0032] The hybrid model building module is used to model the wiring topology as a graph structure based on the GNN graph neural network. Nodes represent meter terminals and edges represent wiring relationships. It combines GCN graph convolution and GRU neural network to process dynamic data and build a GNN-GRU hybrid model.

[0033] The hybrid model optimization module is used to optimize the hyperparameters of the GNN-GRU hybrid model using the improved IWOA whale optimization algorithm to obtain the target GNN-GRU hybrid model.

[0034] The error index output module is used to input the standard multi-source data into the target GNN-GRU hybrid model for training and output the wiring calibrator error index.

[0035] The correction coefficient calculation module is used to construct a nonlinear error correction model based on SVR support vector regression. It takes the error index of the wiring calibrator as input and outputs the correction coefficient.

[0036] Furthermore, in the AI-based wiring calibrator output data processing system, the hybrid model optimization module includes the following sub-modules:

[0037] The generation submodule is used to initialize the whale population. Each individual corresponds to a set of hyperparameters, and the initial population is generated from a set search range by random sampling.

[0038] The evaluation submodule is used to evaluate the model's performance on the validation set as a fitness function, assess the model performance for each set of hyperparameters, update the position of each individual according to the improved IWOA search rules based on the fitness value, and simulate whale encirclement, bubble net attack and random search behavior.

[0039] The resulting submodule is used to output the optimal parameter combination when the change in the optimal fitness value is less than a set threshold in multiple consecutive iterations, thus obtaining the target GNN-GRU hybrid model.

[0040] Furthermore, in the AI-based wiring calibrator output data processing system, the correction coefficient calculation module includes the following sub-modules:

[0041] The output submodule is used to input the error index of the wiring calibrator into the nonlinear error correction model built based on SVR support vector regression to output the correction coefficient.

[0042] The mapping submodule is used to establish the mapping relationship between the correction coefficients and the actual calibration operations, and to convert the generated correction coefficients into executable calibration instructions, including at least adjusting the measurement threshold and compensation coefficient of the calibrator.

[0043] Its beneficial effects are as follows: 1. Compared with traditional calibration processes, it requires less time, improves efficiency, and supports batch calibration, handling the calibration and correction tasks of multiple meters simultaneously through a wireless communication module. 2. By designing a dynamic update mechanism, when the wiring topology changes or error characteristics drift, the model can be automatically retrained to generate correction coefficients adapted to the new scenario, enhancing adaptability to dynamically changing environments and reducing manual intervention costs. 3. It directly converts data processing results into executable calibration instructions and records application effects through calibration logs, achieving closed-loop management from data to operation. This not only improves the automation level of wiring calibration but also provides data support for subsequent model optimization and equipment maintenance, helping to continuously improve the overall efficiency and reliability of power system wiring calibration. Attached Figure Description

[0044] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention.

[0045] Figure 1 This is a schematic diagram of the first embodiment of the AI-based wiring calibrator output data processing method in this invention.

[0046] Figure 2 This is a schematic diagram of a second embodiment of the artificial intelligence-based wiring calibrator output data processing method in this invention.

[0047] Figure 3 This is a schematic diagram of the first embodiment of the AI-based wiring calibration instrument output data processing system in this invention. Detailed Implementation

[0048] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0049] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this specification means the presence of features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0050] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 As shown, the data processing method for the output of a wiring calibrator based on artificial intelligence includes the following steps:

[0051] Step 101: Collect voltage, current and power data, and record wiring topology information. After eliminating noise interference using the sliding window filtering algorithm, remove abnormal data points using the isolated forest algorithm to obtain standard multi-source data.

[0052] Specifically, in this embodiment, voltage, current and power data are collected by the built-in sensors and microprocessor of the calibrator, and wiring topology information is recorded;

[0053] Set the window size of the sliding window filtering algorithm to 10, and arrange the collected voltage, current and power data in chronological order;

[0054] Starting with the first data point, take 10 consecutive data points as the first window, calculate the average value of the data within the window, and use the average value as the filtered data at the center of the window. Repeat this calculation to obtain the processed target filtered data.

[0055] The number of trees in the isolated forest algorithm is set to 100, the sample subset size is 256, the processed target filtered data is input into the isolated forest algorithm, the distribution of abnormal scores of normal data is calculated, and the 95th percentile of abnormal scores is used as the threshold.

[0056] When the outlier score of the data exceeds the threshold, it is identified as an outlier data point; outlier data points are removed to obtain standard multi-source data.

[0057] Specifically:

[0058] Multi-dimensional data collection and recording;

[0059] The sensor and microprocessor work together to capture voltage, current, and power data in real time at a preset acquisition frequency (once every 10 milliseconds). During the acquisition process, the timestamps corresponding to the data must be recorded synchronously to ensure the consistency of various parameters over time, providing a basis for subsequent time-series analysis.

[0060] For wiring topology information, the built-in topology identification module records the physical location distribution of meter terminals (installation serial number, circuit number), connection method between terminals (direct connection, connection via adapter), and wire identification information (color, label number). Simultaneously, it associates basic attribute data of the associated equipment, including meter model, rated parameters, installation date, etc., forming a complete topology file.

[0061] The data caching mechanism is designed to temporarily store the collected real-time data in the local cache. Once a certain amount of data (1 minute's worth) has accumulated, it will be processed in batches to avoid the impact of high-frequency data read and write on system performance, while ensuring the continuity and integrity of the data.

[0062] Sliding window filtering eliminates noise interference;

[0063] Determine the size of the sliding window and set a reasonable window length (containing 5-10 consecutive data points) based on the fluctuation characteristics of the data and the intensity of on-site interference. For high-frequency fluctuating data such as voltage and current, use a smaller window size to preserve detailed changes; for relatively stable data such as power, the window size can be appropriately increased to enhance the filtering effect.

[0064] A sliding window filtering operation is performed to smooth the data on a window-by-window basis. Within each window, the mean or median of the data is calculated and replaced with the original data at the center of the window, thereby reducing random fluctuations caused by transient impulse noise and electromagnetic interference. For example, when a single anomalous jump occurs within a window, the mean calculation can reduce its impact on the overall data trend.

[0065] Set the window sliding step size, usually consistent with the data acquisition interval (10 milliseconds), so that the window slides continuously to cover the entire data sequence, ensuring that the filtered data can completely retain the temporal characteristics of the original data, while effectively eliminating high-frequency noise.

[0066] The Isolation Forest algorithm removes outlier data points;

[0067] An anomaly detection sample set is constructed by dividing the data after sliding window filtering into several sample segments according to the time series. Each segment contains voltage, current, and power data and corresponding topological features within a certain time range (whether the wiring status is stable within this period), forming a multi-dimensional detection sample.

[0068] The Isolation Forest algorithm is used to train the sample set. By randomly selecting features and segmentation thresholds, the algorithm gradually isolates samples that deviate significantly from the normal data distribution. For example, when the current value at a certain moment far exceeds the normal fluctuation range within the same time period and does not match the corresponding voltage and power data, the algorithm will identify it as a potential outlier.

[0069] Anomaly detection thresholds are set, and reasonable anomaly scoring thresholds are determined based on the operating specifications of field equipment and the anomaly occurrence rate of historical data. When the anomaly score of a sample exceeds the threshold, it is identified as an anomaly data point and removed. Simultaneously, the occurrence time, value, and corresponding topology status of the anomaly data are recorded to provide a reference for subsequent equipment fault diagnosis.

[0070] Integration and output of standard multi-source data;

[0071] By aligning multi-dimensional data, the filtered and anomaly-removed voltage, current, and power data are matched with the corresponding wiring topology information by timestamp, ensuring that each set of data can be accurately associated with the topology state at a specific moment, forming a spatiotemporally unified dataset.

[0072] Data format standardization is performed to convert different types of data into a unified format (numerical, categorical coding), and metadata such as units and acquisition device numbers are labeled to facilitate subsequent model reading and processing. For example, the connection relationships in the wiring topology are converted into standardized codes, enabling graph neural networks to directly identify the relationships between nodes.

[0073] Generate standard multi-source data files, integrate all processed data in chronological order, store them in a structured format (table or time-series database), and attach data processing logs to record key parameters such as filter window size and number of anomaly removals to ensure data traceability and provide reliable basic data for subsequent model training and result verification.

[0074] After each new window is formed, the average value of the data within that window is recalculated, and this average value is used as the filtered data corresponding to the center position of that window. This process of moving windows and calculating average values ​​is repeated continuously until the processed target filtered data is obtained.

[0075] Step 102: Based on the GNN graph neural network, the wiring topology is modeled as a graph structure, where nodes represent meter terminals and edges represent wiring relationships. The GCN graph convolution and GRU neural network are combined to process dynamic data and construct a GNN-GRU hybrid model.

[0076] Specifically, in this embodiment, an initial graph structure matrix is ​​constructed, and an adjacency matrix is ​​generated based on the node feature set and the edge connection attributes to quantify the association strength between nodes.

[0077] Spatial feature learning is achieved through multi-layer GCN graph convolution. The first layer graph convolution focuses on local connectivity to extract feature interaction patterns between adjacent terminals. The deep network aggregates global topological information to capture indirect associations between terminals across layers.

[0078] After GCN processing, the features of each node are fused with its own state and spatial association information in the whole topology network to generate a spatial feature vector with a global perspective.

[0079] Specifically:

[0080] Graph structure data preparation;

[0081] The physical attributes of all meter terminals in the wiring topology are analyzed, including basic parameters such as terminal model, interface specification, and rated current. Each terminal is treated as an independent node and assigned a unique identifier.

[0082] Record the actual connection relationship between terminals, and clarify the wire type (live wire, neutral wire, ground wire), wire diameter specification and connection stability (historical loosening records, contact resistance fluctuation range) of each pair of connected terminals. Convert these connection relationships into edges between nodes and label the corresponding connection attributes for each edge.

[0083] Acquire standard multi-source data and supplement each node with real-time monitoring features, including the sliding window mean of voltage / current / power, fluctuation amplitude, etc., to form a node feature set containing static physical attributes and dynamic monitoring data.

[0084] GCN spatial feature extraction;

[0085] An initial graph structure matrix is ​​constructed, and an adjacency matrix is ​​generated based on the node feature set and the edge connection attributes to quantify the association strength between nodes.

[0086] Spatial feature learning is achieved through multi-layer GCN graph convolution: the first layer of graph convolution focuses on local connectivity and extracts the feature interaction patterns between adjacent terminals (the impact of voltage changes at a certain terminal on directly connected terminals); the deep network aggregates global topological information and captures the indirect correlations between terminals across layers (the cooperative change patterns of non-directly connected terminals in the same circuit).

[0087] After GCN processing, each node's features will be integrated with its own state and spatial correlation information in the entire topology network to form a spatial feature vector with a global perspective.

[0088] GRU dynamic feature capture;

[0089] Standard multi-source data is divided into time series to form dynamic data segments with fixed time intervals (100ms). Each segment corresponds to the full topology structure at the same time point.

[0090] The spatial feature vectors output by GCN are arranged in chronological order to form a temporally ordered spatial feature sequence, which is then used as the input to the GRU network.

[0091] The GRU network's gating mechanism captures the dynamic changes in feature sequences: the update gate filters key time-series information (feature changes at the moment of sudden changes in electricity load), and the reset gate forgets irrelevant interference (short-term fluctuations of transient noise residue), ultimately outputting a dynamic feature vector containing the evolutionary patterns of the time dimension.

[0092] Hybrid model integration;

[0093] The design of the feature fusion module aligns the spatial feature vector output by GCN with the dynamic feature vector output by GRU in terms of dimensions, and integrates the two types of features by weighted concatenation (the weights are dynamically adjusted according to the importance of spatial correlation strength and time change rate).

[0094] A model output layer is built, and the fused features are nonlinearly transformed to generate comprehensive features that can simultaneously reflect the spatial distribution of the topological structure and the dynamic change trend of the data, providing a foundation for subsequent error analysis.

[0095] Construct a complete GNN-GRU hybrid model framework, clarify the connection method between GCN and GRU (GCN output as GRU input sequence), the number of network layers and the feature dimension transformation rules, and ensure that spatial features and dynamic features are organically coordinated in the model.

[0096] Step 103: Optimize the hyperparameters of the GNN-GRU hybrid model using the improved IWOA whale optimization algorithm to obtain the target GNN-GRU hybrid model;

[0097] Specifically, in this embodiment, the whale population is initialized, with each individual corresponding to a set of hyperparameters, and the initial population is generated from a set search range through random sampling;

[0098] The model's performance on the validation set is used as the fitness function to evaluate the model performance for each set of hyperparameters. The position of each individual is updated according to the improved IWOA search rule based on the fitness value, simulating whale encirclement, bubble net attack and random search behavior.

[0099] When the change in the optimal fitness value is less than a set threshold in multiple consecutive iterations, the optimal parameter combination is output to obtain the target GNN-GRU hybrid model.

[0100] Specifically:

[0101] Improve the core design of IWOA;

[0102] An adaptive search strategy is introduced to dynamically adjust the step size of individual whales based on the iteration progress during the optimization process. In the early stages of optimization, a larger step size is used to expand the search range and avoid getting trapped in local optima; as the iteration progresses, the step size is gradually reduced to improve the search accuracy for the area surrounding the optimal solution.

[0103] A population diversity maintenance mechanism is integrated, which periodically calculates the similarity of individuals in the population. When the similarity exceeds a set threshold, some individuals are randomly perturbed to increase the diversity of the population and prevent the algorithm from converging too early.

[0104] The design employs an elite retention strategy, in which the top 5% of the best-performing individuals are retained in each iteration and directly enter the next iteration, ensuring that high-quality solutions are not lost and accelerating the convergence speed.

[0105] Determining the search range for hyperparameters;

[0106] Identify the categories of hyperparameters to be optimized in the GNN-GRU hybrid model, including the number of network layers, number of neurons per layer, and activation function type of GCN, the hidden layer dimension and time step of GRU, as well as the model's learning rate and regularization coefficient.

[0107] Based on the model characteristics and practical application scenarios, set a reasonable search range for each set of hyperparameters. For example, the number of network layers in GCN can be set to 2-5, the learning rate range can be set to 0.001-0.01, and the regularization coefficient range can be set to 0.0001-0.01, etc.

[0108] Further refine the range of some hyperparameters, such as limiting the activation function to common types like ReLU, Sigmoid, and Tanh, to ensure that the search direction focuses on effective candidate values.

[0109] Implementation of the optimization process;

[0110] The whale population is initialized, with each individual corresponding to a set of hyperparameters. The initial population is generated from a set search range through random sampling to ensure the uniformity of population distribution.

[0111] The model's performance on the validation set (the inverse of the error exponent) is used as the fitness function to evaluate the model performance for each set of hyperparameters. For each individual, the position is updated according to the improved IWOA search rules based on its fitness value, simulating whale encirclement, bubble net attacks, and random search behavior.

[0112] Set an iteration termination condition: when the change in the optimal fitness value is less than a set threshold in multiple consecutive iterations (10 rounds), or when the maximum number of iterations (200 times) is reached, the optimization process stops.

[0113] Generation of the target model;

[0114] Extract the optimal hyperparameters obtained during the optimization process, including key parameters such as the number of layers and neurons in GCN, the hidden layer dimension of GRU, and the learning rate.

[0115] The GNN-GRU hybrid model is reconstructed based on the optimal hyperparameters, and the model is retrained using the training set to ensure that the model fully learns the data features under the optimal parameter configuration.

[0116] The retrained model is validated by testing its error index on the validation set. If the preset performance index is met (error index is lower than the set threshold), the model is identified as the target GNN-GRU hybrid model. If not, the hyperparameter search range or optimization strategy is adjusted, and the optimization process is re-executed.

[0117] Step 104: Input standard multi-source data into the target GNN-GRU hybrid model for training, and output the wiring calibrator error index;

[0118] Specifically, in this embodiment, standard multi-source data is input into the target GNN-GRU hybrid model, and the corresponding wiring calibrator error index is output. The error index must include at least the overall system error mean, the individual error value of each meter terminal, and the error distribution range under different wiring relationships.

[0119] Specifically:

[0120] The stages of model training;

[0121] Initial training phase: A large learning rate is used to allow the model to quickly learn the basic patterns in the data. This phase involves a relatively small number of iterations (20% of the total iterations), with the main goal of rapidly reducing the model's loss function and avoiding falling into an initial inefficient state.

[0122] Fine-tuning phase: Reduce the learning rate to 1 / 5-1 / 10 of the initial value, extend the iteration cycle, and allow the model to delve deeper into the complex relationships in the data, such as the implicit relationship between changes in wiring topology and meter errors. During this phase, after a certain number of iterations (50 times), the model performance is evaluated using a validation set, and the current optimal state is recorded.

[0123] Convergence and Stabilization Phase: This phase begins when the validation set loss changes less than a set threshold over multiple consecutive iterations (10 iterations). The learning rate is kept stable, and training continues for a small number of iterations (20 iterations) to ensure the model stabilizes near its optimal state and avoids performance degradation due to accidental fluctuations.

[0124] Output specifications for the error index;

[0125] After training, the entire validation set is input into the model, and the corresponding wiring calibrator error index is output. The error index needs to include multi-dimensional information, such as the overall system error mean, the individual error values ​​of each meter terminal, and the error distribution range under different wiring relationships, so as to comprehensively reflect the error characteristics of the calibrator.

[0126] Statistical analysis is performed on the output error index to calculate the maximum, minimum, median, and 95% confidence interval of the error, generating a standardized error report. This report needs to be compared and verified with actual calibration data to ensure that the error index accurately reflects the actual performance of the wiring calibrator.

[0127] Save the error index data output by the model and the corresponding input features as training samples for the SVR error correction model in step 5. At the same time, record the key parameters (number of iterations, optimal loss value) during the model training process to provide a reference for subsequent model optimization or version iteration.

[0128] Step 105: Construct a nonlinear error correction model based on SVR support vector regression, input the wiring calibrator error index, and output the correction coefficient.

[0129] Specifically, in this embodiment, the error index of the wiring calibrator is input into the nonlinear error correction model constructed based on SVR support vector regression to output the correction coefficient;

[0130] Establish a mapping relationship between correction coefficients and actual calibration operations, and convert the generated correction coefficients into executable calibration instructions, including at least adjusting the measurement threshold and compensation coefficient of the calibrator.

[0131] Specifically:

[0132] Preliminary preparations for SVR model building;

[0133] The input and output variables of the model are determined. The error index of the wiring calibrator is used as the input feature of the SVR model, including the overall system error mean, the individual error value of each meter terminal, the error distribution range and other multi-dimensional indicators. The true correction coefficient obtained by precision instruments during the actual calibration process is used as the output target to ensure the correlation between input and output.

[0134] Divide the SVR model into training and test sets to avoid the impact of data distribution bias on the model's generalization ability. The training set is used for model parameter learning, and the test set is used to evaluate the model's prediction accuracy. When dividing the data, it is necessary to ensure that the two sets of data cover the error characteristics under different operating conditions (high load, low load, complex topology, etc.).

[0135] Choose a suitable kernel function type, taking into account the nonlinear characteristics of the error data. The radial basis function (RBF) should be given priority because it can effectively handle high-dimensional and nonlinear data relationships. At the same time, set the initial range of the kernel function parameters to provide a basis for subsequent model optimization.

[0136] Generation and verification of correction coefficients;

[0137] The error index data is input into the trained SVR model to generate corresponding correction coefficients. These correction coefficients need to be output categorized by dimensions such as meter terminals and wiring relationships, clearly specifying the correction value for each object to facilitate subsequent calibration operations.

[0138] The effectiveness of the correction coefficients is verified using a test set. The correction coefficients predicted by the model are applied to the actual wiring calibration process, and the error values ​​before and after correction are compared. If the error after correction is significantly reduced (the decrease exceeds the preset threshold of 80%), the model performance is considered satisfactory. If it is not satisfactory, the parameters of the SVR model need to be adjusted or the dataset needs to be re-split for training.

[0139] The correction coefficients are validated for reasonableness. Based on the physical principles of wiring calibration, outliers that clearly do not conform to reality (correction coefficients exceeding the theoretically reasonable range) are eliminated. For error samples corresponding to outliers, their characteristics are re-analyzed, and training data for such samples is supplemented if necessary to improve the model's ability to handle special cases.

[0140] Application adaptation of correction coefficients;

[0141] Establish a mapping relationship between correction coefficients and actual calibration operations, and convert the generated correction coefficients into executable calibration instructions, such as adjusting the measurement threshold and compensation coefficient of the calibrator, to ensure that the theoretical correction values ​​can be directly applied to the hardware devices.

[0142] The design incorporates a dynamic update mechanism. When the wiring topology changes (adding meter terminals, altering wiring relationships) or the error characteristics drift after long-term use, the SVR model retraining process is automatically triggered. New error index data is input, and updated correction coefficients are generated to ensure the continuity of calibration accuracy.

[0143] Record the application effect of each correction coefficient, including the error comparison before and after correction, calibration time, equipment operating status and other information, to form a calibration log, which provides data support for subsequent model optimization and equipment maintenance.

[0144] Its beneficial effects are as follows: 1. Compared with traditional calibration processes, it requires less time, improves efficiency, and supports batch calibration, handling the calibration and correction tasks of multiple meters simultaneously through a wireless communication module. 2. By designing a dynamic update mechanism, when the wiring topology changes or error characteristics drift, the model can be automatically retrained to generate correction coefficients adapted to the new scenario, enhancing adaptability to dynamically changing environments and reducing manual intervention costs. 3. It directly converts data processing results into executable calibration instructions and records application effects through calibration logs, achieving closed-loop management from data to operation. This not only improves the automation level of wiring calibration but also provides data support for subsequent model optimization and equipment maintenance, helping to continuously improve the overall efficiency and reliability of power system wiring calibration.

[0145] Please see Figure 2 In the AI-based wiring calibration instrument output data processing method, the wiring topology is modeled as a graph structure based on the GNN graph neural network, where nodes represent meter terminals and edges represent wiring relationships. The GCN graph convolution and GRU neural network are combined to process dynamic data, and the construction of the GNN-GRU hybrid model includes the following steps:

[0146] Step 201: Construct an initial graph structure matrix. Based on the node feature set, generate an adjacency matrix by combining the edge connection attributes to quantify the association strength between nodes.

[0147] Step 202: Spatial feature learning is performed through multi-layer GCN graph convolution. The first layer graph convolution focuses on local connectivity to extract feature interaction patterns between adjacent terminals. The deep network aggregates global topology information to capture indirect associations between terminals across layers.

[0148] Step 203: After GCN processing, each node's features are fused with its own state and spatial association information in the whole topology network to generate a spatial feature vector with a global perspective.

[0149] The above describes embodiments of the artificial intelligence-based wiring calibration instrument output data processing method of the present invention. Please refer to [link / reference]. Figure 3 The AI-based wiring calibration instrument output data processing system includes the following modules:

[0150] The multi-source data acquisition module is used to acquire voltage, current and power data, and record wiring topology information. After eliminating noise interference using a sliding window filtering algorithm, it removes abnormal data points using an isolated forest algorithm to obtain standard multi-source data.

[0151] The hybrid model building module is used to model the wiring topology as a graph structure based on the GNN graph neural network. Nodes represent meter terminals and edges represent wiring relationships. It combines GCN graph convolution and GRU neural network to process dynamic data and build a GNN-GRU hybrid model.

[0152] The hybrid model optimization module is used to optimize the hyperparameters of the GNN-GRU hybrid model using the improved IWOA whale optimization algorithm to obtain the target GNN-GRU hybrid model.

[0153] The error index output module is used to input standard multi-source data into the target GNN-GRU hybrid model for training and output the error index of the wiring calibrator.

[0154] The correction coefficient calculation module is used to construct a nonlinear error correction model based on SVR support vector regression. It takes the error index of the wiring calibrator as input and outputs the correction coefficient.

[0155] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for processing output data of a wiring calibrator based on artificial intelligence, characterized in that, The data processing method for the output of the wiring calibrator includes the following steps: Voltage, current and power data are collected and wiring topology information is recorded. After noise interference is eliminated by using a sliding window filtering algorithm, abnormal data points are removed by using an isolated forest algorithm to obtain standard multi-source data. An initial graph structure matrix is ​​constructed, and an adjacency matrix is ​​generated based on the node feature set and the edge connectivity attributes to quantify the association strength between nodes. Spatial feature learning is performed through multi-layer GCN graph convolution. The first layer graph convolution focuses on local connectivity to extract feature interaction patterns between adjacent nodes. The deep network aggregates global topology information to capture indirect associations between nodes across layers. The features of each node after GCN processing are fused with its own state and spatial association information in the whole topology network to generate a spatial feature vector with a global perspective. The GCN graph convolution and GRU neural network are combined to process dynamic data and construct a GNN-GRU hybrid model. The hyperparameters of the GNN-GRU hybrid model are optimized using the improved IWOA whale optimization algorithm to obtain the target GNN-GRU hybrid model. The standard multi-source data is input into the target GNN-GRU hybrid model for training, and the error index of the wiring calibrator is output. A nonlinear error correction model is constructed based on SVR support vector regression, with the error index of the wiring calibrator as input and the correction coefficient as output.

2. The data processing method for the output of the wiring calibrator based on artificial intelligence as described in claim 1, characterized in that, The collected voltage, current, and power data, along with the recorded wiring topology information, are processed using a sliding window filtering algorithm to eliminate noise interference, followed by an isolated forest algorithm to remove outlier data points, resulting in standard multi-source data, including: The calibrator collects voltage, current, and power data using its built-in sensors and microprocessor, and records wiring topology information. Set the window size of the sliding window filtering algorithm to 10, and arrange the collected voltage, current and power data in chronological order; Starting with the first data point, take 10 consecutive data points as the first window, calculate the average value of the data within the window, and use the average value as the filtered data at the center of the window. Repeat this calculation to obtain the processed target filtered data.

3. The data processing method for the output of the wiring calibrator based on artificial intelligence as described in claim 1, characterized in that, The process of collecting voltage, current, and power data, recording wiring topology information, eliminating noise interference using a sliding window filtering algorithm, and removing outlier data points using an isolated forest algorithm to obtain standard multi-source data also includes: The number of trees in the isolated forest algorithm is set to 100, the sample subset size is 256, the processed target filtered data is input into the isolated forest algorithm, the distribution of abnormal scores of normal data is calculated, and the 95th percentile of abnormal scores is used as the threshold. When the outlier score of the data exceeds the threshold, it is identified as an outlier data point; outlier data points are removed to obtain standard multi-source data.

4. The data processing method for the output of the wiring calibrator based on artificial intelligence as described in claim 1, characterized in that, The process of optimizing the hyperparameters of the GNN-GRU hybrid model using the improved IWOA whale optimization algorithm to obtain the target GNN-GRU hybrid model includes: Initialize the whale population, with each individual corresponding to a set of hyperparameters, and generate the initial population from a set search range through random sampling; The model's performance on the validation set is used as the fitness function to evaluate the model performance for each set of hyperparameters. The position of each individual is updated according to the improved IWOA search rule based on the fitness value, simulating whale encirclement, bubble net attack and random search behavior. When the change in the optimal fitness value is less than a set threshold in multiple consecutive iterations, the optimal parameter combination is output to obtain the target GNN-GRU hybrid model.

5. The data processing method for the output of the wiring calibrator based on artificial intelligence as described in claim 1, characterized in that, The step of inputting the standard multi-source data into the target GNN-GRU hybrid model for training and outputting the wiring calibrator error index includes: The standard multi-source data is input into the target GNN-GRU hybrid model, and the corresponding wiring calibrator error index is output. The error index must include at least the overall system error mean, the individual error value of each meter terminal, and the error distribution range under different wiring relationships.

6. The data processing method for the output of the wiring calibrator based on artificial intelligence as described in claim 1, characterized in that, The nonlinear error correction model constructed based on SVR support vector regression takes the error index of the wiring calibrator as input and outputs the correction coefficient, including: Input the error index of the wiring calibrator into the nonlinear error correction model constructed based on SVR support vector regression to output the correction coefficient; Establish a mapping relationship between correction coefficients and actual calibration operations, and convert the generated correction coefficients into executable calibration instructions, including at least adjusting the measurement threshold and compensation coefficient of the calibrator.

7. A data processing system for the output of a wiring calibrator based on artificial intelligence, characterized in that: The data processing system for the output of the wiring calibrator includes the following modules: The multi-source data acquisition module is used to acquire voltage, current and power data, and record wiring topology information. After eliminating noise interference using a sliding window filtering algorithm, it removes abnormal data points using an isolated forest algorithm to obtain standard multi-source data. The hybrid model building module is used to construct an initial graph structure matrix. Based on the node feature set, it generates an adjacency matrix by combining the edge connectivity attributes to quantify the association strength between nodes. Spatial feature learning is performed through multi-layer GCN graph convolution. The first-layer graph convolution focuses on local connectivity to extract feature interaction patterns between adjacent nodes. The deep network aggregates global topological information to capture indirect associations between nodes across layers. The features of each node after GCN processing are fused with its own state and spatial association information in the whole topology network to generate a spatial feature vector with a global perspective. The GCN graph convolution and GRU neural network are combined to process dynamic data and construct a GNN-GRU hybrid model. The hybrid model optimization module is used to optimize the hyperparameters of the GNN-GRU hybrid model using the improved IWOA whale optimization algorithm to obtain the target GNN-GRU hybrid model. The error index output module is used to input the standard multi-source data into the target GNN-GRU hybrid model for training and output the wiring calibrator error index. The correction coefficient calculation module is used to construct a nonlinear error correction model based on SVR support vector regression. It takes the error index of the wiring calibrator as input and outputs the correction coefficient.

8. The AI-based wiring calibration instrument output data processing system as described in claim 7, characterized in that, The hybrid model optimization module includes the following sub-modules: The generation submodule is used to initialize the whale population. Each individual corresponds to a set of hyperparameters, and the initial population is generated from a set search range by random sampling. The evaluation submodule is used to evaluate the model's performance on the validation set as a fitness function, assess the model performance for each set of hyperparameters, update the position of each individual according to the improved IWOA search rules based on the fitness value, and simulate whale encirclement, bubble net attack and random search behavior. The resulting submodule is used to output the optimal parameter combination when the change in the optimal fitness value is less than a set threshold in multiple consecutive iterations, thus obtaining the target GNN-GRU hybrid model.

9. The AI-based wiring calibration instrument output data processing system as described in claim 7, characterized in that, The correction coefficient calculation module includes the following sub-modules: The output submodule is used to input the error index of the wiring calibrator into the nonlinear error correction model built based on SVR support vector regression to output the correction coefficient. The mapping submodule is used to establish the mapping relationship between the correction coefficients and the actual calibration operations, and to convert the generated correction coefficients into executable calibration instructions, including at least adjusting the measurement threshold and compensation coefficient of the calibrator.

Images