Power metering misalignment evaluation method and system considering multi-source measurement data fusion

Abstract

The disclosure provides a power metering misalignment evaluation method and system considering multi-source measurement data fusion, relating to the technical field of power metering evaluation, comprising: acquiring multi-source power metering data, which is preprocessed as an initial sample set; extracting sample feature data based on the initial sample set, and constructing a sample feature distribution representation learning model based on the sample feature data; setting a grouping similarity threshold, grouping the sample feature data, and calculating the grouped class mean sample feature data; constructing a complementary distance measurement function; obtaining normal sample feature data based on the sample feature distribution representation learning model; inputting the normal sample feature data into a region-oriented mean grouping class feature extractor to obtain sample features; calculating the complementary distance measurement value based on the complementary distance measurement function, inputting the softmax function to output the recognized sample class; comparing the recognized sample class with the actual user sample class to determine whether the power metering data is misaligned.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of electric power metering evaluation, in particular to an electric energy metering misalignment evaluation method and system considering multi-source measurement data fusion. BACKGROUND

[0002] The statements in this section merely provide background information related to the present disclosure and do not necessarily constitute the prior art.

[0003] With the construction of new power systems, high proportion of new energy access and large-scale application of power electronic equipment, higher requirements are put forward for the rapid response, transmission error, accurate calibration, misalignment judgment and reliable collection of measurement devices. The measurement accuracy of the power system is directly related to the fairness of the power market transaction. Especially under the intermittent and volatile output of new energy and the high-speed switching of power electronic equipment, persistent fundamental dynamic, wide range variation, harmonic reactive power and wide frequency domain complex information are generated. The current metering device is insufficient in response and anti-interference ability, which leads to the deterioration of the quality of massive multi-source electric power measurement data, and further leads to the difficulty in carrying out the upper data fusion application work.

[0004] For example, the existing distributed photovoltaic grid-connected peak shaving control strategy based on massive new energy grid-connected measurement data is proposed to solve the problem of power peak and valley caused by load change of distribution network. The segmented operation mode is proposed to ensure the stable operation of the power grid. However, the implementation of this strategy is based on the accuracy of electric energy metering data, and the problem of misalignment data processing of electric energy metering data is not involved in the paper. At the same time, in the traditional supervised learning-based abnormal data identification model, a large amount of labeled data is usually needed for training, which is not only time-consuming and laborious, but also high in cost. Moreover, for new energy grid connection, it also leads to the difficulty in identifying the misalignment of electric energy meter, the dynamic nature of the operation error calculation model and the decline of the misalignment identification accuracy. Traditional distance measurement methods also have their own limitations, and there are still problems such as not fully considering the multi-class imbalance, metering abnormality and high cost of labeled samples of multi-source measurement data. SUMMARY

[0005] In order to solve the above problems, the present disclosure proposes an electric energy metering misalignment evaluation method and system considering multi-source measurement data fusion, which adopts a comparison learning combined with small sample learning method to compare the similarity and difference between samples, constructs a sample feature distribution representation learning model and a mean grouping class feature extractor based on sufficient sample data, and then adjusts the parameters for the region, thereby improving the regional applicability, learning the intrinsic features of the data without explicit labels, extracting useful information from a large amount of unlabeled data, and reducing the dependence on labeled data.

[0006] According to some embodiments, the present disclosure adopts the following technical solutions:

[0007] The power metering misalignment evaluation method considering multi-source measurement data fusion comprises:

[0008] Multi-source power metering data is acquired and preprocessed as an initial sample set;

[0009] Sample feature data is extracted based on the initial sample set, and a sample feature distribution representation learning model is constructed based on the sample feature data;

[0010] The sample feature data is grouped based on a set grouping similarity threshold, and grouped class mean sample feature data is calculated; a complementary distance measurement function is constructed based on the grouped class mean sample feature data and the sample feature data;

[0011] According to a small amount of normal sample data of users in different regions, sample features are obtained based on the sample feature distribution representation learning model;

[0012] The complementary distance measurement value of the sample feature and the grouped class mean sample feature data is calculated based on the complementary distance measurement function, and the recognized sample class is obtained; the recognized sample class is compared with the actual user sample class, and whether the power metering data is misaligned is obtained.

[0013] According to some embodiments, the present disclosure adopts the following technical solutions:

[0014] The power metering misalignment evaluation system considering multi-source measurement data fusion comprises:

[0015] A model construction module is configured to acquire multi-source power metering data, pre-process the data as an initial sample set, extract sample feature data based on the initial sample set, and construct a sample feature distribution representation learning model based on the sample feature data;

[0016] A class grouping module is configured to group the sample feature data based on a set grouping similarity threshold, and calculate and obtain grouped class mean sample feature data;

[0017] A measurement function construction module is configured to construct a complementary distance measurement function based on the grouped class mean sample feature data and the sample feature data;

[0018] An evaluation module is configured to obtain sample features based on the sample feature distribution representation learning model according to a small amount of normal sample data of users in different regions, calculate the complementary distance measurement value of the sample feature and the grouped class mean sample feature data based on the complementary distance measurement function, obtain the recognized sample class, and compare the recognized sample class with the actual user sample class to obtain whether the power metering data is misaligned.

[0019] According to some embodiments, the present disclosure adopts the following technical solutions:

[0020] A computer program product comprising a computer program which, when executed by a processor, implements the power metering misalignment evaluation method considering multi-source measurement data fusion.

[0021] According to some embodiments, the present disclosure adopts the technical solutions as follows:

[0022] A non-transitory computer-readable storage medium for storing computer instructions, which, when executed by a processor, implement the power metering misalignment evaluation method considering multi-source measurement data fusion.

[0023] According to some embodiments, the present disclosure adopts the technical solutions as follows:

[0024] An electronic device comprising a processor, a memory, and a computer program; wherein the processor is connected with the memory, and the computer program is stored in the memory; when the electronic device is running, the processor executes the computer program stored in the memory, so that the electronic device implements the power metering misalignment evaluation method considering multi-source measurement data fusion.

[0025] Compared with the prior art, the present disclosure has the beneficial effects that:

[0026] The power metering misalignment evaluation method considering multi-source measurement data fusion of the present disclosure can efficiently utilize data, especially unlabeled data, by adopting contrast learning to solve the problems existing in traditional supervised learning. Contrast learning can learn the intrinsic features of data without explicit labels by comparing the similarities and differences between samples, can extract useful information from a large amount of unlabeled data, and reduce the dependence on labeled data, thereby significantly improving the utilization rate of data.

[0027] The power metering misalignment evaluation method considering multi-source measurement data fusion of the present disclosure adopts small sample learning to evaluate the power metering misalignment, borrows the ideas of contrast learning and meta-learning, and based on sufficient sample data, first constructs a sample feature distribution representation learning model and a mean grouping category feature extractor, and then performs parameter fine-tuning for the region, thereby improving the regional applicability, and then based on the small amount of sample data provided by the region, maps the sample to a feature space, measures the sample and the feature center in the space, and according to the accuracy of sample classification, identifies whether the metering is misaligned, especially in processing time-varying small sample data, which shows high feasibility and effectiveness, and solves the problems of difficult identification of power meter misalignment caused by new energy grid connection, dynamic operation error calculation model, and decreased misalignment identification accuracy.

[0028] The power metering misalignment evaluation method considering multi-source measurement data fusion of the present disclosure adopts a complementary distance measurement method, simultaneously considers the influence of cosine similarity and Euclidean distance on feature measurement, simultaneously measures features from two aspects of angle and geometric spatial distance, and jointly expresses the local feature measurement distance through a measurement function as the final measurement result of sample data, realizes accurate classification of sample data, and solves the respective limitations of traditional distance measurement methods. BRIEF DESCRIPTION OF DRAWINGS

[0029] The accompanying drawings, which form a part of this disclosure, are intended to provide further understanding of the present disclosure and are incorporated herein for illustrative purposes. The schematic embodiments of the present disclosure and their descriptions are used to explain the present disclosure and do not constitute improper limitations on the present disclosure.

[0030] Figure 1 The flowchart of the misalignment evaluation method of the embodiments of the present disclosure;

[0031] Figure 2 The architecture diagram of the contrast learning method of the embodiments of the present disclosure. DETAILED DESCRIPTION

[0032] The present disclosure will be further described below in combination with the drawings and embodiments.

[0033] It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the present disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as generally understood by those skilled in the art to which the present disclosure belongs.

[0034] It should be noted that the terms used herein are only for the purpose of describing specific embodiments and are not intended to limit the exemplary embodiments according to the present disclosure. As used herein, the singular form is intended to include the plural form unless the context clearly indicates otherwise, and furthermore, it should be understood that when the terms "comprise" and / or "include" are used in the specification, there is a presence of the features, steps, operations, devices, components and / or combinations thereof.

[0035] Embodiment 1

[0036] In an embodiment of the present disclosure, a power metering misalignment evaluation method considering multi-source measurement data fusion is provided. The method fully considers the problems of multi-class imbalance, metering abnormality and high cost of labeled samples existing in multi-source measurement data, adopts a contrast learning combined with small sample learning method, constructs a power metering misalignment evaluation method considering multi-source measurement data fusion, and improves the quality and availability of multi-source measurement data through model parameter adaptation, feature extraction, similarity measurement, abnormal value detection and elimination, provides high-quality data support for provincial measurement data fusion application, and the method steps include:

[0037] Step 1: Acquire multi-source power metering data, preprocess it, and use it as the initial sample set;

[0038] Step 2: Extract sample feature data based on the initial sample set. Based on the sample feature data, calculate the distance metric value of the sample feature data using a similarity function, obtain the contrastive learning loss function, and thus construct a sample feature distribution representation learning model.

[0039] Step 3: Set a group similarity threshold, group the sample feature data, and calculate the average value of the sample features as the group category mean sample feature data;

[0040] Step 4: Construct a complementary distance metric function based on the group category mean sample feature data and sample feature data;

[0041] Step 5: Obtain a small amount of normal sample data from users in different regions and input it into the sample feature distribution representation learning model to obtain normal sample feature data; input the normal sample feature data into the region-oriented mean grouping category feature extractor to obtain sample features;

[0042] Step 6: Based on the complementary distance metric function, calculate the complementary distance metric value between the sample features and the group category mean sample features, input it into the softmax function and output the identified sample category; compare the identified sample category with the actual user sample category to determine whether the electricity metering data is inaccurate.

[0043] As one embodiment, this disclosure uses an electricity consumption information collection system and a marketing business application system as data processing platforms, and uses distributed photovoltaic power metering data as a modeling and implementation example. The specific implementation process is as follows:

[0044] Step 1: Acquire multi-source electricity metering data based on the new generation electricity information collection system. This multi-source electricity metering data consists of multi-source distributed photovoltaic (PV) electricity metering data. After aggregation and organization, preprocessing is performed. Preprocessing first involves data cleaning, specifically deleting duplicate and missing data. Then, data integration is performed, merging multiple data sources into one data store and deleting redundant data. Next, data normalization is performed to eliminate dimensional differences, completing the construction of the initial sample set. The categories of the multi-source distributed PV electricity metering data include: distributed PV operation mode, PV business user type, electricity metering, and environmental factors. See Table 1 below for details.

[0045] Table 1. Categories of Distributed Photovoltaic Multi-Source Power Metering Data

[0046]

[0047] Step 2: Extract sample feature data based on the initial sample set. Based on the sample feature data, calculate the distance metric value of the sample feature data using a similarity function, obtain the contrastive learning loss function, and thus construct a sample feature distribution representation learning model.

[0048] Specifically, based on the initial sample set, an encoder combined with an attention mechanism is used to extract sample feature data, which is then input into an initial sample feature distribution representation learning model constructed based on contrastive learning. This process brings the features of samples of the same category closer together and widens the features of samples of different categories, thereby obtaining the spatial distribution of electricity metering sample feature data. The specific process is as follows:

[0049] Step 21: Based on the initial sample set, extract sample feature data using an encoder combined with an attention mechanism;

[0050] The distributed photovoltaic operation mode, photovoltaic business user type, and environmental factors of the initial sample set are used as input sample data for the attention mechanism. The electricity metering data of the initial sample set are used as input sample data for the encoder. After the encoder combines with the attention mechanism, it outputs encoded data. The encoded data is input into the generator. Combined with the attention mechanism's influence parameters of the same initial sample set, the distributed photovoltaic operation mode, photovoltaic business user type, and environmental factors, the generator outputs reproduced sample data as sample feature data. The loss function is used for feedback parameter tuning until the error between the input sample data and the reproduced sample data is less than a set threshold. After training, the encoder combined with the attention mechanism is output as a sample feature data extraction model.

[0051] Step 22: Based on the initial sample set, input the sample feature data into the initial sample feature distribution representation learning model built based on contrastive learning, calculate the feature distance metric value using the similarity function, and obtain the contrastive learning loss function;

[0052] Among them, the sample feature distribution representation learning model structure built based on contrastive learning is as follows: Figure 2 As shown, it contains two data processing branches, upper and lower. Both branches include an encoder combined with an attention mechanism (Attention-Encoder) module, a feature data module, a similarity function calculation module, and a loss function calculation module. The upper and lower branches are symmetrical and can share parameters.

[0053] The encoder combined with the attention-encoder module is the feature data extraction model constructed in step 21.

[0054] The process involves inputting sample data into an initial sample feature distribution representation learning model built based on contrastive learning. The specific data processing includes two types: First, sample data from the same user at different times are input into the upper and lower branches respectively to obtain sample feature data. Through similarity function calculation and loss function feedback, the distance between the extracted feature data of the sample set of the same user in the projection space is reduced. Second, sample data from different users are input into the upper and lower branches respectively to obtain sample feature data. Through similarity function calculation and loss function feedback, the distance between the extracted feature data of the sample sets of different users in the projection space is increased.

[0055] Specifically, the data from two electricity metering samples of the same user are represented as follows: and The two different user electricity metering sample data are represented as follows: and The sample feature data extraction model is used to encode the electricity metering sample data into vectors:

[0056]

[0057]

[0058]

[0059] in, This represents a sample feature data extraction model built by combining an encoder with an attention mechanism. , , This is the encoded sample feature vector.

[0060] Calculating electricity metering values ​​using InfoNCE losses Eigenvectors and electricity metering values Comparative learning loss function between feature vectors :

[0061]

[0062] Among them, the dot product method is used to... and Calculate the similarity between them. The temperature hyperparameter represents the sample difficulty identification penalty coefficient, which increases the distance between positive samples and highly similar negative samples; N represents the number of different input sample groups. For the encoded and Feature vectors of different classes of samples;

[0063] Step 23: Update the encoder parameters and attention mechanism parameters by backpropagation based on the contrastive learning loss function until the loss function is less than the set threshold. Then fix the model parameters and complete the construction of the sample feature distribution representation learning model.

[0064] Step 3: Set a group similarity threshold, group the sample feature data, and calculate the average value of the sample features as the group category mean sample feature data;

[0065] Given that the characteristic distribution distance of similar distributed photovoltaic users in the same area is often very small, a group similarity threshold B is set, and users whose similarity is less than this threshold are grouped together, totaling 100 groups. K Groups, calculate the characteristic average as k Mean sample feature data for each group category;

[0066] The mean sample characteristics of the group categories are: :

[0067]

[0068] in, Represents the sample feature vector matrix, containing K One feature; L is the total number of samples within a group. Indicates the first l one sample K The vector of values ​​for each feature; .

[0069] Step 4: Construct a complementary distance metric function based on the group category mean sample feature data and sample feature data;

[0070] Furthermore, based on the sample feature data of group category mean and the sample feature data of electricity metering, a complementary distance metric function is constructed using cosine distance and Euclidean distance to evaluate the similarity of small-sample learning classification. The specific process is as follows:

[0071] Step 41: Based on the feature data of the group category mean samples and the feature data of the electricity metering samples to be classified, calculate the measurement coefficient using the cosine similarity algorithm. ;

[0072]

[0073] in, Represents the sample feature vector; The sample represents the electrical energy metering data to be classified; k represents the mean sample of the kth class group. The sample feature vector represents the electrical energy metering sample q to be classified; This represents the feature vector of sample k, which represents the mean of the k-th group. This represents the cosine similarity calculation function, which obtains... and The cosine similarity value; This represents the threshold activation function, whose value range is (0,1).

[0074] Step 42: Construct a complementary distance metric function by combining the Euclidean distance function with the metric coefficients. ;

[0075]

[0076] in, Used for balance and Euclidean distance:

[0077] Step 43: Based on the sample feature data of the group category mean and the sample feature data of electricity metering, calculate the complementary distance metric value using the complementary distance metric function, input it into the softmax function, and output the group category of the electricity metering sample. y .

[0078]

[0079] in, b The training parameters are initialized to 0; softmax is a multi-class probability function, with each vector's probability ranging from (0,1), and the sum of its probabilities is 1; the group with the highest probability value in the softmax vector is then... y Category;

[0080] If the sample measurement data is normal, it will be assigned to the correct user group category; if the sample measurement data is abnormal, it will be grouped into another group. An incorrect group category indicates that the measurement data is abnormal.

[0081] Step 5: Obtain a small amount of normal sample data from users in different regions and input it into the sample feature distribution representation learning model to obtain normal sample feature data; input the normal sample feature data into the region-oriented mean grouping category feature extractor to obtain sample features;

[0082] The region-oriented mean grouping category feature extractor uses a complementary distance metric function to calculate the distance between a small number of normal sample data of users in different regions and the mean grouping category sample features, and feeds it back to the initial sample feature distribution representation learning model based on contrastive learning constructed in step 22. By feeding back the parameters, the fine-tuning parameters include the parameters of the encoder combined with the attention mechanism (Attention-Encoder) module, the similarity function calculation module, the loss function calculation module, and the softmax function in step 43, thereby realizing the applicability of the model to the region.

[0083] Step 6: Based on the complementary distance metric function, input the sample features and the group category mean sample feature data into the complementary distance metric function, calculate the complementary distance metric value between the sample features and the group category mean sample feature data, input the softmax function to output the identified sample category; compare the identified sample category with the actual user sample category to determine whether the electricity metering data is inaccurate.

[0084] If the identified sample category is the same as the actual user sample category, the electricity metering data is normal and will be used for subsequent load forecasting tasks; if the identified sample category is different from the actual user sample category, the electricity metering data is abnormal and will be removed.

[0085] Simulation Experiment

[0086] To verify the effectiveness of the method proposed in this disclosure for assessing electricity metering inaccuracies, a comparative explanation is provided with existing research methods for identifying abnormal photovoltaic metering inaccuracies. The comparison method is as follows:

[0087] (1) k-means clustering comparison method: select photovoltaic users in the same area and evaluate the inaccuracy of photovoltaic power metering based on the sample distance deviation value;

[0088] (2) SVM: A photovoltaic power metering inaccuracy assessment model constructed using Support Vector Machine (SVM);

[0089] (3) LSTM: A photovoltaic power metering inaccuracy assessment model constructed using the Long Short-Term Memory (LSTM) algorithm;

[0090] (4) The method of the present disclosure embodiment;

[0091] To demonstrate the generalizability of the method described in this disclosure, photovoltaic power metering user data from three different regions were collected, and three sets of data sample sets were constructed as follows:

[0092] Data set 1: Contains 861 samples from region A;

[0093] Data set 2: 59 samples from region B were obtained;

[0094] Data set 3: 902 samples from region C were obtained;

[0095] The accuracy verification results of the method proposed in this disclosure are shown in Table 2.

[0096] Table 2 Comparison of Accuracy Rates in Electricity Metering Inaccuracy Assessment

[0097]

[0098] As shown in Table 2, the proposed method demonstrates superior recognition performance in simulation data sets 1 and 3, effectively tracking dynamic changes in photovoltaic power generation and identifying abnormal electricity metering data. Particularly in data set 2, even with a small sample size, parameter tuning based on the pre-trained model still allows for accurate identification of abnormal metering data.

[0099] As one embodiment, the energy metering inaccuracy assessment method of this disclosure, which considers the fusion of multi-source measurement data, is based on an energy metering inaccuracy data identification model. The training process of this energy metering inaccuracy data identification model is as follows: Figure 1 As shown, specifically:

[0100] 1) Use the electricity consumption information acquisition system to obtain multi-source measurement data, and construct an initial sample set after preprocessing;

[0101] 2) Based on the initial sample set, an encoder combined with an attention mechanism is used to extract sample feature data, which is then input into a sample feature distribution representation learning model built based on contrastive learning. This brings the features of samples of the same category closer together and the features of samples of different categories farther apart, thereby obtaining the spatial distribution of the features of electricity metering samples.

[0102] 3) Set a group similarity threshold and group the data. Collect sample feature data that are less than the group similarity threshold as the same group category and calculate the average feature data as the average sample feature data of the group category.

[0103] 4) Based on the characteristic data of the group category mean samples and the characteristic data of the electricity metering samples, a complementary distance metric function is constructed using cosine distance and Euclidean distance to evaluate the similarity of small sample learning classification;

[0104] 5) Use a small amount of normal sample data from users in different regions as a support sample set, and input it into a sample feature distribution representation learning model built based on contrastive learning, and output the data that exceeds the threshold.

[0105] 6) Based on the data exceeding the threshold, the complementary distance metric function is used to calculate the distance with the features of the mean grouped category samples, and the encoder, contrastive learning and softmax function are fed back to fine-tune the parameters, update the model and output the data category exceeding the threshold again;

[0106] 7) Finally, output the trained model for identifying inaccurate electricity metering data, and obtain the characteristic data of electricity metering to be evaluated for identification and evaluation.

[0107] Example 2

[0108] One embodiment of this disclosure provides a power metering inaccuracy assessment system that considers the fusion of multi-source measurement data, including:

[0109] The model building module is used to acquire multi-source power metering data, preprocess it, and use it as an initial sample set; extract sample feature data based on the initial sample set, and build a sample feature distribution representation learning model based on the sample feature data.

[0110] The category grouping module is used to group sample feature data based on a set grouping similarity threshold and calculate the mean sample feature data of each group category.

[0111] The metric function construction module is used to construct complementary distance metric functions based on the grouped category mean sample feature data and sample feature data;

[0112] The evaluation module obtains sample features based on a sample feature distribution representation learning model using a small amount of normal sample data from users in different regions; it calculates the complementary distance metric between the sample features and the mean sample feature data of the group categories based on the complementary distance metric function to obtain the identified sample category; and it compares the identified sample category with the actual user sample category to obtain the result of whether the electricity metering data is inaccurate.

[0113] Example 3

[0114] One embodiment of this disclosure provides a computer program product, including a computer program that, when executed by a processor, implements the aforementioned method for assessing electrical energy metering inaccuracies considering the fusion of multi-source measurement data.

[0115] Example 4

[0116] One embodiment of this disclosure provides a non-transitory computer-readable storage medium for storing computer instructions. When executed by a processor, the computer instructions implement the power metering inaccuracy assessment method that considers the fusion of multi-source measurement data.

[0117] Example 5

[0118] An electronic device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to cause the electronic device to perform the energy metering inaccuracy assessment method that considers the fusion of multi-source measurement data.

[0119] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0120] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0121] While the specific embodiments of this disclosure have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of this disclosure. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of this disclosure are still within the scope of protection of this disclosure.

Claims

1. A method for assessing the inaccuracy of electricity metering considering the fusion of multi-source measurement data, characterized in that, include: Acquire multi-source power metering data, preprocess it, and use it as the initial sample set; Based on the initial sample set, sample feature data is extracted, and a sample feature distribution representation learning model is constructed based on the sample feature data. The sample feature data are grouped based on a set grouping similarity threshold, and the mean sample feature data of each group category is calculated and obtained. Based on the mean sample feature data of each group category and the sample feature data, a complementary distance metric function is constructed. Based on a small amount of normal sample data from users in different regions, sample features are obtained by learning a model based on sample feature distribution representation. The complementary distance metric between the sample features and the group category mean sample features is calculated based on the complementary distance metric function to obtain the identified sample category. The identified sample category is compared with the actual user sample category to determine whether the electricity metering data is inaccurate. The step of constructing a complementary distance metric function based on the group category mean sample feature data and sample feature data includes: calculating the metric coefficient using cosine similarity based on the group category mean sample feature data and sample feature data. in, Represents the sample feature vector; The sample represents the electrical energy metering data to be classified; k represents the mean sample of the kth class group. The sample feature vector represents the electrical energy metering sample q to be classified; This represents the feature vector of sample k, which represents the mean of the k-th group. This represents the cosine similarity calculation function, which obtains... and The cosine similarity value; This represents the threshold activation function, whose value range is (0,1). A complementary distance metric function is constructed by combining the Euclidean distance function with metric coefficients, including: in, Used for balance and The size of the Euclidean distance; The extraction of sample feature data based on the initial sample set includes: The encoder is combined with the attention mechanism to extract sample feature data. The distributed photovoltaic operation mode, photovoltaic business user type and environmental factors of the initial sample set are used as input sample data for the attention mechanism. The electricity metering data of the initial sample set is used as input sample data for the encoder and outputs coded data. The coded data is input into the generator. Combined with the attention mechanism influence parameters of distributed photovoltaic operation mode, photovoltaic business user type and environmental factors output by the attention mechanism, the reproduced sample data is output as sample feature data.

2. The method for assessing electrical energy metering inaccuracies considering multi-source measurement data fusion as described in claim 1, characterized in that, The acquired multi-source electricity metering data refers to multi-source distributed photovoltaic electricity metering data, and the categories of multi-source distributed photovoltaic electricity metering data include: distributed photovoltaic operation mode, photovoltaic business user type, electricity metering, and environmental factors.

3. The method for assessing electrical energy metering inaccuracies considering the fusion of multi-source measurement data as described in claim 1, characterized in that, Based on sample feature data, a similarity function is used to calculate the distance metric of the sample feature data, and a contrastive learning loss function is obtained to construct a sample feature distribution representation learning model, including: The sample feature data is input into the initial sample feature distribution representation learning model based on contrastive learning. The feature distance metric is calculated using the similarity function, and the contrastive learning loss function is obtained. The encoder parameters and attention mechanism parameters are updated by backpropagation based on the contrastive learning loss function until the loss function is less than a set threshold. Then the model parameters are fixed, and the final sample feature distribution representation learning model is completed.

4. A power metering inaccuracy assessment system considering multi-source measurement data fusion, based on the power metering inaccuracy assessment method considering multi-source measurement data fusion as described in any one of claims 1-3, characterized in that, include: The model building module is used to acquire multi-source power metering data, which is then preprocessed and used as the initial sample set. Based on the initial sample set, sample feature data is extracted, and a sample feature distribution representation learning model is constructed based on the sample feature data. The category grouping module is used to group sample feature data based on a set grouping similarity threshold and calculate the mean sample feature data of each group category. The metric function construction module is used to construct complementary distance metric functions based on the grouped category mean sample feature data and sample feature data; The evaluation module obtains sample features based on a sample feature distribution representation learning model using a small amount of normal sample data from users in different regions; it calculates the complementary distance metric between the sample features and the mean sample feature data of the group categories based on the complementary distance metric function to obtain the identified sample category; and it compares the identified sample category with the actual user sample category to obtain the result of whether the electricity metering data is inaccurate.

5. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the method for assessing electrical energy metering inaccuracies that considers the fusion of multi-source measurement data as described in any one of claims 1-3.

6. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium is used to store computer instructions, which, when executed by a processor, implement the power metering inaccuracy assessment method considering the fusion of multi-source measurement data as described in any one of claims 1-3.

7. An electronic device, characterized in that, include: The device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to perform the energy metering inaccuracy assessment method considering multi-source measurement data fusion as described in any one of claims 1-3.

Images