Electric energy meter error prediction method and system based on multi-scale dense network

Abstract

The invention relates to an electric energy meter error prediction method and system based on a multi-scale dense network, and the method comprises the steps: obtaining the data of a to-be-detected electric energy meter, including manufacturer information, historical errors before a prediction day, and environment data after the prediction day; to-be-detected electric energy meter data is preprocessed and then spliced, and multi-scale features are obtained; inputting the multi-scale features into a pre-trained multi-scale dense network, and performing feature extraction on the multi-scale features by using dense blocks of convolution kernels of different sizes in the multi-scale dense network to obtain multi-scale fusion features; calculating an attention mechanism weight of the multi-scale fusion feature based on an attention mechanism, and calculating a weighted feature based on the attention mechanism weight and the multi-scale fusion feature; and inputting the weighted features into a long-short-term memory network, and outputting an acquired error prediction value of a prediction day of the to-be-detected electric energy meter.

Description

Technical Field

[0001] This invention relates to the field of electricity meter prediction technology, and mainly to an electricity meter error prediction method and system based on multi-scale dense networks. Background Technology

[0002] As a core terminal device for power grid metering, electricity billing, and dispatch optimization, the accuracy of smart meters directly affects the economic interests of power companies, the fair electricity experience for users, and the safe and stable operation of the power grid. In practical applications, the error of smart meters is affected by a combination of factors: at the environmental level, long-term exposure to complex conditions such as extreme temperature and humidity and electromagnetic interference can lead to performance drift of internal components; at the product level, differences in manufacturing processes, core component selection, and factory calibration standards among different manufacturers result in inherently different error baseline characteristics; at the operational level, long-term load fluctuations and equipment aging and wear further exacerbate the dynamic changes in error; if the trend of meter error cannot be accurately predicted, it may lead to deviations in electricity billing, errors in power grid dispatching decisions, and even the escalation of equipment failures.

[0003] However, existing error prediction methods mainly rely on traditional statistical models or simple deep learning models, which are difficult to fully capture the complex relationships between multi-source features and the dynamic characteristics of time series. Traditional statistical models, such as autoregressive integral moving average models, linear regression, and support vector machines, establish linear or simple nonlinear mapping relationships based on the statistical laws of historical error data, and achieve prediction by fitting statistical features such as the mean and variance of the data. However, meter errors are the result of the combined effects of multiple factors such as manufacturer processes, environmental temperature and humidity, and historical operating conditions. There are strong nonlinear coupling relationships between these factors, and traditional statistical models cannot capture such complex correlations, resulting in large prediction biases. On the other hand, simple deep learning models can only extract features at a fixed scale, cannot identify global patterns, lack dense connection structures, have low feature reuse efficiency, and are difficult to capture error correlation information across time scales simultaneously, easily missing key features. Summary of the Invention

[0004] To address the problems existing in the prior art, this invention proposes a method and system for predicting electricity meter errors based on multi-scale dense networks.

[0005] The technical solution of the present invention is as follows: On one hand, this invention proposes a method for predicting electricity meter errors based on multi-scale dense networks, the method comprising: Obtain data from the electricity meter to be tested, including manufacturer information and forecast date. Historical errors and future predictions Environmental data for the day; The data from the electricity meter to be tested is preprocessed and then stitched together to obtain multi-scale features. Multi-scale features are input into a pre-trained multi-scale dense network, and feature extraction is performed using dense blocks of convolutional kernels of different sizes in the multi-scale dense network to obtain multi-scale fused features. The attention mechanism weights of the multi-scale fusion features are calculated based on the attention mechanism, and the weighted features are calculated based on the attention mechanism weights and the multi-scale fusion features. The weighted features are input into a long short-term memory network, which outputs a prediction of the future date of the electricity meter to be tested. Predicted daily error value.

[0006] Preferably, the preprocessing specifically includes: The manufacturer information is categorical data, which is encoded using preset embedding weights and embedding biases to obtain an embedding vector; The environmental data is numerical data, which is normalized based on its own mean and standard deviation to obtain normalized environmental data; Data cleaning is performed on historical error and environmental data, including handling missing and outlier values.

[0007] Preferably, the multi-scale dense network includes an input layer, at least three feature extraction layers with convolutional kernels of different sizes, and an output layer, wherein each feature extraction layer includes multiple dense blocks.

[0008] Preferably, feature extraction of multi-scale features is performed using dense blocks of convolutional kernels of different sizes in a multi-scale dense network. The specific steps are as follows: In the current feature extraction layer, the first... The output of the block-dense block is passed to the first Connect the dense blocks to obtain the output of the current feature extraction layer; The outputs of each feature extraction layer are concatenated along the channel dimension to obtain multi-scale fused features.

[0009] Preferably, the attention mechanism weights for multi-scale fusion features are calculated based on the attention mechanism, and the specific steps are as follows: The multi-scale fused features are linearly transformed and then input into the activation function to obtain the attention mechanism weights; The attention mechanism weights and multi-scale fusion features are weighted and calculated to obtain weighted features.

[0010] Preferably, L1 regularization and L2 regularization are added to the loss function of the Long Short-Term Memory network, and the calculation method is as follows: ; In the formula, Represents the loss function; Indicates the first The predicted error value for the day; Indicates the first The actual value of the error per day; These represent the weight coefficients for L1 regularization; The norm of L1 regularization; These represent the weights for L2 regularization; This represents the norm of L2 regularization.

[0011] Preferably, the weighted features are input into the Long Short-Term Memory network, and the specific steps are as follows: The long short-term memory network captures the temporal dependencies in the weighted features to obtain the final hidden state; The final hidden state is linearly transformed and output to predict the future date of the energy meter under test. Predicted daily error value.

[0012] Preferably, it also includes interpretability analysis based on the prediction results, the specific steps of which are as follows: Based on the gradient weighting algorithm, the gradient weight of the feature map of each channel in the multi-scale fusion feature is calculated. Sort all feature map gradient weights in descending order, output the multi-scale fused features corresponding to the top T feature map gradient weights, and generate an interpretable report.

[0013] Preferably, the method also includes using Monte Carlo Dropout to perform multiple forward propagations on the predicted values ​​to obtain multiple sets of prediction error vectors; Calculate the mean and standard deviation of the prediction error vector for each day, and output the confidence interval.

[0014] On the other hand, the present invention also provides an energy meter error prediction system based on a multi-scale dense network, the system comprising: The data acquisition module acquires data from the electricity meter under test, including manufacturer information and the forecast date. Historical errors and future predictions Environmental data for the day; The feature extraction module preprocesses the data from the electricity meter to be detected and then stitches them together to obtain multi-scale features. Multi-scale features are input into a pre-trained multi-scale dense network, and feature extraction is performed using dense blocks of convolutional kernels of different sizes in the multi-scale dense network to obtain multi-scale fused features. The attention mechanism weights of the multi-scale fusion features are calculated based on the attention mechanism, and the weighted features are calculated based on the attention mechanism weights and the multi-scale fusion features. The error prediction module inputs weighted features into a long short-term memory network and outputs a predicted future date for the energy meter under test. Predicted daily error value.

[0015] The present invention has the following beneficial effects: 1. This invention provides a method and system for predicting electricity meter errors based on multi-scale dense networks. The multi-scale dense network can capture local details and global patterns in electricity meter data simultaneously through dense blocks of convolutional kernels of different sizes. Through the dense connection structure, the output of all previous layers is used as the input of subsequent layers, realizing feature reuse and gradient backpropagation through multiple paths. This reduces the problem of gradient vanishing caused by skip connections between layers and improves the model's ability to model complex nonlinear relationships. 2. This invention provides a method and system for predicting electricity meter errors based on multi-scale dense networks. It uses embedding encoding and weight bias assignment for categorical features to transform non-numerical information into quantifiable vectors. It normalizes numerical features using mean and standard deviation to eliminate feature weight imbalance caused by differences in units. It achieves unified input and collaborative modeling of heterogeneous features through feature concatenation, solving the problems of coarse processing and poor fusion effect of existing technologies for heterogeneous features. It fully leverages the complementary value of multi-source data such as manufacturer information, environmental parameters, and historical errors, enhances the compatibility of categorical and numerical features, and avoids modeling failure caused by conflicts in units. 3. This invention provides a method and system for predicting electricity meter errors based on multi-scale dense networks. It utilizes the gating mechanism of LSTM to accurately capture long-term dependencies in long-sequence historical error data; it employs L1+L2 combined regularization to suppress the overfitting risk of complex networks under limited samples, enhancing the modeling ability for long-period error trends and accurately capturing long-term dependencies such as meter aging and seasonal electricity consumption; it improves the generalization performance of the model and enhances the stability of time-series prediction. 4. This invention provides a method and system for predicting electricity meter errors based on multi-scale dense networks. By introducing an interpretability analysis module, it can intuitively display the key features affecting error prediction, thereby improving the transparency and credibility of model decision-making. 5. This invention provides a method and system for predicting electricity meter errors based on multi-scale dense networks. Through an uncertainty quantification mechanism, it outputs the confidence interval of the prediction results, assists users in assessing the reliability of the prediction, and supports risk warning and manual review decision-making. Attached Figure Description

[0016] Figure 1 This is a detailed flowchart of an embodiment of the present invention; Figure 2 This is a diagram of a multi-scale dense network structure according to an embodiment of the present invention; Figure 3 The figure shows the experimental results of an embodiment of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] It should be understood that the step numbers used in the text are for ease of description only and are not intended to limit the order in which the steps are performed.

[0019] It should be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0020] The terms “comprising” and “including” indicate the presence of the described feature, whole, step, operation, element and / or component, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or collections thereof.

[0021] The term “and / or” refers to any combination of one or more of the associated listed items, as well as all possible combinations, and includes these combinations.

[0022] Example 1: See Figure 1 This invention provides a method for predicting electricity meter errors based on multi-scale dense networks, the method comprising: S1. Obtain data from the electricity meter to be tested, including manufacturer information and forecast date. Historical errors and future predictions Environmental data for the day; The manufacturer information includes manufacturer identification, manufacturer qualification certification information, production batch, core component configuration information, etc. The environmental data includes environmental data; S2. The data of the electricity meter to be tested is preprocessed and then spliced ​​to obtain multi-scale features; The preprocessing specifically includes: The manufacturer information is categorical data, which is encoded using preset embedding weights and embedding biases to obtain an embedding vector. The calculation method is as follows: ; In the formula, Represents an embedding vector; Indicates the preset embedding weights; Indicates the preset embedding bias; The environmental data is numerical data, which is normalized based on its mean and standard deviation to obtain normalized environmental data. The calculation method is as follows: ; In the formula, This represents the normalized environmental data; Represents environmental data; This represents the average value of environmental data; The standard deviation of environmental data; This represents a preset constant; Data cleaning is performed on historical error and environmental data, including handling missing and outlier values. The multi-scale features are represented as follows: ,in Indicates the date of the forecast Historical errors of the day; S3. Input the multi-scale features into a pre-trained multi-scale dense network; The multi-scale dense network includes an input layer, at least three feature extraction layers with convolutional kernels of different sizes, and an output layer, wherein each feature extraction layer includes multiple dense blocks; Please see Figure 2 This is a schematic diagram of a three-layer feature extraction layer. The first feature extraction layer has a 1×1 convolution kernel, the second feature extraction layer has a 3×3 convolution kernel, the third feature extraction layer has a 5×5 convolution kernel, and so on. The kernel size of the feature extraction layer is ; S4. Utilize dense blocks of convolutional kernels of different sizes in a multi-scale dense network to extract features from multi-scale features. In the current feature extraction layer, the first... The output of the block-dense block is passed to the first The dense blocks are connected to obtain the output of the current feature extraction layer. The calculation method is as follows: ; In the formula, Indicates the first Layer Feature Extraction Layer Output of block-dense blocks; Indicates the convolution operation; express function; The outputs of each feature extraction layer are concatenated along the channel dimension to obtain multi-scale fused features, calculated as follows: ; In the formula, Indicates multi-scale fusion features; Indicates the concatenation function; Indicates the first Layer Feature Extraction Layer The output of the block-dense block, i.e., the first... The output of the last dense block in the layer feature extraction layer; Indicates the number of dense blocks; S5. Calculate the attention mechanism weights of multi-scale fusion features based on the attention mechanism; The multi-scale fused features are linearly transformed and then input into the activation function to obtain the attention mechanism weights, which are calculated as follows: ; In the formula, Indicates the weights of the attention mechanism; express function; The weight matrix represents the preset first attention mechanism; The weight matrix represents the preset second attention mechanism; express function; This indicates the bias of the preset attention mechanism; S6. Weight the attention mechanism weights and multi-scale fusion features to obtain the weighted features. The calculation method is as follows: ; In the formula, Indicates weighted features; Indicates the first A multi-scale fusion feature; S7. Input the weighted features into the Long Short-Term Memory network, and output the predicted future date of the electricity meter to be tested. Predicted daily error value; S71. Add L1 regularization and L2 regularization to the loss function of the Long Short-Term Memory network. The calculation method is as follows: ; In the formula, Represents the loss function; Indicates the first The predicted error value for the day; Indicates the first The actual value of the error per day; These represent the weight coefficients for L1 regularization; The norm of L1 regularization; These represent the weights for L2 regularization; The norm of L2 regularization; S72. Input the weighted features into the Long Short-Term Memory network. The specific steps are as follows: The Long Short-Term Memory (LSTM) network captures the temporal dependencies in the weighted features to obtain the final hidden state, calculated as follows: ; ; In the formula, This represents the expanded features, i.e., the input features of the Long Short-Term Memory network; Indicates will Perform a one-dimensional unfolding; Indicates time step The hidden state; Indicates time step The unfolding characteristics; Indicates time step cellular state; Represents the Long Short-Term Memory network; The final hidden state is linearly transformed and output to predict the future date of the energy meter under test. The daily error prediction value is calculated as follows: ; In the formula, Indicates a prediction of the future Predicted daily error value; Indicates the preset linear weights; Indicates the preset linear bias; Indicates time step The hidden state, i.e. the final hidden state; S8 also includes interpretability analysis based on the prediction results, the specific steps of which are as follows: Based on the gradient weighting algorithm, the gradient weight of the feature map for each channel in the multi-scale fused feature is calculated as follows: , ; In the formula, Indicates the first Gradient weights of feature maps for each channel; This represents the total number of pixels in the feature map; Indicates the height of the feature map; Indicates the width of the feature map; Indicates the first Each channel has a pixel value in the feature map. The activation value at that location; The index indicating the height of the feature map; An index representing the width of the feature map; Sort all feature map gradient weights in descending order, output the multi-scale fusion features corresponding to the top T feature map gradient weights, and generate an interpretable report. S9 also includes using Monte Carlo Dropout to perform multiple forward propagations of the predicted values ​​to obtain multiple sets of prediction error vectors; Calculate the mean and standard deviation of the prediction error vector for each day, as follows: ; ; In the formula, Indicates the first The mean of the prediction error vector; This indicates the number of prediction error vectors; Indicates the first The group prediction error vector is in the th The predicted error value for the day; Indicates the first The standard deviation of the prediction error vector; The output confidence interval is If the width of the confidence interval exceeds the preset threshold If the prediction result is low confidence, manual review is recommended. In one specific embodiment, test data provided by a typical environment of a smart energy meter was selected for verification. The test data came from the error value of a three-phase prepaid smart energy meter, which was calibrated using a standard meter. The operating error of the energy meter could be calculated daily. The test data covered the period from April 2018 to February 2024, with a sampling frequency of once a day. The environmental parameter data came from the environmental monitoring system of the deployment site, including environmental data such as temperature and humidity for 7 days after the prediction date, with a sampling period of daily. The experiment was conducted using the historical error of 30 consecutive days before the prediction date and the daily average temperature and relative humidity for 7 days after the prediction date. Please see Figure 3 Experimental results show that the method proposed in this invention achieves an average prediction mean square error (MSE) of 0.0285%², a standard deviation of ±0.0021%², a coefficient of determination (R²) of 0.7651, and an adjusted coefficient of determination (Adjusted R²) of 0.9497 under five-fold cross-validation. This indicates that the method proposed in this invention performs well in terms of prediction accuracy and stability, and can accurately identify the changing trend of meter error.

[0023] Example 2: This embodiment provides an energy meter error prediction system based on a multi-scale dense network, the system comprising: The data acquisition module acquires data from the electricity meter under test, including manufacturer information and the forecast date. Historical errors and future predictions Environmental data for the day; The feature extraction module preprocesses the data from the electricity meter to be detected and then stitches them together to obtain multi-scale features. Multi-scale features are input into a pre-trained multi-scale dense network, and feature extraction is performed using dense blocks of convolutional kernels of different sizes in the multi-scale dense network to obtain multi-scale fused features. The attention mechanism weights of the multi-scale fusion features are calculated based on the attention mechanism, and the weighted features are calculated based on the attention mechanism weights and the multi-scale fusion features. The error prediction module inputs weighted features into a long short-term memory network and outputs a predicted future date for the energy meter under test. Predicted daily error value.

[0024] Example 3: This embodiment provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements a method for predicting the error of an energy meter based on a multi-scale dense network as described in any one of Embodiment 1.

[0025] Example 4: This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements a method for predicting electricity meter errors based on a multi-scale dense network as described in any one of Embodiment 1.

[0026] In this application embodiment, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent the existence of A alone, A and B simultaneously, or B alone. A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one of the following" and similar expressions refer to any combination of these items, including any combination of singular or plural items. For example, at least one of a, b, and c can represent: a, b, c, a and b, a and c, b and c, or a and b and c, where a, b, and c can be single or multiple.

[0027] Those skilled in the art will recognize that the units and algorithm steps described in the embodiments disclosed herein can be implemented using electronic hardware, computer software, or a combination of electronic hardware and software. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0028] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0029] In the several embodiments provided in this application, any function, if implemented as a software functional unit and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0030] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for predicting electricity meter errors based on multi-scale dense networks, characterized in that, The method includes: Obtain data from the electricity meter to be tested, including manufacturer information and forecast date. Historical errors and future predictions Environmental data for the day; The data from the electricity meter to be tested is preprocessed and then stitched together to obtain multi-scale features. Multi-scale features are input into a pre-trained multi-scale dense network, and feature extraction is performed using dense blocks of convolutional kernels of different sizes in the multi-scale dense network to obtain multi-scale fused features. The attention mechanism weights of the multi-scale fusion features are calculated based on the attention mechanism, and the weighted features are calculated based on the attention mechanism weights and the multi-scale fusion features. The weighted features are input into a long short-term memory network, which outputs a prediction of the future date of the electricity meter to be tested. Predicted daily error value.

2. The method for predicting electricity meter errors based on multi-scale dense networks according to claim 1, characterized in that, The preprocessing specifically includes: The manufacturer information is categorical data, which is encoded using preset embedding weights and embedding biases to obtain an embedding vector; The environmental data is numerical data, which is normalized based on its own mean and standard deviation to obtain normalized environmental data; Data cleaning is performed on historical error and environmental data, including handling missing and outlier values.

3. The method for predicting electricity meter errors based on multi-scale dense networks according to claim 1, characterized in that, The multi-scale dense network includes an input layer, at least three feature extraction layers with convolutional kernels of different sizes, and an output layer, wherein each feature extraction layer includes multiple dense blocks.

4. The method for predicting electricity meter errors based on multi-scale dense networks according to claim 3, characterized in that, Feature extraction of multi-scale features is performed using dense blocks of convolutional kernels of different sizes in a multi-scale dense network. The specific steps are as follows: In the current feature extraction layer, the first... The output of the block-dense block is passed to the first Connect the dense blocks to obtain the output of the current feature extraction layer; The outputs of each feature extraction layer are concatenated along the channel dimension to obtain multi-scale fused features.

5. The method for predicting electricity meter errors based on multi-scale dense networks according to claim 1, characterized in that, The attention mechanism weights for multi-scale fusion features are calculated based on the attention mechanism, and the specific steps are as follows: The multi-scale fused features are linearly transformed and then input into the activation function to obtain the attention mechanism weights; The attention mechanism weights and multi-scale fusion features are weighted and calculated to obtain weighted features.

6. The method for predicting electricity meter errors based on multi-scale dense networks according to claim 1, characterized in that, L1 and L2 regularization are added to the loss function of the Long Short-Term Memory network, and the calculation method is as follows: ； In the formula, Represents the loss function; Indicates the first The predicted error value for the day; Indicates the first The actual value of the error per day; These represent the weight coefficients for L1 regularization; The norm of L1 regularization; These represent the weights for L2 regularization; This represents the norm of L2 regularization.

7. The method for predicting electricity meter errors based on multi-scale dense networks according to claim 1, characterized in that, The weighted features are input into the Long Short-Term Memory network, and the specific steps are as follows: The long short-term memory network captures the temporal dependencies in the weighted features to obtain the final hidden state; The final hidden state is linearly transformed and output to predict the future date of the energy meter under test. Predicted daily error value.

8. The method for predicting electricity meter errors based on multi-scale dense networks according to claim 1, characterized in that, It also includes interpretability analysis based on the prediction results, the specific steps of which are as follows: Based on the gradient weighting algorithm, the gradient weight of the feature map of each channel in the multi-scale fusion feature is calculated. Sort all feature map gradient weights in descending order, output the multi-scale fused features corresponding to the top T feature map gradient weights, and generate an interpretable report.

9. The method for predicting electricity meter errors based on multi-scale dense networks according to claim 1, characterized in that, It also includes using Monte Carlo Dropout to perform multiple forward propagations of the predicted values ​​to obtain multiple sets of prediction error vectors; Calculate the mean and standard deviation of the prediction error vector for each day, and output the confidence interval.

10. An energy meter error prediction system based on a multi-scale dense network, characterized in that, The system includes: The data acquisition module acquires data from the electricity meter under test, including manufacturer information and the forecast date. Historical errors and future predictions Environmental data for the day; The feature extraction module preprocesses the data from the electricity meter to be detected and then stitches them together to obtain multi-scale features. Multi-scale features are input into a pre-trained multi-scale dense network, and feature extraction is performed using dense blocks of convolutional kernels of different sizes in the multi-scale dense network to obtain multi-scale fused features. The attention mechanism weights of the multi-scale fusion features are calculated based on the attention mechanism, and the weighted features are calculated based on the attention mechanism weights and the multi-scale fusion features. The error prediction module inputs weighted features into a long short-term memory network and outputs a predicted future date for the energy meter under test. Predicted daily error value.

Images