Method and system for evaluating metering error of electric energy meter

Abstract

The invention discloses an electric energy meter metering error assessment method and system, and the method comprises the steps: collecting the historical power data of a to-be-measured electric energy meter and corresponding environment data, extracting feature data, inputting the feature data into a DBN neural network, and calculating a relation parameter representing the relation between the feature data and a metering error; constructing an input quantity by utilizing the relation parameters; the DBN neural network outputs a predicted electric energy meter metering error by using the input quantity and the corresponding weight parameter; constructing a loss function based on the electric energy meter metering error predicted by the DBN neural network and the true value of the electric energy meter metering error, and training the network to obtain a trained DBN neural network; and utilizing the trained DBN neural network to predict the metering error of the to-be-measured electric energy meter to obtain a metering error evaluation result. The method has the advantage that the error evaluation result is more accurate.

Description

Technical Field

[0001] The present invention relates to the technical field of on-line monitoring of electric power metering, and particularly relates to a method and system for evaluating the metering error of an electric energy meter. Background Art

[0002] As an important measuring device in the power system, especially for high-voltage transmission lines, the functions of high-voltage electric energy meters in the power system mainly include the following aspects: Electric energy measurement: High-voltage electric energy meters are used to accurately measure the electric energy consumption in high-voltage power systems, which is crucial for power supply companies, industrial users, and commercial users, as it directly relates to the calculation of electricity bills and the generation of bills; Load management: By monitoring the electricity usage in real time, electric energy meters can help manage and optimize the power load, avoid overloading, and improve the reliability and efficiency of the system; Equipment protection: The use of high-voltage electric energy meters helps monitor the operating status of power equipment, prevent overload and other abnormal conditions, and thus protect power equipment from damage; Data recording and analysis: The data recorded by high-voltage electric energy meters can be used to analyze electricity consumption patterns, predict future power demands, and conduct energy efficiency evaluations, etc. Electric energy meters are widely used in the fields of electric energy measurement and power metering. Their accuracy has an important impact on power grid operation, electric energy metering, and power quality analysis. Therefore, it is crucial to accurately evaluate the metering error of electric energy meters. Traditionally, this kind of evaluation usually relies on manual off-line experiments and theoretical derivations. This method is not only time-consuming and laborious, but also limited by the control of the experimental environment and the assumptions of theoretical models, and may not be able to truly reflect the performance of the device in actual operation.

[0003] In recent years, with the progress of data acquisition technology and the development of data analysis algorithms, data-driven evaluation methods have gradually become the mainstream in power system monitoring and analysis. For the error evaluation of electric energy meters, the on-line detection method has obvious advantages: The on-line detection method can evaluate based on the output data of the electric energy meter collected in real time, and can reflect the actual performance and error situation of the device in real time. This real-time nature makes the evaluation results more representative and reliable. Traditional manual off-line evaluation is often limited by the complexity of controlling experimental conditions and environments, while the on-line detection method can conduct evaluations in complex operating environments and adapt to the performance changes of electric energy meters under different working conditions. The on-line detection method can utilize big data technology and machine learning algorithms to comprehensively analyze multi-dimensional information in the output data of electric energy meters, such as waveform characteristics, frequency response, temperature influence, etc., so as to more comprehensively evaluate the error characteristics of the device. The data-driven evaluation method can more precisely analyze the error characteristics of electric energy meters through big data samples and advanced algorithms, and can provide higher evaluation accuracy and precision compared with traditional methods.

[0004] By abandoning the traditional manual off-line evaluation method and turning to a data-driven on-line detection method for the error evaluation of electric energy meters, the real-time performance, accuracy, and comprehensiveness of the evaluation can be effectively improved, providing more reliable support and guarantee for the operation management of power systems and electric energy metering.

[0005] The literature "Research on the Error Evaluation Model of Electric Energy Meters Combining Improved SVR and Forgetting Factor Recursive Least Square Algorithm [J]. Wang Hao, Yang Peng, Li Chong. Journal of Electric Power Science and Technology, 2023, 38(05): 206-215. DOI: 10.19781 / j.issn.1673-9140.2023.05.021." proposed an error evaluation model of electric energy meters combining the Sparrow Search Algorithm (SSA), Support Vector Regression (SVR), and Forgetting Factor Recursive Least Square Algorithm (FMRLS). This method first classifies the power distribution areas using the improved K-Means algorithm, imports the classified samples into the SVR model optimized by SSA for training to establish a power distribution area line loss rate prediction model; then substitutes the obtained line loss rate into the improved line loss model to construct an electric energy meter error solution equation, and uses the FMRLS algorithm to solve the error equation to estimate the electric energy meter error. The original models used do not deeply integrate the characteristics of each algorithm, and it is difficult to improve the accuracy of the model combination, resulting in low accuracy of the evaluation results of electric energy meter measurement errors. Summary of the Invention

[0006] The technical problem to be solved by the present invention lies in the low accuracy of the evaluation results of the existing technology for the error evaluation method of electric energy meter measurement.

[0007] The present invention solves the above technical problems through the following technical means: An error evaluation method for electric energy meters, comprising the following steps:

[0008] Step 1: Collect the historical power data of the electric energy meter to be measured and the corresponding environmental data, extract the characteristic data, input it into the DBN neural network, and calculate the relationship parameters representing the relationship between the characteristic data and the measurement error.

[0009] Step 2: Use the relationship parameters to construct the input quantity.

[0010] Step 3: The DBN neural network outputs the predicted measurement error of the electric energy meter using the input quantity and the corresponding weight parameters.

[0011] Step 4: Based on the predicted measurement error of the electric energy meter by the DBN neural network and the true value of the electric energy meter measurement error, construct a loss function, continuously adjust the weight parameters of the DBN neural network to train the network and calculate the corresponding loss function value, and stop training when the loss function value is the smallest to obtain the trained DBN neural network.

[0012] Step 5: Use the trained DBN neural network to predict the measurement error of the electricity meter to be measured, and obtain the measurement error evaluation result.

[0013] The present invention selects relevant characteristic data of the electricity meter to construct the input of the DBN neural network, and constructs a loss function to train the network and optimize the model, so that the error prediction result output by the model is more accurate. According to the real-time environmental data and historical power data, the real-time error of the electricity meter can be calculated using the DBN neural network, getting rid of the offline detection of traditional methods and improving the accuracy of error evaluation.

[0014] Further, the step 1 includes:

[0015] Collect the electricity meter error obtained by offline detection in the most recent month of the electricity meter to be measured, denoted as W, W = {w 1 , w 2 ,..., w n}, the number of data points is n, and collect the historical power data and environmental data at the corresponding time to extract characteristic data. The characteristic data includes voltage data, current data, grid frequency data, and environmental temperature data, denoted as V, I, F, T respectively. V = {v 1 , v 2 ,..., v n}, I = {i 1 , i 2 ,..., i n}, F = {f 1 , f 2 ,..., f n}, T = {t 1 , t 2 ,..., t n}, and the standard deviations of V, I, F, T are denoted as σ(V), σ(I), σ(F), σ(T) respectively;

[0016] Denote the relationship parameters representing the relationship between W and V, I, F, T as r1, r2, r3, r4 respectively, and the calculation formulas are as follows:

[0017]

[0018] Among them, C is the permutation and combination symbol, and count() is the counting symbol.

[0019] Even further, the step 2 includes:

[0020] Construct the input

[0021] Furthermore, the DBN neural network includes an input layer, a hidden layer, and an output layer connected in sequence. Feature data is input into the input layer, and the expression of the output layer is σ* = Relu(σ + X), where σ* is the output layer, which outputs the predicted metering error of the watt-hour meter, σ is the weight parameter output by the hidden layer, and Relu is the activation function.

[0022] Furthermore, the process of constructing the loss function in step 4 is as follows:

[0023] Step 401: Denote the set of the metering errors of the watt-hour meters predicted by the DBN neural network as U, and the set of the true values of the metering errors of the watt-hour meters as U*. Calculate the mean square error between U and U*. U i * represents the true value of the metering error of the watt-hour meter at the i-th time point, and U i represents the predicted value of the metering error of the watt-hour meter at the i-th time point;

[0024] Step 402: Fit the slope k of U using the least squares method, fit the slope k* of U* using the least squares method, and calculate the slope difference using the slopes k and k*.

[0025] Step 403: Calculate the residual L between the straight line fitted from U* and the true value of the metering error of the watt-hour meter. Collect and record the residual L of all the watt-hour meters in the substation and take the average value as the first statistical average residual L'.

[0026] Step 404: Perform fast Fourier transforms on U* and U respectively, take the reciprocal of the main frequency to obtain the period, denoted as p* and p respectively, and calculate the period difference using p* and p.

[0027] Step 405: After performing the fast Fourier transform on U*, extract the frequencies above the preset amplitude, denote the number of the extracted frequencies as J, collect and record the J corresponding to all the watt-hour meters in the substation and take the average value as the second statistical average residual J'.

[0028] Step 406: Calculate the first weight coefficient b and the second weight coefficient c using L, L', J, and J'.

[0029] Step 407: Construct the loss function M = MSE(a) + bΔk + cΔp using the mean square error MSE(a), the slope difference Δk, the period difference Δp, the first weight coefficient b, and the second weight coefficient c.

[0030] Furthermore, the formula for calculating the residual L between the straight line fitted from U* and the true value of the metering error of the watt-hour meter in step 403 is

[0031]

[0032] Furthermore, for the first weight coefficient b and the second weight coefficient c in step 406, the formula is as follows

[0033]

[0034] The present invention also provides an electric energy meter measurement error evaluation system, including:

[0035] A relationship parameter acquisition module, configured to collect historical power data and corresponding environmental data of the electric energy meter to be measured, extract feature data, input it into the DBN neural network, and calculate relationship parameters representing the relationship between the feature data and the measurement error;

[0036] An input quantity construction module, configured to construct an input quantity by using the relationship parameters;

[0037] A model prediction module, configured to output the predicted measurement error of the electric energy meter by using the DBN neural network with the input quantity and the corresponding weight parameters;

[0038] A model training module, configured to construct a loss function based on the predicted measurement error of the electric energy meter by the DBN neural network and the true value of the measurement error of the electric energy meter, continuously adjust the weight parameters of the DBN neural network to train the network and calculate the corresponding loss function value, and stop training when the loss function value is the smallest to obtain a trained DBN neural network;

[0039] An error evaluation module, configured to predict the measurement error of the electric energy meter to be measured by using the trained DBN neural network to obtain a measurement error evaluation result.

[0040] Furthermore, the relationship parameter acquisition module is further configured to:

[0041] Collect the electric energy meter error obtained by offline detection of the electric energy meter to be measured in the most recent month, denoted as W, W = {w 1 , w 2 ,..., w n}, the number of data points is n, and collect historical power data and environmental data at the corresponding time to extract feature data. The feature data includes voltage data, current data, grid frequency data, and environmental temperature data, denoted as V, I, F, T respectively. V = {v 1 , v 2 ,..., v n}, I = {i 1 , i 2 ,..., i n}, F = {f 1 , f 2 ,..., f n}, T = {t 1 , t 2 ,..., t n}, the standard deviations of V, I, F, and T are denoted as σ(V), σ(I), σ(F), and σ(T) respectively;

[0042] The relationship parameters representing the relationships between W and V, I, F, and T are denoted as r1, r2, r3, and r4 respectively, and the calculation formulas are as follows:

[0043]

[0044] Among them, C is the permutation and combination symbol, and count() is the counting symbol.

[0045] Furthermore, the input quantity construction module is also used for:

[0046] Construct an input quantity

[0047] Furthermore, the DBN neural network includes an input layer, a hidden layer, and an output layer connected in sequence. Feature data is input into the input layer, and the expression of the output layer is σ* = Relu(σ + X), where σ* is the output layer, and the output is the predicted metering error of the electric energy meter. σ is the weight parameter output by the hidden layer, and Relu is the activation function.

[0048] Furthermore, the model training module is also used for:

[0049] Step 401: Denote the set of the metering errors of the electric energy meter predicted by the DBN neural network as U, and the set of the true values of the metering errors of the electric energy meter as U*. Calculate the mean square error between U and U*. U i * represents the true value of the metering error of the electric energy meter at the i-th time point, and U i represents the predicted value of the metering error of the electric energy meter at the i-th time point;

[0050] Step 402: Fit the slope k of U using the least squares method, fit the slope k* of U* using the least squares method, and calculate the slope difference using the slope k and the slope k*.

[0051] Step 403: Calculate the residual L between the straight line fitted by U* and the true value of the metering error of the electric energy meter. Collect and record the residuals L of all the electric energy meters in the substation and take the average value as the first statistical average residual L'.

[0052] Step 404: Perform fast Fourier transforms on U* and U respectively, take the reciprocal of the main frequency to obtain the periods, denoted as p* and p respectively, and calculate the period difference using p* and p.

[0053] Step 405: After performing a fast Fourier transform on U*, extract the frequencies above a preset amplitude. Denote the number of the extracted frequencies as J, collect and record J corresponding to all the watt-hour meters in the substation, and take the average value as the second statistical average residual J'.

[0054] Step 406: Calculate the first weight coefficient b and the second weight coefficient c using L, L', J, and J'.

[0055] Step 407: Construct a loss function M = MSE(a) + bΔk + cΔp using the mean square error MSE(a), the slope difference Δk, the period difference Δp, the first weight coefficient b, and the second weight coefficient c.

[0056] Furthermore, the formula for calculating the residual L between the straight line fitted by U* and the true value of the watt-hour meter measurement error in step 403 is

[0057]

[0058] Furthermore, the first weight coefficient b and the second weight coefficient c in step 406 are as follows

[0059]

[0060] The advantages of the present invention are as follows:

[0061] (1) The present invention selects the relevant characteristic data of the watt-hour meter to construct the input of the input DBN neural network, and constructs a loss function to train the network and optimize the model, so that the error prediction result output by the model is more accurate. According to the real-time environmental data and historical power data, the real-time error of the watt-hour meter can be calculated using the DBN neural network, getting rid of the offline detection of the traditional method and improving the accuracy of error evaluation.

[0062] (2) The present invention introduces the residual network idea into the DBN neural network, and imports the initial information into the visible layer, that is, the input layer, to avoid the phenomenon of gradient disappearance caused by the network being too deep, and further improve the accuracy of network evaluation.

[0063] (3) The optimization objective of the loss function of the existing DBN neural network is the sum of the products of three mean square errors and their weights, which are MSEo (the mean square error of the initial condition), MSEb (the mean square error of the boundary condition), and MSEf (the mean square error of the equation control). However, the present invention designs the optimization objective as the sum of the products of several characteristics such as the network mean square error, the slope difference, and the reciprocal difference of the main frequency and the weights, which can take into account the network reliability and better adapt to different working conditions. Description of the Drawings

[0064] Figure 1It is a model framework diagram of a method for evaluating the measurement error of an electric energy meter disclosed in an embodiment of the present invention. Detailed implementation manners

[0065] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Obviously, the described embodiments are some, rather than all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0066] Embodiment 1

[0067] As Figure 1 shown, Embodiment 1 of the present invention provides a method for evaluating the measurement error of an electric energy meter, including the following steps:

[0068] S1. Collect the historical power data of the electric energy meter to be measured and the corresponding environmental data, extract the feature data, input it into the DBN neural network, and calculate the relationship parameters representing the relationship between the feature data and the measurement error; the specific process is as follows:

[0069] Collect the error of the electric energy meter obtained through off-line detection in the most recent month of the electric energy meter to be measured, denoted as W, W = {w 1 , w 2 ,..., w n}, the number of data points is n, and collect the historical power data and environmental data at the corresponding time to extract the feature data. The feature data includes voltage data, current data, grid frequency data, and environmental temperature data, denoted as V, I, F, T respectively. V = {v 1 , v 2 ,..., v n}, I = {i 1 , i 2 ,..., i n}, F = {f 1 , f 2 ,..., f n}, T = {t 1 , t 2 ,..., t n}, and the standard deviations of V, I, F, T are denoted as σ(V), σ(I), σ(F), σ(T) respectively;

[0070] The relationship parameters representing the relationship between W and V, I, F, T are denoted as r1, r2, r3, r4 respectively, and the calculation formulas are as follows:

[0071]

[0072]

[0073] Among them, C is the permutation and combination symbol, and count() is the counting symbol.

[0074] S2. Construct the input quantity by using the relationship parameters

[0075] S3. The DBN neural network outputs the predicted metering error of the electric energy meter by using the input quantity and the corresponding weight parameters; the specific process is as follows:

[0076] The DBN neural network includes an input layer, a hidden layer, and an output layer connected in sequence. The feature data is input into the input layer. In the present invention, the feature data V, I, F, and T are used to construct the input quantity and imported into the visible layer, that is, the input layer. The hidden layer is composed of 4 layers of RBMs, and a short connection (that is, the output layer) is introduced into the last layer to the hidden layer to alleviate the gradient disappearance problem in the deep network. The expression of the output layer is σ* = Relu(σ + X), where σ* is the output layer, and the output is the predicted metering error of the electric energy meter, σ is the weight parameter output by the hidden layer, and Relu is the activation function.

[0077] S4. Based on the metering error of the electric energy meter predicted by the DBN neural network and the true value of the metering error of the electric energy meter, construct a loss function, continuously adjust the weight parameters of the DBN neural network to train the network and calculate the corresponding loss function value, and stop training when the loss function value is the smallest to obtain the trained DBN neural network; the specific process is as follows:

[0078] S401. Denote the set of the metering errors of the electric energy meter predicted by the DBN neural network as U, and denote the set of the true values of the metering errors of the electric energy meter as U*. Calculate the mean square error between U and U* U i * represents the true value of the metering error of the electric energy meter at the i-th time point, and U i represents the predicted value of the metering error of the electric energy meter at the i-th time point;

[0079] S402. Fit the slope k of U by using the least square method, fit the slope k* of U* by using the least square method, and calculate the slope difference by using the slope k and the slope k*

[0080] S403. Calculate the residual L between the straight line fitted by U* and the true value of the metering error of the electric energy meter, collect and record the residual L of all the electric energy meters in the substation and take the average value as the first statistical average residual L'; the formula for calculating the residual L between the straight line fitted by U* and the true value of the metering error of the electric energy meter is

[0081] S404. Perform fast Fourier transforms on U* and U respectively, take the reciprocal of the main frequency to obtain the period, denoted as p* and p respectively, and calculate the period difference using p* and p.

[0082] S405. After performing a fast Fourier transform on U*, extract the frequencies above a preset amplitude. Denote the number of the extracted frequencies as J. Collect and record the J values corresponding to all the watt-hour meters in the substation and take the average as the second statistical average residual J'. In this embodiment, the preset amplitude is 0.075.

[0083] S406. Calculate the first weight coefficient b and the second weight coefficient c using L, L', J, and J'. The formulas are as follows

[0084]

[0085] S407. Construct a loss function M = MSE(a) + bΔk + cΔp using the mean square error MSE(a), the slope difference Δk, the period difference Δp, the first weight coefficient b, and the second weight coefficient c.

[0086] S408. The feedback mechanism part trains the neural network to obtain a set of σ values that minimize M. The formula for minimizing M is σ = arg (σ) minM(σ). The specific training process is to import the historical data of the recent month, continuously adjust the weight parameters σ of the output of the hidden layer in the DBN neural network to train the network and calculate the corresponding loss function values. Stop training when the loss function value is minimized, and use the corresponding weight parameters to adjust the DBN neural network to obtain a trained DBN neural network.

[0087] S5. Use the trained DBN neural network to predict the measurement error of the watt-hour meter to be measured and obtain the measurement error evaluation result. Specifically, import the real-time voltage, current, temperature, and frequency of the watt-hour meter to be measured into the input layer for prediction, and the output value is the predicted value of the watt-hour meter error. Through simulation experiments, the measurement error evaluation results of the watt-hour meter of the present invention are shown in Table 1.

[0088] Table 1 Comparison table of the measurement error evaluation results of the watt-hour meter and the true value results

[0089]

[0090]

[0091] As can be seen from Table 1, the error evaluation deviation of the method of the present invention is small. To further quantify the accuracy, the root mean square error is used to characterize it as The root mean square error is small, indicating that the evaluated value of the watt-hour meter obtained by the method of the present invention is very close to the true value and can be used for practical applications.

[0092] Through the above technical solutions, the present invention selects relevant characteristic data of the electric energy meter to construct the input of the input DBN neural network, and constructs a loss function to train the network and optimize the model, so that the error prediction result output by the model is more accurate. According to the real-time environmental data and historical power data, the real-time error of the electric energy meter can be calculated by using the DBN neural network, getting rid of the offline detection of the traditional method and improving the accuracy of error evaluation.

[0093] Embodiment 2

[0094] Based on Embodiment 1, Embodiment 2 of the present invention further provides an electric energy meter measurement error evaluation system, including:

[0095] A relationship parameter acquisition module, configured to collect historical power data and corresponding environmental data of the to-be-tested electric energy meter, extract characteristic data, input the characteristic data into the DBN neural network, and calculate relationship parameters representing the relationship between the characteristic data and the measurement error;

[0096] An input quantity construction module, configured to construct an input quantity by using the relationship parameters;

[0097] A model prediction module, configured to use the DBN neural network to output the predicted measurement error of the electric energy meter by using the input quantity and the corresponding weight parameters;

[0098] A model training module, configured to construct a loss function based on the predicted measurement error of the electric energy meter by the DBN neural network and the true value of the measurement error of the electric energy meter, continuously adjust the weight parameters of the DBN neural network to train the network, and calculate the corresponding loss function value, and stop training when the loss function value is the smallest to obtain a trained DBN neural network;

[0099] An error evaluation module, configured to use the trained DBN neural network to predict the measurement error of the to-be-tested electric energy meter and obtain a measurement error evaluation result.

[0100] Specifically, the relationship parameter acquisition module is further configured to:

[0101] Collect the electric energy meter error obtained by offline detection of the to-be-tested electric energy meter in the most recent month, denoted as W, W = {w 1 , w 2 ,..., w n}, the number of data points is n, and collect the historical power data and environmental data at the corresponding time, extract characteristic data, and the characteristic data includes voltage data, current data, grid frequency data, and environmental temperature data, denoted as V, I, F, T respectively, V = {v 1 , v 2 ,..., v n}, I = {i 1 , i 2 ,..., i n}, F = {f 1 , f 2 ,..., f n}, T = {t 1 , t 2 ,..., t n}, and the standard deviations of V, I, F, and T are denoted as σ(V), σ(I), σ(F), and σ(T) respectively;

[0102] The relationship parameters representing the relationships between W and V, I, F, and T are denoted as r1, r2, r3, and r4 respectively, and the calculation formulas are as follows:

[0103]

[0104] where C is the permutation and combination symbol, and count() is the counting symbol.

[0105] More specifically, the input quantity construction module is further configured to:

[0106] Construct the input quantity

[0107] More specifically, the DBN neural network includes an input layer, a hidden layer, and an output layer connected in sequence. Feature data is input into the input layer, and the expression of the output layer is σ* = Relu(σ + X), where σ* is the output layer, and the output is the predicted metering error of the electric energy meter, σ is the weight parameter output by the hidden layer, and Relu is the activation function.

[0108] More specifically, the model training module is further configured to:

[0109] Step 401: Denote the set of the metering errors of the electric energy meters predicted by the DBN neural network as U, and the set of the true values of the metering errors of the electric energy meters as U*. Calculate the mean square error between U and U* U i * represents the true value of the metering error of the electric energy meter at the i-th time point, and U i represents the predicted value of the metering error of the electric energy meter at the i-th time point;

[0110] Step 402: Fit the slope k of U using the least squares method, fit the slope k* of U* using the least squares method, and calculate the slope difference using the slopes k and k*

[0111] Step 403: Calculate the residual L between the straight line fitted by U* and the true value of the metering error of the electric energy meter, collect and record the residual L of all the electric energy meters in the substation, and take the average value as the first statistical average residual L';

[0112] Step 404: Perform fast Fourier transforms on U* and U respectively, take the reciprocal of the main frequency to obtain the periods, denoted as p* and p respectively, and calculate the period difference using p* and p.

[0113] Step 405: After performing a fast Fourier transform on U*, extract the frequencies above a preset amplitude, record the number of the extracted frequencies as J, collect and record the J corresponding to all the watt-hour meters in the substation, and take the average as the second statistical average residual J'.

[0114] Step 406: Calculate the first weight coefficient b and the second weight coefficient c using L, L', J, and J'.

[0115] Step 407: Construct a loss function M = MSE(a) + bΔk + cΔp using the mean square error MSE(a), the slope difference Δk, the period difference Δp, the first weight coefficient b, and the second weight coefficient c.

[0116] More specifically, the formula for calculating the residual L between the straight line fitted by U* and the true value of the watt-hour meter measurement error in step 403 is

[0117] More specifically, for the first weight coefficient b and the second weight coefficient c in step 406, the formulas are as follows

[0118]

[0119] The above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for evaluating measurement error of an electric energy meter, characterized in that: The following steps are involved: Step 1: Collect historical power data of the electric energy meter to be tested and the corresponding environmental data to extract feature data and input them into the DBN neural network, and calculate the relationship parameters that characterize the relationship between the feature data and the metering error; Step 2: Use relational parameters to construct input quantities; Step 3: The DBN neural network uses the input quantity and the corresponding weight parameters to output the predicted electric energy meter measurement error; Step 4: construct a loss function based on the metering error of the electric energy meter predicted by the DBN neural network and the true value of the metering error of the electric energy meter, continuously adjust the weight parameters of the DBN neural network to train the network and calculate the corresponding loss function value, stop training when the loss function value is the smallest, and obtain a trained DBN neural network; Step 5: Use the trained DBN neural network to predict the metering error of the electric energy meter to obtain the metering error evaluation result.

2. The method for evaluating the measurement error of an electric energy meter according to claim 1, characterized in that: The step one comprises: The meter error of the energy meter to be tested is obtained by offline testing in the last month, denoted as W, where W = {w1,w2,...,w n }, the number of data points is n, and the historical power data and environmental data of the corresponding time are collected to extract feature data. The feature data includes voltage data, current data, grid frequency data, and ambient temperature data, which are denoted as V, I, F, and T respectively. V = {v1, v2, ..., v n }, I={i1,i2,...,i n }, F={f1,f2,...,f n }, T={t1,t2,...,t n }, the standard deviations of V, I, F, and T are denoted as σ(V), σ(I), σ(F), and σ(T), respectively; The relationship parameters representing the relationship between W and V, I, F, and T are recorded as r1, r2, r3, and r4 respectively, and the calculation formula is as follows: Among them, C is the permutation and combination symbol, and count() is the counting symbol.

3. The method for evaluating the measurement error of an electric energy meter according to claim 2, characterized in that: The second step comprises:

4. The method for evaluating the measurement error of an electric energy meter according to claim 3, characterized in that: The DBN neural network includes an input layer, a hidden layer and an output layer connected in sequence. The feature data is input to the input layer. The expression of the output layer is σ*=Relu(σ+X), where σ* is the output layer, and its output is the predicted metering error of the electric energy meter. σ is the weight parameter of the hidden layer output, and Relu is the activation function.

5. The method for evaluating the measurement error of an electric energy meter according to claim 4, characterized in that: The process of constructing the loss function in step 4 is: Step 401: The set of metering errors predicted by the DBN neural network is recorded as U, the set of true values ​​of metering errors is recorded as U*, and the mean square error between U and U* is calculated. U i * represents the true value of the metering error at the ith time point, U i represents the predicted value of the electric energy meter measurement error at the i-th time point; Step 402: Fit U to get slope k by using the least square method, fit U* to get slope k* by using the least square method, and calculate the slope difference using slope k and slope k*. Step 403, calculating the residual L between the straight line fitted by U* and the true value of the metering error of the electric energy meter, collecting and recording the residual L of all the electric energy meters in the substation and taking the average value as the first statistical average residual L'; Step 404: Perform fast Fourier transform on U* and U respectively, take the inverse of the main frequency to obtain the period, record it as p* and p respectively, and use p* and p to calculate the period difference Step 405: extract frequencies above a preset amplitude after performing a fast Fourier transform on U*, record the number of extracted frequencies as J, collect and record J corresponding to all electric energy meters in the substation and take the average value as the second statistical average residual J'; Step 406, using L, L', J, J' to calculate a first weight coefficient b and a second weight coefficient c; Step 407: construct a loss function M=MSE(a)+bΔk+cΔp using the mean square error MSE(a), the slope difference Δk, the period difference Δp, the first weight coefficient b and the second weight coefficient c.

6. The method for evaluating the measurement error of an electric energy meter according to claim 5, characterized in that: The formula for calculating the residual L between the straight line fitted by U* and the true value of the electric energy meter measurement error in step 403 is:

7. The method for evaluating the measurement error of an electric energy meter according to claim 5, characterized in that: In step 406, the first weight coefficient b and the second weight coefficient c are expressed as follows:

8. An electric energy meter measurement error evaluation system, characterized in that: include: The relationship parameter acquisition module is used to collect the historical power data of the electric energy meter to be tested and the corresponding environmental data to extract the characteristic data and input it into the DBN neural network, and calculate the relationship parameters that characterize the relationship between the characteristic data and the metering error; An input quantity construction module, used for constructing input quantities using relational parameters; Model prediction module, used for DBN neural network to output predicted electric energy meter measurement error using input quantity and corresponding weight parameters; The model training module is used to construct a loss function based on the metering error of the electric energy meter predicted by the DBN neural network and the true value of the metering error of the electric energy meter, continuously adjust the weight parameters of the DBN neural network to train the network and calculate the corresponding loss function value, and stop training when the loss function value is the smallest to obtain a trained DBN neural network; The error evaluation module is used to predict the metering error of the electric energy meter to be tested by using the trained DBN neural network to obtain the metering error evaluation result.

9. The electric energy meter measurement error evaluation system according to claim 8, characterized in that: The relationship parameter acquisition module is also used for: The meter error of the energy meter to be tested is obtained by offline testing in the last month, denoted as W, where W = {w1,w2,...,w n }, the number of data points is n, and the historical power data and environmental data of the corresponding time are collected to extract feature data. The feature data includes voltage data, current data, grid frequency data, and ambient temperature data, which are denoted as V, I, F, and T respectively. V = {v1, v2, ..., v n }, I={i1,i2,...,i n }, F={f1,f2,...,f n }, T={t1,t2,...,t n }, the standard deviations of V, I, F, and T are denoted as σ(V), σ(I), σ(F), and σ(T), respectively; The relationship parameters representing the relationship between W and V, I, F, and T are recorded as r1, r2, r3, and r4 respectively, and the calculation formula is as follows: Among them, C is the permutation and combination symbol, and count() is the counting symbol.

10. The method for evaluating the measurement error of an electric energy meter according to claim 9, characterized in that: The input quantity building module is also used to:

Images