Electric energy meter error prediction method and system based on big data analysis

Abstract

The present invention discloses an error prediction method and system for an electric energy meter based on big data analysis, and relates to the field of electric power measurement technology. The method comprises: first, reading the device attributes, voltage fluctuation threshold, and power load threshold of the target electric energy meter, dividing the parameters into multiple voltage fluctuation intervals and load ratio intervals, and constructing multiple power usage scenarios. Next, taking the device attributes and power usage scenarios as constraints, a sample training set is obtained based on the electric power big data, and trained through a BP neural network to obtain multiple error predictors. Finally, by establishing an error prediction model, the real-time voltage fluctuation coefficient, power load ratio, and environmental factors are input, and the prediction error coefficient of the electric energy meter is output. The technical problem of inaccurate error prediction of existing electric energy meters in complex environments is solved, and the technical effect of making the measurement of electric energy meters under different voltage fluctuations, environmental conditions, and load conditions more accurate is achieved.

Description

Technical Field

[0001] The present application relates to the field of electric power measurement technology, and in particular to an electric energy meter error prediction method and system based on big data analysis. Background Art

[0002] During the operation of the power system, the problem of the accumulation of energy meter errors is particularly prominent. After long-term use, the energy meter is affected by the external environment and power load fluctuations, resulting in gradually increasing errors. Especially in industrial scenarios, the complex power consumption environment and frequent voltage fluctuations make the accuracy control of the energy meter face greater challenges. The traditional energy meter error management method usually relies on regular manual calibration, which makes it difficult to detect and correct errors in time during actual operation, and cannot meet the needs of accurate measurement of energy meters under complex working conditions. At the same time, there is a lack of in-depth analysis of the causes of energy meter errors and effective use of real-time data, and it is impossible to comprehensively and accurately predict the error changes of energy meters under variable working conditions, resulting in a relatively simple and lagging error management method, which cannot effectively deal with the measurement problems of energy meters under complex working conditions.

[0003] In the current related technologies, there is a technical problem of inaccurate error prediction of electric energy meters in complex environments. Summary of the invention

[0004] The present application provides an electric energy meter error prediction method and system based on big data analysis, which first reads the device properties, voltage fluctuation threshold and power load threshold of the target electric energy meter, and divides these parameters into multiple voltage fluctuation intervals and load ratio intervals, thereby constructing multiple power usage scenarios. Then, with the device properties and power usage scenarios as constraints, a sample training set is obtained based on the power big data, and trained through the BP neural network to obtain multiple error predictors. Finally, by establishing an error prediction model, inputting the real-time voltage fluctuation coefficient, power load ratio and environmental factors, and outputting the prediction error coefficient of the electric energy meter, the error prediction accuracy of the electric energy meter in a complex environment is improved, and the technical effect of making the electric energy meter more accurate in measurement under different voltage fluctuations, environmental conditions and load conditions is achieved.

[0005] This application provides an electric energy meter error prediction method based on big data analysis, including:

[0006] The device attribute characteristics, voltage fluctuation threshold and power load threshold of the target electric energy meter are read; the voltage fluctuation threshold and the power load threshold are divided according to preset scalars to determine multiple voltage fluctuation intervals and multiple load ratio intervals, and multiple power usage scenarios are constructed based on the multiple voltage fluctuation intervals and multiple load ratio intervals; with the device attribute characteristics and multiple power usage scenarios as constraints and the environmental impact on the error of the electric energy meter as a guide, a sample training set is obtained based on the power big data retrieval, and the BP neural network is supervised and trained using the sample training set to obtain multiple error predictors; an error prediction model is built based on the multiple power usage scenarios and multiple error predictors, the real-time voltage fluctuation coefficient, the real-time power load ratio and the real-time environmental factor set are input into the error prediction model, and the prediction error coefficient of the target electric energy meter is output.

[0007] This application also provides an electric energy meter error prediction system based on big data analysis, including:

[0008] A feature reading module, the feature reading module is used to read the device attribute characteristics, voltage fluctuation threshold and power load threshold of the target electric energy meter; a load threshold division module, the load threshold division module is used to divide the voltage fluctuation threshold and the power load threshold according to a preset scalar, determine multiple voltage fluctuation intervals and multiple load ratio intervals, and construct multiple power usage scenarios based on the multiple voltage fluctuation intervals and multiple load ratio intervals; an error predictor acquisition module, the error predictor acquisition module is used to use the device attribute characteristics and multiple power usage scenarios as constraints, and the environmental impact of the electric energy meter error as a guide, to obtain a sample training set based on the power big data retrieval, and use the sample training set to supervise the BP neural network training to obtain multiple error predictors; a prediction error coefficient output module, the prediction error coefficient output module is used to build an error prediction model based on the multiple power usage scenarios and multiple error predictors, input the real-time voltage fluctuation coefficient, the real-time power load ratio and the real-time environmental factor set into the error prediction model, and output the prediction error coefficient of the target electric energy meter.

[0009] The electric energy meter error prediction method and system based on big data analysis proposed in this application first reads the device properties, voltage fluctuation threshold and power load threshold of the target electric energy meter, and divides these parameters into multiple voltage fluctuation intervals and load ratio intervals, so as to construct multiple power usage scenarios. Then, with the device properties and power usage scenarios as constraints, a sample training set is obtained based on the power big data, and trained through the BP neural network to obtain multiple error predictors. Finally, by establishing an error prediction model, inputting the real-time voltage fluctuation coefficient, power load ratio and environmental factors, and outputting the prediction error coefficient of the electric energy meter, the error prediction accuracy of the electric energy meter in a complex environment is improved, and the technical effect of making the electric energy meter more accurate in measurement under different voltage fluctuations, environmental conditions and load conditions is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solution of the embodiment of the present invention, the accompanying drawings of the embodiment of the present invention will be briefly introduced below. A flow chart is used in the present application to illustrate the operations performed by the system according to the embodiment of the present application. It should be understood that the preceding or following operations are not necessarily performed accurately in order. On the contrary, various steps can be processed in reverse order or simultaneously as needed. At the same time, other operations can also be added to these processes, or one or more operations can be removed from these processes.

[0011] Figure 1 A schematic flow chart of an electric energy meter error prediction method based on big data analysis provided in an embodiment of the present application.

[0012] Figure 2 A schematic diagram of the structure of an electric energy meter error prediction system based on big data analysis provided in an embodiment of the present application.

[0013] Explanation of reference numerals: feature reading module 10 , load threshold division module 20 , error predictor acquisition module 30 , prediction error coefficient output module 40 . DETAILED DESCRIPTION

[0014] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below.

[0015] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings. The described embodiments should not be regarded as limiting the present application. All other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of this application.

[0016] In the following description, reference is made to "some embodiments", which describe a subset of all possible embodiments, but it is understood that "some embodiments" may be the same subset or different subsets of all possible embodiments, and may be combined with each other without conflict, and the terms "first\second" involved are merely to distinguish similar objects and do not represent a specific ordering of objects. The terms "including" and "having" and any variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or modules that are not clearly listed or inherent to these processes, methods, products, or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those generally understood by technicians in the technical field of this application. The terms used herein are for the purpose of describing the embodiments of the present application only.

[0017] The present application embodiment provides an electric energy meter error prediction method based on big data analysis, such as Figure 1 As shown, the method includes:

[0018] Step S100, read the device attribute characteristics, voltage fluctuation threshold and power load threshold of the target electric energy meter. Specifically, in the complex environment of the factory workshop, first determine the target electric energy meter for error prediction, which is usually an electric energy meter installed in the workshop distribution cabinet for measuring the power consumption of various equipment. Then read its device attribute characteristics, including determining the specific model and type of the electric energy meter, understanding the specification parameters such as the measurement range and accuracy level, checking the accuracy of internal components, determining the assembly method in the distribution cabinet, understanding the factory calibration method and determining the service life. Then read the voltage fluctuation threshold, which requires analyzing the factory workshop power system, referring to relevant standards and specifications, considering the impact of special equipment and combining historical data to determine. Finally, read the power load threshold, analyze the production process to determine the maximum and minimum loads, consider the load change rate and combine it with the actual power consumption to determine it, so as to better understand the performance of the electric energy meter under different load conditions.

[0019] In a possible implementation, the device attribute characteristics, voltage fluctuation threshold and power load threshold of the target electric energy meter are read, and step S100 further includes step S110, wherein the device attribute characteristics at least include device type, specification characteristics, component accuracy, assembly method, factory calibration method and service life. Specifically, the device attribute characteristics at least include device type, specification characteristics, component accuracy, assembly method, factory calibration method and service life, wherein determining the device type of the target electric energy meter is the basis of the solution, and different types of electric energy meters differ in design, function and applicable scenarios. For example, smart electronic electric energy meters may have higher accuracy and more functions, while traditional mechanical electric energy meters may have limitations in some aspects. By clarifying the type of device, it can provide an important reference basis for subsequent analysis and prediction; the specification characteristics include the measurement range, accuracy level, rated current and voltage of the electric energy meter. The parameters determine the range of electricity that the electric energy meter can accurately measure and the accuracy level. In complex environments such as factory workshops, it is necessary to select electric energy meters with appropriate specifications and characteristics according to actual electricity demand to ensure the accuracy of measurement; the accuracy of components affects the performance of the electric energy meter. When considering the voltage fluctuation coefficient, high-precision components can improve the responsiveness of the electric energy meter to voltage changes and reduce measurement errors. For example, high-quality voltage transformers and current transformers can more accurately measure the voltage and current in the power grid, thereby improving the overall accuracy of the electric energy meter. The manufacturing accuracy of the circuit board and the performance of the electronic components will also affect the stability and reliability of the electric energy meter; the assembly method will affect the electric energy meter The correct assembly method can ensure the stable and reliable connection between the electricity meter and the power grid and reduce the impact of external interference on the measurement. For example, vertical installation and horizontal installation will lead to different errors. Therefore, it is necessary to select the appropriate assembly method according to the design requirements of the electricity meter and the actual installation environment; the factory calibration method determines the initial accuracy of the electricity meter. Different manufacturers use different calibration methods and standards, which affects the performance of the electricity meter in actual use. Understanding the factory calibration method can help us evaluate the accuracy of the electricity meter and recalibrate or adjust it when necessary. The service life is an important factor affecting the performance of the electricity meter. With the passage of time, the parts of the electricity meter may age and wear, resulting in a decrease in measurement accuracy. When predicting errors, it is necessary to consider the impact of service life on the performance of the electricity meter so that it can be maintained and replaced in time.

[0020] Step S120, the voltage fluctuation threshold is a voltage fluctuation coefficient threshold of the power supply grid of the target electric energy meter, wherein the voltage fluctuation coefficient is calculated based on the voltage fluctuation frequency and the voltage fluctuation amplitude. Specifically, the voltage fluctuation coefficient is calculated based on the voltage fluctuation frequency and the voltage fluctuation amplitude. In the power supply grid of the target electric energy meter, voltage fluctuation is inevitable. The voltage fluctuation frequency indicates the number of times the voltage changes within a certain period of time, while the voltage fluctuation amplitude indicates the magnitude of the voltage change. By calculating these two parameters, the voltage fluctuation coefficient can be obtained, which is used to evaluate the stability of the grid voltage. Nonlinear loads (such as motors, inverters, etc.) will generate non-sinusoidal currents, resulting in the inability of the electric energy meter to measure accurately. Nonlinear loads will cause the voltage and current waveforms in the grid to distort, thereby affecting the electric energy meter's measurement of voltage and current. Especially in environments such as factory workshops, nonlinear loads are more common, so it is necessary to consider their impact on the voltage fluctuation coefficient to improve the measurement accuracy of the electric energy meter. The electric energy meter under light load or no-load conditions, especially the mechanical electric energy meter, may not be able to accurately reflect the extremely low power consumption, resulting in a large relative error. In the factory workshop, there will be some equipment running under light load or no-load conditions. At this time, special attention should be paid to the measurement error of the electric energy meter. The influence of light load or no-load on voltage fluctuation coefficient can be reduced by selecting appropriate types and specifications of electric energy meters, or by adopting special measurement methods. Impact loads (such as large motor starting, welding machines, etc.) will cause excessive instantaneous power changes. If the measurement response of the electric energy meter is not timely, it may lead to error accumulation. Impact loads will cause the voltage and current in the power grid to change dramatically in an instant, which puts higher requirements on the measurement capability of the electric energy meter. When designing and selecting electric energy meters, it is necessary to consider their adaptability to impact loads to ensure the accuracy of the measurement.

[0021] Step S130, the power load threshold is the power load ratio threshold of the target power meter's power equipment. Specifically, the power load threshold is the power load ratio threshold of the target power meter's power equipment. The threshold represents the load range of the power equipment connected to the power meter during normal operation. By determining the power load ratio threshold, the performance of the power meter under different load conditions can be understood, providing an important reference for error prediction. Determining the power load ratio threshold requires consideration of multiple factors. First, it is necessary to analyze the production process of the factory workshop to understand the power demand and load characteristics of each device. Secondly, according to the capacity and operating conditions of the equipment, the maximum and minimum power loads are determined. Then, the rate of change of the load and the power consumption in different time periods are considered. Finally, a reasonable power load ratio threshold is determined in combination with the actual power monitoring data. When determining the power load ratio threshold, factors such as the type, specification and accuracy level of the power meter need to be considered to ensure the accuracy and practicality of the threshold.

[0022] Step S200, divide the voltage fluctuation threshold and the power load threshold according to the preset scalar, determine multiple voltage fluctuation intervals and multiple load ratio intervals, and construct multiple power consumption scenarios based on the multiple voltage fluctuation intervals and multiple load ratio intervals. Specifically, first determine the preset scalar, divide the voltage fluctuation threshold of the target electric energy meter power supply network into multiple voltage fluctuation intervals according to the scalar, and the voltage fluctuation coefficient is calculated based on the voltage fluctuation frequency and amplitude, such as dividing it with a specific fluctuation amplitude as one interval. At the same time, the power load threshold of the target electric energy meter electrical equipment is also divided into multiple load ratio intervals based on the preset scalar, for example, with a specific load ratio as one interval. Then, randomly select an interval from the multiple voltage fluctuation intervals to represent a specific voltage fluctuation situation, and then randomly select an interval from the multiple load ratio intervals to represent a specific power load situation, and fuse the two to construct a power consumption scenario, and then repeat this process continuously, randomly select different intervals for fusion, and construct multiple different power consumption scenarios to simulate various different power consumption situations, so as to provide support for improving the accuracy of the error prediction of the electric energy meter.

[0023] In a possible implementation, the voltage fluctuation threshold and the power load threshold are divided according to a preset scalar, a plurality of voltage fluctuation intervals and a plurality of load ratio intervals are determined, and a plurality of power consumption scenarios are constructed based on the mixing of the plurality of voltage fluctuation intervals and the plurality of load ratio intervals. Step S200 further includes step S210, dividing the voltage fluctuation threshold according to a first preset scalar, dividing the power load threshold according to a second preset scalar, and determining a plurality of voltage fluctuation intervals and a plurality of load ratio intervals. Specifically, firstly, the roles of the first preset scalar and the second preset scalar are clarified. The first preset scalar, as the step length for dividing the voltage fluctuation threshold, determines how many intervals the voltage fluctuation threshold is divided into and the size of each interval. For example, if the first preset scalar is small, the divided voltage fluctuation intervals will be more detailed and can more accurately reflect different degrees of voltage fluctuation. Similarly, the second preset scalar, as the step length for dividing the power load threshold, is used to determine the granularity of the load ratio interval. According to the first preset scalar, the voltage fluctuation threshold is divided. Assuming that the range of the voltage fluctuation threshold is from A to B, the first preset scalar is used to divide the voltage fluctuation threshold into two intervals. A preset scalar is C, then the range can be divided according to the step size of C to obtain multiple voltage fluctuation intervals. For example, the first interval may be from A to A+C, the second interval is from A+C to A+2C, and so on. In this way, multiple different voltage fluctuation intervals are determined, each interval represents the voltage fluctuation within a certain range, and the power load threshold is divided according to the second preset scalar. Similarly, according to the range of the power load threshold and the second preset scalar, it is divided into multiple load ratio intervals, each interval corresponds to a different power load ratio range, reflecting different power load conditions.

[0024] Step S220, randomly select a first voltage fluctuation interval from the multiple voltage fluctuation intervals, randomly select a first load ratio interval from the multiple load ratio intervals, fuse the first voltage fluctuation interval and the first load ratio interval to construct a first power usage scenario, and add it to the multiple power usage scenarios. Specifically, randomly select an interval from the multiple voltage fluctuation intervals that have been divided, and use it as the first voltage fluctuation interval. The random selection process can increase the diversity of power usage scenarios and simulate different voltage fluctuation conditions. Also randomly select an interval from the multiple load ratio intervals and define it as the first load ratio interval. Considering various different power load conditions, the first voltage fluctuation interval and the first load ratio interval are fused to construct the first power usage scenario. The fusion process can be done in a simple combination manner, such as merging the feature descriptions of the two intervals together to form a power usage scenario that can simultaneously reflect voltage fluctuations and power load conditions, and adding the constructed first power usage scenario to the set of multiple power usage scenarios. With the continuous repetition of this process, more different power usage scenarios can be constructed to provide rich samples for subsequent electric energy meter error prediction.

[0025] Step S300, taking the device attribute characteristics and multiple power usage scenarios as constraints, taking the environmental impact on the energy meter error as a guide, obtaining a sample training set based on the power big data retrieval, and using the sample training set to supervise the BP neural network to obtain multiple error predictors. Specifically, in this scheme, the device attribute characteristics (including device type, specification characteristics, component accuracy, assembly method, factory calibration method and service life, etc.) and multiple power usage scenarios (constructed by dividing the voltage fluctuation threshold and the power load threshold) are constrained, and the environmental impact on the energy meter error is guided. Using the power big data platform, according to these constraints, search and filter out data with the same device type and running in similar power usage scenarios, and further filter with environmental impact as a guide, and the filtered data is composed of a sample training set. Then initialize the BP neural network, determine its structure and randomly initialize the weights and biases, input the data in the sample training set batch by batch for supervised training, and adjust the network weights and biases through the back propagation algorithm until the network performance reaches the standard. The training process is repeated for different power usage scenarios to obtain multiple error predictors, each of which is targeted at a specific power usage scenario and can better predict the error of the electric energy meter in the scenario.

[0026] In a possible implementation, the device attribute characteristics and multiple power usage scenarios are constrained, and the environmental impact on the error of the electric energy meter is used as a guide. A sample training set is obtained based on the power big data retrieval, and the BP neural network is supervised and trained using the sample training set to obtain multiple error predictors. Step S300 further includes step S310, configuring environmental influencing factors according to the environmental impact on the error of the electric energy meter, wherein the environmental influencing factors at least include ambient temperature, ambient humidity, magnetic field strength, and vibration impact characteristics. Specifically, it is first recognized that environmental factors will affect the error of the electric energy meter. In complex environments such as factory workshops, different environmental conditions will cause deviations in the measurement results of the electric energy meter. It is clarified that environmental influencing factors include ambient temperature, ambient humidity, magnetic field strength, and vibration impact characteristics. Changes in ambient temperature may affect the performance of the electronic components inside the electric energy meter, thereby affecting the measurement accuracy. When the ambient humidity is high, the insulation performance of the electric energy meter may be reduced, resulting in errors. In an environment with a strong magnetic field, the magnetic field may interfere with the normal operation of the electric energy meter, making its measurement results inaccurate. Vibration and shock characteristics are more common in some specific industrial environments. For example, the operation of mechanical equipment in a factory workshop may generate vibration and shock, which may also affect the performance of the electric energy meter.

[0027] Step S320, randomly select a first electricity usage scenario from the multiple electricity usage scenarios, use the device type, specification features, component accuracy, assembly method, factory calibration method, service life and the first electricity usage scenario as constraints, and use the environmental influencing factors as a guide to obtain a first sample training set based on power big data retrieval. Specifically, randomly selecting an electricity usage scenario from the multiple electricity usage scenarios that have been constructed as the first electricity usage scenario can increase the diversity of samples and avoid overfitting problems caused by training only for specific scenarios. Use the device type, specification features, component accuracy, assembly method, factory calibration method, service life and the first electricity usage scenario as constraints. The constraints can ensure that the retrieved sample data has similar properties and operating environment to the target electricity meter. For example, if the target electricity meter is an industrial electricity meter of a specific model, then when retrieving sample data, electricity meter data of the same or similar model is given priority; if the target electricity meter has been used for a certain number of years, then it is also selected. Electricity meter data with similar service life can be selected to improve the relevance of samples, and retrieval can be guided by environmental influencing factors. Electricity meter data that are greatly affected by environmental factors such as ambient temperature, ambient humidity, magnetic field strength, and vibration impact characteristics can be found in the power big data. This can make the retrieved sample data more targeted and better reflect the relationship between environmental factors and electricity meter errors. Through the above constraints and guiding factors, retrieval is performed in the power big data to obtain qualified sample data and form the first sample training set. The sample training set contains electricity meter data under various combinations of device attribute characteristics, power usage scenarios, and environmental factors, as well as corresponding error values.

[0028] Step S330, supervised training is performed on the BP neural network using the first sample training set to obtain a first error predictor, which is added to the plurality of error predictors. Specifically, the BP neural network is initialized, and the structure of the network is determined, including the number of nodes in the input layer, the hidden layer, and the output layer. The nodes in the input layer may correspond to parameters such as device attribute characteristics, power usage scenarios, and environmental factors, and the nodes in the output layer may correspond to error prediction values ​​of the electric energy meter. The weights and biases of the network are randomly initialized, and the BP neural network is supervised and trained using the first sample training set. The data in the sample training set is input into the network batch by batch, and the output value of the network is calculated and compared with the actual error value to obtain an error signal. According to the error signal, the weights and biases of the network are adjusted using a back propagation algorithm so that the output of the network gradually approaches the actual error value. The process is repeated until the performance of the network reaches a certain standard, such as the error is less than a preset value or after a certain number of training times. After training, a first error predictor is obtained. The error predictor can predict the error of the electric energy meter according to parameters such as input device attribute characteristics, power usage scenarios and environmental factors. The first error predictor is added to a set of multiple error predictors. As the above process is repeated, multiple different error predictors can be obtained for different power usage scenarios and environmental factors, providing richer tools for subsequent electric energy meter error prediction.

[0029] In a possible implementation, a first electricity usage scenario is randomly selected from the multiple electricity usage scenarios, and the device type, specification features, component accuracy, assembly method, factory calibration method, service life and the first electricity usage scenario are constrained, and the environmental influencing factors are guided, based on the power big data retrieval to obtain a first sample training set, step S320 further includes step S321, performing a similarity comparison on the retrieved data according to the device type to obtain a first similarity. Specifically, first, the source of the retrieval data to be similarly compared is clarified. These data are usually retrieved from the power big data according to certain preliminary conditions. The device type of the target electric energy meter is used as a benchmark to compare the device type of the electric energy meter in the retrieval data. Various similarity measurement methods can be used, such as calculating the text similarity of the device type name, comparing key technical parameters, etc., so as to obtain a first similarity. The similarity value reflects the degree of similarity between the electric energy meter in the retrieval data and the target electric energy meter in terms of device type.

[0030] Step S322, if the first similarity is greater than a predetermined similarity threshold, then the retrieved data is sequentially compared for similarity according to the specification features, component accuracy, assembly method, factory calibration method, service life, and the first power usage scenario. If the feature comparison similarities all meet the predetermined similarity threshold, then the retrieved data is added to the sample environment parameter set. Specifically, when the first similarity is greater than the predetermined similarity threshold, it means that the electric energy meter in the retrieved data is relatively similar to the target electric energy meter in terms of device type. At this time, multiple features such as specification features, component accuracy, assembly method, factory calibration method, service life, and the first power usage scenario are further sequentially compared for similarity. For each feature, a suitable similarity measurement method is also used. For example, for specification features, the proximity of parameters such as measurement range and accuracy level can be compared; for component accuracy, the manufacturing accuracy of key components can be compared. Only when the comparison similarities of all features meet the predetermined similarity threshold, the retrieved data is added to the sample environment parameter set, and the sample environment parameter set will be used for subsequent further analysis and processing.

[0031] Step S323, randomly select a first sample environmental parameter from the sample environmental parameter set, and use the device type, specification features, component accuracy, assembly method, factory calibration method, service life, first power usage scenario and the first sample environmental parameter as constraints, and obtain a sample error coefficient set based on the power big data retrieval. Specifically, a sample environmental parameter is randomly selected from the sample environmental parameter set as the first sample environmental parameter. The random selection process can increase the diversity of samples. Use the device type, specification features, component accuracy, assembly method, factory calibration method, service life, first power usage scenario and the first sample environmental parameter as constraints, and search again in the power big data. The goal of this search is to obtain a sample error coefficient set, that is, a data set containing error coefficients of electric energy meters under various conditions. The error coefficient can reflect the error of the electric energy meter under different environments and operating conditions.

[0032] Step S324, high-frequency feature analysis is performed on the sample error coefficient set, and a first sample error coefficient is output. Specifically, multiple sample error coefficients in the sample error coefficient set are analyzed, and firstly, mean calculation is performed to obtain the mean of the error coefficient. The mean can be used as a reference value to reflect the overall level of the sample error coefficient set. Based on the mean of the error coefficient, deviation calculation is performed on each sample error coefficient in the sample error coefficient set to determine multiple coefficient deviations. Sample error coefficients corresponding to coefficient deviations less than a predetermined deviation threshold are selected to construct a standard error coefficient set. The standard error coefficient set contains relatively stable and reliable sample error coefficients. The mean of the standard error coefficient set is calculated to obtain a first sample error coefficient. The first sample error coefficient is a relatively accurate and representative error coefficient value obtained after high-frequency feature analysis.

[0033] Step S325, constructing first sample training data according to the first sample environmental parameter and the first sample error coefficient, and adding it to the first sample training set. Specifically, the first sample environmental parameter and the first sample error coefficient are combined to construct the first sample training data, the data includes the error coefficient information of the electric energy meter under specific environmental parameters, and the first sample training data is added to the first sample training set. The first sample training set will be used for supervised training of the BP neural network to obtain the error predictor. With the continuous repetition of the above process, the first sample training set can be continuously enriched to improve the accuracy and reliability of the training.

[0034] In a possible implementation, the sample error coefficient set is subjected to high-frequency feature analysis, and the first sample error coefficient is output. Step S324 further includes step S3241, in which the sample error coefficient set is subjected to mean calculation to obtain the error coefficient mean. Specifically, first, it is clarified that the sample error coefficient set is a set of error coefficients obtained by collecting data from electric energy meters under specific conditions. The error coefficient reflects the degree of measurement error of the electric energy meter under different conditions. When performing mean calculation, all sample error coefficients in the sample error coefficient set are added together and then divided by the number of samples. For example, if the sample error coefficient set has n sample error coefficients, namely e1, e2, ..., en, then the error coefficient mean M=(e1+e2+...+en) / n. This mean represents the overall average level of the sample error coefficient set, and provides a benchmark value for subsequent deviation calculation.

[0035] Step S3242, taking the error coefficient mean as a reference, respectively performing deviation calculations on the multiple sample error coefficients in the sample error coefficient set to determine multiple coefficient deviations. Specifically, taking the error coefficient mean calculated in the first step as a reference, for each sample error coefficient in the sample error coefficient set, calculating its deviation from the mean, the deviation calculation method may be an absolute deviation or a relative deviation, and by performing deviation calculations on each sample error coefficient, multiple coefficient deviations are obtained, and the deviation value reflects the degree of difference between each sample error coefficient and the overall average level.

[0036] Step S3243, select the sample error coefficients corresponding to the coefficient deviations less than the predetermined deviation threshold to construct a standard error coefficient set, and perform mean calculation on the standard error coefficient set to obtain the first sample error coefficient. Specifically, a deviation threshold is pre-set, and the threshold can be determined according to actual conditions and experience, and is used to screen out relatively stable and reliable sample error coefficients. From multiple coefficient deviations, select the sample error coefficients corresponding to the coefficient deviations less than the predetermined deviation threshold. The sample error coefficients are considered to be relatively close to the overall average level and have small fluctuations. The selected sample error coefficients are constructed into a standard error coefficient set, and the standard error coefficient set contains relatively reliable error coefficient data. The standard error coefficient set is mean calculated. Similarly, a method similar to the first step is used to add all sample error coefficients in the standard error coefficient set, and then divide by the number of samples to obtain the first sample error coefficient. The first sample error coefficient is a relatively accurate and representative error coefficient value obtained after screening and processing, and can be used for subsequent electric energy meter error prediction and analysis.

[0037] In a possible implementation, supervised training is performed on the BP neural network using the first sample training set to obtain a first error predictor, which is added to the multiple error predictors. Step S330 further includes step S331, dividing the first sample training set into Q equal parts, selecting Q times with replacement, constructing a first training set, and iteratively selecting Q times using the same method to obtain Q training sets. Specifically, first, the first sample training set is divided into Q parts in order to increase the diversity of samples so that different subsets can be used for training and verification in the future. Selecting with replacement means that at each selection, the selected sample still has a chance to be selected again at the next selection. Selecting with replacement is performed Q times to construct a first training set, so that the first training set has a certain degree of randomness and contains different sample combinations. Iterative selection is performed Q times using the same method, and a new training set is constructed each time, and finally Q different training sets are obtained, which provides a basis for subsequent multiple rounds of training of the BP neural network.

[0038] Step S332, using the sample environmental parameters as input and the sample error coefficient as output, the BP neural network is supervised and trained and verified using the Q training sets to obtain Q convergence error prediction branches. Specifically, the input and output are clarified, the sample environmental parameters are used as the input of the BP neural network, and the environmental parameters may include various factors that affect the error of the electric energy meter, such as ambient temperature, humidity, magnetic field strength, etc. The sample error coefficient is used as the output, that is, the target value that the BP neural network needs to learn, and the BP neural network is supervised and trained using the Q training sets. For each training set, the sample environmental parameters are input into the network, and the predicted sample error coefficient is calculated through the forward propagation of the network, and then compared with the actual sample error coefficient to obtain an error signal. According to the error signal, the weight and bias of the network are adjusted using the back propagation algorithm, so that the output of the network gradually approaches the actual sample error coefficient. In the process of supervised training, verification training is also required. Each training set can be divided into a training subset and a verification subset. The training subset is used for supervised training, and then the trained network is verified using the verification subset to evaluate the performance of the network. By continuously adjusting the network parameters and conducting multiple rounds of training until the network's performance on the validation set reaches a certain standard, Q converged error prediction branches are obtained. Each branch is obtained through training with a different training set and has a certain degree of independence and difference.

[0039] Step S333, based on the Q converged error prediction branches, the first error predictor is constructed by integrating, wherein the output of the first error predictor is the mean of the output results of the Q converged error prediction branches. Specifically, after obtaining the Q converged error prediction branches, they are integrated to construct the first error predictor. The integration method can adopt a simple averaging method, that is, the output results of the Q branches are averaged, and for the new input sample environment parameters, they are respectively input into the Q converged error prediction branches to obtain Q output results. Then the mean of the Q results is calculated as the final output of the first error predictor, which can integrate the prediction results of multiple branches to improve the accuracy and stability of the prediction. By integrating and constructing the first error predictor, it can better adapt to different input conditions and improve the prediction ability of the electric energy meter error.

[0040] Step S400, build an error prediction model based on the multiple power usage scenarios and multiple error predictors, input the real-time voltage fluctuation coefficient, the real-time power load ratio and the real-time environmental factor set into the error prediction model, and output the prediction error coefficient of the target electric energy meter. Specifically, first, it is clear that there are multiple power usage scenarios and multiple error predictors. The power usage scenario is constructed by dividing the voltage fluctuation threshold and the power load threshold of the target electric energy meter. Each power usage scenario reflects a different combination of voltage fluctuations and power load conditions. The error predictor is obtained by using the device attribute characteristics and specific power usage scenarios as constraints, and the environmental impact on the electric energy meter error as a guide, based on the power big data retrieval to obtain a sample training set, and using the sample training set to supervise the BP neural network training. Each power usage scenario is regarded as a subnode in the classification decision tree. The subnode represents a different classification of power usage conditions. For example, a power usage scenario may be a high power scenario. The voltage fluctuation and high load situation of another power consumption scenario may be low voltage fluctuation and medium load situation. By taking different power consumption scenarios as child nodes, classification and judgment can be gradually performed according to the input real-time data to determine the current power consumption scenario. For each power consumption scenario child node, the corresponding error predictor is used as the leaf node of the child node. The error predictor is specially trained for a specific power consumption scenario and can accurately predict the error coefficient of the target electric energy meter under the scenario. After determining the current power consumption scenario, the error predictor corresponding to the scenario can be directly called for error prediction. By establishing a mapping relationship between the power consumption scenario and the error predictor, the entire Error prediction model, the mapping relationship can be a one-to-one relationship, that is, each power consumption scenario corresponds to a unique error predictor. When building the model, it is necessary to ensure the accuracy and reliability of this mapping relationship so that the corresponding predictor can be found quickly and accurately in the actual prediction process. The real-time voltage fluctuation coefficient, real-time power load ratio and real-time environmental factor set are input into the error prediction model. The real-time data reflects the actual power consumption and environmental factors of the current target power meter. Based on the input real-time data, starting from the root node of the model, it is gradually compared with each power consumption scenario sub-node to determine which power consumption scenario the current data best conforms to. For example, by comparing the real-time voltage fluctuation coefficient The current electricity usage scenario is determined by combining the voltage fluctuation range and load ratio corresponding to the number and power load ratio of each electricity usage scenario. Once the current electricity usage scenario is determined, the error predictor corresponding to the scenario can be directly called for prediction. The real-time environmental factor set is used as input, and the error predictor outputs the prediction error coefficient of the target electricity meter. The error prediction model outputs the prediction error coefficient of the target electricity meter. The prediction error coefficient can provide an important reference for error monitoring and calibration of electricity meters in complex environments such as factory workshops. By timely understanding the error situation of the electricity meter, corresponding measures can be taken for adjustment and maintenance to ensure the measurement accuracy and reliability of the electricity meter.

[0041] The embodiment of the present application adopts the method of first reading the device properties, voltage fluctuation threshold and power load threshold of the target electric energy meter, and dividing these parameters into multiple voltage fluctuation intervals and load ratio intervals, so as to construct multiple power usage scenarios. Then, with the device properties and power usage scenarios as constraints, a sample training set is obtained based on the power big data, and trained through the BP neural network to obtain multiple error predictors. Finally, by establishing an error prediction model, inputting the real-time voltage fluctuation coefficient, power load ratio and environmental factors, and outputting the prediction error coefficient of the electric energy meter, the error prediction accuracy of the electric energy meter in a complex environment is improved, and the technical effect of making the electric energy meter more accurate in measuring under different voltage fluctuations, environmental conditions and load conditions is achieved.

[0042] In the above, refer to Figure 1 The error prediction method of electric energy meter based on big data analysis according to an embodiment of the present invention is described in detail. Figure 2 An electric energy meter error prediction system based on big data analysis according to an embodiment of the present invention is described.

[0043] The electric energy meter error prediction system based on big data analysis according to the embodiment of the present invention is used to solve the technical problem of inaccurate error prediction of existing electric energy meters in complex environments, and achieves the technical effect of improving the error prediction accuracy of electric energy meters in complex environments, making the measurement of electric energy meters under different voltage fluctuations, environmental conditions and load conditions more accurate. The electric energy meter error prediction system based on big data analysis includes: a feature reading module 10, a load threshold division module 20, an error predictor acquisition module 30, and a prediction error coefficient output module 40.

[0044] The feature reading module 10 is used to read the device attribute features, voltage fluctuation threshold and power load threshold of the target electric energy meter.

[0045] The load threshold division module 20 is used to divide the voltage fluctuation threshold and the power load threshold according to a preset scalar, determine multiple voltage fluctuation intervals and multiple load proportion intervals, and construct multiple power consumption scenarios based on the multiple voltage fluctuation intervals and multiple load proportion intervals.

[0046] The error predictor acquisition module 30 is used to obtain a sample training set based on power big data retrieval, using the device attribute characteristics and multiple power usage scenarios as constraints, and the environmental impact of the power meter error as a guide, and use the sample training set to supervise the BP neural network to obtain multiple error predictors.

[0047] The prediction error coefficient output module 40 is used to build an error prediction model based on the multiple power usage scenarios and multiple error predictors, input the real-time voltage fluctuation coefficient, real-time power load ratio and real-time environmental factor set into the error prediction model, and output the prediction error coefficient of the target electric energy meter.

[0048] The specific configuration of the feature reading module 10 will be described in detail below. As described above, the device attribute characteristics, voltage fluctuation threshold and power load threshold of the target electric energy meter are read, and the feature reading module 10 further includes: a device attribute characteristic composition unit, the device attribute characteristic composition unit is used for the device attribute characteristics including at least device type, specification characteristics, component accuracy, assembly method, factory calibration method and service life; a voltage fluctuation threshold target unit, the voltage fluctuation threshold target unit is used for the voltage fluctuation threshold being the voltage fluctuation coefficient threshold of the power supply network of the target electric energy meter, wherein the voltage fluctuation coefficient is calculated based on the voltage fluctuation frequency and the voltage fluctuation amplitude; a power load threshold target unit, the power load threshold target unit is used for the power load threshold being the power load ratio threshold of the electric equipment of the target electric energy meter.

[0049] Below, the specific configuration of the load threshold division module 20 will be described in detail. As described above, the voltage fluctuation threshold and the power load threshold are divided according to a preset scalar, multiple voltage fluctuation intervals and multiple load ratio intervals are determined, and multiple power usage scenarios are constructed based on the multiple voltage fluctuation intervals and multiple load ratio intervals. The load threshold division module 20 further includes: a power load threshold division unit, the power load threshold division unit is used to divide the voltage fluctuation threshold according to a first preset scalar, divide the power load threshold according to a second preset scalar, and determine multiple voltage fluctuation intervals and multiple load ratio intervals; a first power usage scenario construction unit, the first power usage scenario construction unit is used to randomly select a first voltage fluctuation interval from the multiple voltage fluctuation intervals, randomly select a first load ratio interval from the multiple load ratio intervals, and fuse the first voltage fluctuation interval and the first load ratio interval to construct a first power usage scenario, and add it to the multiple power usage scenarios.

[0050] The specific configuration of the error predictor acquisition module 30 will be described in detail below. As described above, with the device attribute characteristics and multiple power usage scenarios as constraints, with the environmental impact of the electric energy meter error as a guide, a sample training set is obtained based on the power big data retrieval, and the BP neural network is supervised and trained using the sample training set, and multiple error predictors are obtained further including: an influencing factor configuration unit, the influencing factor configuration unit is used to configure environmental influencing factors according to the environmental impact of the electric energy meter error, wherein the environmental influencing factors at least include ambient temperature, ambient humidity, magnetic field strength and vibration impact characteristics; a first sample training set acquisition unit, the first sample training set acquisition unit is used to randomly select a first power usage scenario from the multiple power usage scenarios, with the device type, specification characteristics, component accuracy, assembly method, factory calibration method, service life and the first power usage scenario as constraints, with the environmental influencing factors as a guide, based on the power big data retrieval to obtain the first sample training set; a first error predictor acquisition unit, the first error predictor acquisition unit is used to supervise the BP neural network using the first sample training set to obtain a first error predictor, and add it to the multiple error predictors.

[0051] Among them, a first electricity usage scenario is randomly selected from the multiple electricity usage scenarios, and the device type, specification characteristics, component accuracy, assembly method, factory calibration method, service life and the first electricity usage scenario are constrained, and the environmental influencing factors are used as a guide. Based on the power big data retrieval, a first sample training set is obtained, and the first sample training set acquisition unit further includes: a first similarity acquisition subunit, the first similarity acquisition subunit is used to perform a similarity comparison on the retrieved data according to the device type to obtain a first similarity; a retrieval data similarity comparison subunit, the retrieval data similarity comparison subunit is used to perform a similarity comparison on the retrieved data in turn according to the specification characteristics, component accuracy, assembly method, factory calibration method, service life and the first electricity usage scenario if the first similarity is greater than a predetermined similarity threshold, and if the feature comparison similarities all meet the requirements, The predetermined similarity threshold is met, the retrieval data is added to the sample environment parameter set; a sample error coefficient set acquisition subunit, the sample error coefficient set acquisition subunit is used to randomly select a first sample environment parameter from the sample environment parameter set, and use the device type, specification characteristics, component accuracy, assembly method, factory calibration method, service life, first power usage scenario and first sample environment parameter as constraints, and acquire the sample error coefficient set based on the power big data retrieval; a high-frequency feature analysis subunit, the high-frequency feature analysis subunit is used to perform high-frequency feature analysis on the sample error coefficient set, and output a first sample error coefficient; a first sample training data construction subunit, the first sample training data construction subunit is used to construct the first sample training data according to the first sample environment parameter and the first sample error coefficient, and add it to the first sample training set.

[0052] Among them, a high-frequency feature analysis is performed on the sample error coefficient set to output a first sample error coefficient, and the high-frequency feature analysis subunit further includes: a mean calculation microunit, the mean calculation microunit is used to perform mean calculation on the sample error coefficient set to obtain an error coefficient mean; a deviation calculation microunit, the deviation calculation microunit is used to perform deviation calculation on multiple sample error coefficients in the sample error coefficient set based on the error coefficient mean, and determine multiple coefficient deviations; a standard error coefficient set construction microunit, the standard error coefficient set construction microunit is used to select sample error coefficients corresponding to coefficient deviations less than a predetermined deviation threshold to construct a standard error coefficient set, and perform mean calculation on the standard error coefficient set to obtain the first sample error coefficient.

[0053] Among them, the BP neural network is supervised and trained using the first sample training set to obtain a first error predictor, which is added to the multiple error predictors. The first error predictor acquisition unit further includes: an iterative selection subunit, the iterative selection subunit is used to divide the first sample training set into Q equal parts, and select Q times with replacement to construct a first training set, and iteratively select Q times using the same method to obtain Q training sets; a convergence error prediction branch acquisition subunit, the convergence error prediction branch acquisition subunit is used to use the sample environment parameters as input and the sample error coefficient as output, and use the Q training sets to perform supervised training and verification training on the BP neural network respectively to obtain Q convergence error prediction branches;

[0054] A first error predictor construction subunit is used to construct the first error predictor based on the Q convergence error prediction branches, wherein the output of the first error predictor is the average of the output results of the Q convergence error prediction branches.

[0055] The electric energy meter error prediction system based on big data analysis provided in the embodiment of the present invention can execute the electric energy meter error prediction method based on big data analysis provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0056] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, any number of different modules may be used and run on the user terminal and / or server, and the various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of the present invention.

[0057] The above specific implementation manner does not constitute a limitation to the protection scope of the present application. It should be understood by those skilled in the art that various modifications, combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present application should be included in the protection scope of the present application. In some cases, the actions or steps recorded in the present application can be performed in an order different from that in the embodiment and can still achieve the desired results. In addition, the process depicted in the accompanying drawings does not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

Claims

1. The electric energy meter error prediction method based on big data analysis is characterized by: include: Reading device attribute characteristics, voltage fluctuation threshold and power load threshold of the target electric energy meter; Dividing the voltage fluctuation threshold and the power load threshold according to a preset scalar, determining a plurality of voltage fluctuation intervals and a plurality of load ratio intervals, and constructing a plurality of power usage scenarios based on the plurality of voltage fluctuation intervals and the plurality of load ratio intervals; Taking the device attribute characteristics and multiple power usage scenarios as constraints, and taking the environmental impact on the electric energy meter error as a guide, a sample training set is obtained based on the power big data retrieval, and the BP neural network is supervised and trained using the sample training set to obtain multiple error predictors; Building an error prediction model based on the multiple power usage scenarios and multiple error predictors, inputting the real-time voltage fluctuation coefficient, the real-time power load ratio and the real-time environmental factor set into the error prediction model, and outputting the prediction error coefficient of the target electric energy meter; Build multiple electricity usage scenarios, including: The voltage fluctuation threshold is divided according to a first preset scalar, and the power load threshold is divided according to a second preset scalar to determine a plurality of voltage fluctuation intervals and a plurality of load ratio intervals; A first voltage fluctuation interval is randomly selected from the multiple voltage fluctuation intervals, a first load ratio interval is randomly selected from the multiple load ratio intervals, and the first voltage fluctuation interval and the first load ratio interval are integrated to construct a first power usage scenario, which is added to the multiple power usage scenarios.

2. The electric energy meter error prediction method based on big data analysis according to claim 1 is characterized in that: The device attribute characteristics include at least device type, specification characteristics, component accuracy, assembly method, factory calibration method and service life; The voltage fluctuation threshold is a voltage fluctuation coefficient threshold of the power supply grid of the target electric energy meter, wherein the voltage fluctuation coefficient is calculated based on the voltage fluctuation frequency and the voltage fluctuation amplitude; The power load threshold is a power load ratio threshold of the power equipment of the target electric energy meter.

3. The electric energy meter error prediction method based on big data analysis according to claim 2 is characterized in that: Taking the device attribute characteristics and multiple power usage scenarios as constraints, and taking the environmental impact on the electric energy meter error as a guide, a sample training set is obtained based on the power big data retrieval, and the BP neural network is supervised and trained using the sample training set to obtain multiple error predictors, including: According to the environmental factors affecting the electric energy meter error, the environmental factors at least include environmental temperature, environmental humidity, magnetic field strength and vibration impact characteristics; Randomly selecting a first power usage scenario from the multiple power usage scenarios, taking the device type, specification features, component accuracy, assembly method, factory calibration method, service life and the first power usage scenario as constraints, and taking the environmental influencing factors as a guide, obtaining a first sample training set based on power big data retrieval; The BP neural network is supervisedly trained using the first sample training set to obtain a first error predictor, which is added to the multiple error predictors.

4. The electric energy meter error prediction method based on big data analysis according to claim 3 is characterized in that: Get the first sample training set, including: Performing a similarity comparison on the search data according to the device type to obtain a first similarity; If the first similarity is greater than a predetermined similarity threshold, then the search data is sequentially compared for similarity according to the specification features, component accuracy, assembly method, factory calibration method, service life, and the first power usage scenario; if the feature comparison similarities all meet the predetermined similarity threshold, then the search data is added to the sample environment parameter set; Randomly select a first sample environmental parameter from the sample environmental parameter set, and obtain a sample error coefficient set based on power big data retrieval with the device type, specification characteristics, component accuracy, assembly method, factory calibration method, service life, first power usage scenario, and first sample environmental parameter as constraints; Performing high-frequency feature analysis on the sample error coefficient set, and outputting a first sample error coefficient; First sample training data is constructed according to the first sample environmental parameter and the first sample error coefficient, and added to the first sample training set.

5. The electric energy meter error prediction method based on big data analysis according to claim 4 is characterized in that: Performing high-frequency feature analysis on the sample error coefficient set and outputting a first sample error coefficient includes: Calculating the mean of the sample error coefficient set to obtain the error coefficient mean; Based on the error coefficient mean, deviation calculations are performed on a plurality of sample error coefficients in the sample error coefficient set to determine a plurality of coefficient deviations; The sample error coefficients corresponding to the coefficient deviations less than a predetermined deviation threshold are selected to construct a standard error coefficient set, and the mean of the standard error coefficient set is calculated to obtain the first sample error coefficient.

6. The electric energy meter error prediction method based on big data analysis according to claim 4 is characterized in that: Construct a first error predictor, including: The first sample training set is divided into Q equal parts, and Q parts are selected with replacement to construct a first training set, and the same method is used to iteratively select Q parts to obtain Q training sets; Taking the sample environment parameters as input and the sample error coefficient as output, the BP neural network is supervised and verified by using the Q training sets to obtain Q convergence error prediction branches; The first error predictor is constructed based on the integration of the Q convergence error prediction branches, wherein the output of the first error predictor is the average of the output results of the Q convergence error prediction branches.

7. The electric energy meter error prediction system based on big data analysis is characterized by: The system is used to implement the electric energy meter error prediction method based on big data analysis according to any one of claims 1 to 6, and the system comprises: A feature reading module, the feature reading module is used to read the device attribute characteristics, voltage fluctuation threshold and power load threshold of the target electric energy meter; A load threshold division module, the load threshold division module is used to divide the voltage fluctuation threshold and the power load threshold according to a preset scalar, determine multiple voltage fluctuation intervals and multiple load ratio intervals, and construct multiple power consumption scenarios based on the multiple voltage fluctuation intervals and multiple load ratio intervals; An error predictor acquisition module, the error predictor acquisition module is used to obtain a sample training set based on power big data retrieval, using the device attribute characteristics and multiple power usage scenarios as constraints, and the environmental impact of the electric energy meter error as a guide, and supervise the BP neural network with the sample training set to obtain multiple error predictors; A prediction error coefficient output module, wherein the prediction error coefficient output module is used to build an error prediction model based on the multiple power usage scenarios and multiple error predictors, input the real-time voltage fluctuation coefficient, the real-time power load ratio and the real-time environmental factor set into the error prediction model, and output the prediction error coefficient of the target electric energy meter.

Images