Building electric energy meter energy consumption acquisition error identification and correction method

Abstract

The invention discloses a building electric energy meter energy consumption acquisition error identification and correction method, which is characterized in that an error identification and correction model is established by improving a BP neural network, the influence of manual intervention is eliminated, and the correction of loss data of power equipment is realized. Meanwhile, the models are refined and applied to power sub-meter equipment and general meter equipment, and a sub-meter equipment BP neural network model and a general meter equipment BP neural network model are established. The method comprises the following steps: firstly, correcting the deviation of sub-table equipment data through an improved BP neural network, adding the sub-table data according to lines to obtain total table data, and then correcting the deviation of the total table equipment data through the improved BP neural network to further reduce the error. In addition, data collected by different devices are processed in a unified mode through a data standardization method, the influence of different device collection modes is reduced, and finally error correction of the energy consumption data is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of energy management, and is particularly applicable to a method for identifying and correcting energy consumption collection errors of building electric energy meters. Background Art

[0002] The need for high-precision energy management also places higher demands on the accuracy of energy metering data. Regarding the metering of electric energy, power equipment, such as transformers, capacitors, and power lines will inevitably produce power losses. How to accurately measure the use of electric energy has always been an important direction for improving the efficiency of electric energy.

[0003] In the existing literature, "Degradation Trend Prediction of Smart Electricity Meters Based on IGA-BP Neural Network" proposed a research method for the degradation trend of smart electricity meters using an improved genetic algorithm to optimize the BP neural network model (IGA-BP), realizing the backward prediction of the degradation trend of smart electricity meters and improving the metering accuracy of electricity meters. However, this method requires manual setting of the error threshold, which is subjective and will affect the final error detection results.

[0004] The paper "Error State Assessment Model of Smart Energy Meters in Low Voltage Distribution Areas Based on Improved BP Neural Network" aims at the error estimation problem of smart energy meters in distribution network areas and proposes a smart energy meter error estimation method based on particle swarm optimization BP neural network. However, this method only analyzes the error and does not conduct an in-depth analysis of error correction.

[0005] Therefore, there is a need for a relatively objective method for backward prediction of the error of the electric energy meter, as well as a method for intelligently correcting the measurement error of the electric energy meter. Summary of the invention

[0006] The present invention aims to provide a method for identifying and correcting energy consumption collection errors of building electric energy meters, aiming to solve the problems of prediction and deviation correction of electric energy meter measurement errors.

[0007] To achieve the above object, the present invention adopts the following technical solutions: The method for identifying and correcting energy consumption collection errors of building electric energy meters described in the present invention comprises the following steps: S1, respectively establish BP neural network models of the electric power equipment sub-meter and the total meter, and initialize the weight and threshold of each node unit of each layer of the BP neural network model; S2, when training the BP neural network model, the input forward propagation and error back propagation method are adopted, and the weight and threshold of each node unit in each layer of the BP neural network model are optimized and updated in combination with the elastic gradient descent method, so as to obtain the BP neural network error prediction model of the power equipment sub-meter and the total meter respectively; S3, using the output result of the prediction model to correct the data of the electric power equipment sub-meter and the main meter.

[0008] Furthermore, the input forward propagation is to calculate the net input and net output of the error data input into the BP neural network model in each node unit of each layer.

[0009] Furthermore, the error back propagation is to calculate the error of each node unit in each layer based on the net input and net output of each node unit in each layer.

[0010] Furthermore, in step S2, the data of the power equipment sub-meters are corrected using the output results of the BP neural network error prediction model of the power equipment sub-meters to obtain the power equipment sub-meter data; at the same time, the data of each power equipment sub-meter and the line error are added to obtain the difference between the power equipment total meter data and the actual reading of the power equipment total meter as the input of the BP neural network model of the power equipment total meter.

[0011] Furthermore, the input data of the BP neural network model includes the error Ti of the electric energy meter itself, the combined error Ta of the transformer and the voltage drop error Tc of the secondary circuit of the voltage transformer.

[0012] The advantage of the present invention is that it goes deep into the energy system, establishes an error identification and correction model by improving the BP neural network, gets rid of the influence of human intervention, and realizes the correction of the loss data of the power equipment. At the same time, the present invention refines the model and applies it to the power sub-meter equipment and the main meter equipment, and establishes the BP neural network model of the sub-meter equipment and the BP neural network model of the main meter equipment. First, the sub-meter equipment data is corrected by improving the BP neural network, and the sub-meter data is summed up by line to obtain the main meter data, and then the main meter equipment data is corrected by improving the BP neural network, and the error is further reduced. The present invention also uses a data standardization method to uniformly process the data collected by different devices, reduce the influence of different equipment collection methods, and finally realize the error correction of energy consumption data. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 It is a schematic diagram of the process of the method of the present invention.

[0014] Figure 2 It is a schematic diagram of the BP neural network model structure of the method of the present invention. DETAILED DESCRIPTION

[0015] The technical solutions in the embodiments of the present invention are described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0016] Example 1 like Figure 1 As shown, the method for identifying and correcting energy consumption collection errors of building electric energy meters of the present invention comprises the following steps: S1, respectively establish BP neural network models of the electric power equipment sub-meter and the total meter, and initialize the weight and threshold of each node unit in each layer of the BP neural network model.

[0017] The input layer of the BP neural network model is the measurement error of the electric energy metering device, including the error Ti of the electric energy meter itself, the synthetic error Ta of the transformer and the voltage drop error Tc of the secondary circuit of the voltage transformer. The present invention uses 3 nodes to represent the input data, which is X=(x1, x2, x3) T The inventors determined the number of node units in the hidden layer of the BP neural network model through an empirical formula. The empirical formula is as follows: Among them, N is the number of input nodes and M is the number of output nodes.

[0018] The output layer of the BP neural network model mainly obtains the predicted value of the electric energy meter measurement error, which is used to compare with the actual measurement value of the electric energy metering device and perform outlier correction.

[0019] like Figure 2 The figure shows the BP neural network model structure schematic diagram of the electric power equipment sub-meter and the main meter of the present invention.

[0020] Among them, x1 …x j …x m for The input of the input layer. The connection weight between the input layer and the hidden layer is W ij , which means the input layer j nodes to the hidden layer i The connection weight from the hidden layer to the output layer is W ki , represents the hidden layer i Nodes to the output layer k The weights between nodes are i=1,…,q,q is the total number of node units in the hidden layer. The output threshold of each unit in the hidden layer is θ i , represents the hidden layer i The output threshold of the output layer is a k , represents the threshold of the kth node unit in the output layer, k = 1,…,L,L is the total number of node units in the output layer. The activation function of the hidden layer is ф (x) , the activation function of the output layer is Ψ(x) O1, O k , O L is the output of the output layer.

[0021] S2, when training the BP neural network model, the input forward propagation and error back propagation method are adopted, and the weights and thresholds of each node unit in each layer of the BP neural network model are optimized and updated in combination with the elastic gradient descent method, and the BP neural network error prediction models of the power equipment sub-meter and the total meter are obtained respectively.

[0022] The input forward propagation is to calculate the net input and net output of the error data in the input BP neural network model at each node unit in each layer. The process of the hidden layer input forward propagation in the BP neural network model of the present invention can be expressed as: in, net i represents the hidden layer i The input of the node, y i represents the hidden layer i The output of each node. θ j Modified to θ i ，y 1 Modified to y i ) The process of forward propagation of the output layer in the BP neural network model of the present invention can be expressed as: in, O k Represents the output layer k The output of a node.

[0023] According to this process, for each node unit in the hidden layer or output layer j , relative to the previous layer node unit i , computing node unit j The net input is: The net output is: The error back propagation is to calculate the error of each node unit in each layer according to the net input and net output of each node unit in each layer.

[0024] First, the error energy function is the sum of squares of the error between the output vector of the output layer node of the sample and the actual measured value: The process of error back propagation is that the output layer calculates the error value of the neuron output of each layer layer by layer, and then modifies the weight and threshold of each layer according to the error gradient descent method, so that the final output of the network is as close to the expected value as possible.

[0025] Correction of the output layer weights , the correction amount of the output layer threshold Hidden layer weight correction , the correction value of hidden layer threshold .in, η It is called the network learning rate and satisfies 0<η<1.

[0026] According to the specific application of the present invention in the electric power field, the measurement error data of the electric energy metering device has the characteristics of randomness, instability, and nonlinear distribution. These data characteristics can easily cause the traditional BP neural network to oscillate during the learning process, making it difficult to obtain the global optimal solution, and the prediction results are largely limited by the gradient of the weight adjustment. The elastic gradient descent method introduces a momentum term to reflect the previous accumulation of weights, and the adjustment of weights is related to the update value. The gradient only affects the adjustment direction of the weights, which can effectively avoid the problem of slow network convergence caused by the gradient size, and well meet the characteristics of the measurement error data of the electric energy metering device.

[0027] The present invention combines the elastic gradient descent method to optimize and update the weights and thresholds of each node unit in each layer of the BP neural network model. During the training process of the BP neural network model, if the directions of consecutive iterations are inconsistent, the change in the weights is reduced, and the reduction factor is F J , if the gradient direction of consecutive iterations is consistent, then the change in weight is increased, and the increment factor is F When the gradient g(k) When it is 0, the weight change does not change, and the formula is: The relationship between weight and update value is: The corresponding ij The weight adjustment formula at +1 moment is: The present invention divides the sample data set into a training data set and a verification data set. When the number of training cycles reaches a predetermined requirement or the error accuracy of the model training result reaches a preset requirement, the training is stopped to obtain the BP neural network error prediction model of each power equipment sub-meter and the total meter.

[0028] S3, using the output result of the prediction model to correct the data of the electric power equipment sub-meter and the main meter.

[0029] It should be emphasized that in the present invention, the data of the electric power equipment sub-meter is corrected according to the output results of the BP neural network error prediction model of the electric power equipment sub-meter. After obtaining the corrected electric power equipment sub-meter data, the corrected electric power equipment sub-meter data are added and summed, and then added to the line error to obtain the electric power equipment total meter data. The difference between the electric power equipment total meter data and the actual reading of the electric power equipment total meter will be used as the input data of the BP neural network model of the electric power equipment total meter, and participate in the training of the BP neural network model of the electric power equipment total meter, and the BP neural network error prediction model of the electric power equipment total meter is obtained, which is used to correct the data of the electric power equipment total meter.

[0030] According to the method for identifying and correcting the energy consumption collection error of the building electric energy meter described in the present invention, the electric energy error data and the data of each error source measured by the main meter and the standby meter of the metering device in a certain year are used for experiments. The data of the first quarter, the second quarter and the third quarter are used as the training sample set, and the data of the fourth quarter are used as the test sample set. The training parameters mainly include the network learning rate, the number of training steps and the training target. The comprehensive error is predicted, and the difference between the predicted value and the measured value of the prediction model of the main meter and the standby meter is calculated. In the prediction process, the elastic gradient descent method is used to optimize the neural network, and the abnormal error value generated in the metering process is corrected, thereby reducing the error caused by the electric energy metering device and improving the accuracy of electric energy metering.

Claims

1. A method for identifying and correcting energy consumption collection errors of building electric energy meters, characterized in that: The following steps are involved: S1, respectively establish BP neural network models of the electric power equipment sub-meter and the total meter, and initialize the weight and threshold of each node unit of each layer of the BP neural network model; S2, when training the BP neural network model, the input forward propagation and error back propagation method are adopted, and the weight and threshold of each node unit in each layer of the BP neural network model are optimized and updated in combination with the elastic gradient descent method, so as to obtain the BP neural network error prediction model of the power equipment sub-meter and the total meter respectively; S3, using the output result of the prediction model to correct the data of the electric power equipment sub-meter and the main meter.

2. A method for identifying and correcting energy consumption collection errors of building electric energy meters according to claim 1, characterized in that: The input forward propagation is to calculate the net input and net output of the error data input into the BP neural network model in each node unit of each layer.

3. A method for identifying and correcting energy consumption collection errors of building electric energy meters according to claim 2, characterized in that: The error back propagation is to calculate the error of each node unit in each layer according to the net input and net output of each node unit in each layer.

4. A method for identifying and correcting energy consumption collection errors of building electric energy meters according to claim 2, characterized in that: In step S2, the data of the power equipment sub-meters are corrected using the output results of the BP neural network error prediction model of the power equipment sub-meters to obtain the power equipment sub-meter data; at the same time, the data of each power equipment sub-meter is added to the line error to obtain the difference between the power equipment total meter data and the actual reading of the power equipment total meter as the input of the BP neural network model of the power equipment total meter.

5. A method for identifying and correcting energy consumption collection errors of building electric energy meters according to claim 1, characterized in that: The input data of the BP neural network model include the error Ti of the electric energy meter itself, the synthetic error Ta of the transformer and the voltage drop error Tc of the secondary circuit of the voltage transformer.

Images