CN113485986B - Electric power data restoration method - Google Patents

Abstract

The invention discloses a power data restoration method, which utilizes an SOM neural network to classify and process power data in a historical power data set; obtaining influence factors of the electric power data types meeting the relevance threshold by using the Pearson correlation coefficient theory; inputting influencing factors of the missing data into a trained LSTM neural network to obtain the electric power data type of the missing data; and repairing the data by adopting different methods according to the type of the power data of the missing data. According to the invention, the complex nonlinearity of the power data is considered, the characteristic of nonlinear problem can be processed by utilizing the strong learning ability of the neural network, the restoration of the power data is realized, and the classification efficiency and accuracy can be effectively improved.

Description

A method for repairing power data

Technical Field

The invention relates to a power data repair method, belonging to the technical field of power data detection and repair.

Background technique

The continuous development and progress of digital technology has generated a large amount of power data in the power system. However, power data is often missing due to external interference, transmission errors, equipment abnormalities, network delays, etc., which will affect the accuracy and timeliness of data processing in the power system.

Existing repair methods mostly use traditional machine learning methods. However, faced with the increasingly complex power grid structure, especially the access of new energy power generation systems and electric vehicles, and the application of demand response mechanisms, traditional machine learning methods are unable to cope with the repair of highly random power data.

In order to ensure the optimized and stable operation of the intelligent power system, complete and correct power data is needed to provide support to avoid affecting the safe and stable operation of the power system. Technical personnel in this field urgently need to repair the data of the power system with higher accuracy.

Summary of the invention

Purpose: In order to overcome the deficiencies in the prior art, the present invention provides a method for repairing power data.

Technical solution: To solve the above technical problems, the technical solution adopted by the present invention is:

A method for repairing power data comprises the following steps:

Step S1, obtaining a historical power data set, and using a SOM neural network to classify the power data in the historical power data set to obtain a power data type.

Step S2: Use the Pearson correlation coefficient theory to perform correlation analysis on the power data type and the influencing factors, obtain the influencing factors of the power data type that meet the correlation threshold, and use the influencing factors corresponding to the power data type as feature values.

Step S3: Use the power data type and the corresponding characteristic value as training samples to train the LSTM neural network to obtain a trained LSTM neural network.

Step S4: Input the influencing factors of the missing data into the trained LSTM neural network to obtain the power data type of the missing data.

Step S5: Use different methods to repair the data according to the type of power data of the missing data.

As a preferred solution, the SOM neural network consists of two layers, an input layer and an output layer, and the input layer and the output layer are fully connected. When classifying the input power data, the neurons in the input layer work together to compete for the response opportunities of the input power data, thereby obtaining the output neurons. By updating the weights, the weights around the neurons in the output layer will be adjusted accordingly. After the power data is adjusted, each neuron in the input layer learns a specific category, thereby realizing the learning and classification of the input power data by the output layer. The weight value range is usually [0,1].

As a preferred solution, the Pearson correlation coefficient theory is calculated as follows:

Where: ρ _XY is the correlation coefficient, ranging from [0,1]; X, Y are continuous variables; n is the number of continuous variable samples; and are the average values of continuous variables respectively; ρ _XY = 0, indicating that the variables are independent of each other; ρ _XY > 0, indicating that the variables are positively correlated; ρ _XY < 0, indicating that the variables are negatively correlated; the larger the |ρ _X,Y |, the stronger the correlation between the variables, and vice versa.

As a preferred solution, the LSTM neural network is constructed as follows:

The LSTM neural network includes three gate structures to realize the prediction of power data type, namely forget gate, input gate and output gate.

(1) Forget gate. It is mainly used to filter input data and calculate the degree of data retention. After processing through the sigmoid neural layer, it outputs a value in the range of [0,1]. The larger the value, the more components are retained, and vice versa. The calculation formula of the forget gate is as follows:

f _t =σ(W _xf x _t +W _hf h _t-1 +B _f )

Where: _ft is the output of the forget gate; _Wxf and _Whf are the network parameters that the forget gate needs to learn; _Bf is the bias of the forget gate; σ is the sigmoid function:

(2) Input gate. It is mainly used to update the data state. It mainly includes two parts of information. One part is the data to be saved selected by the sigmoid function; the other part is the new information generated by the tanh function for the current input _xt . _C′t combines the two parts of information to generate a new memory state. The calculation formula is as follows:

i _t =δ(W _xi x _t +W _hi h _t-1 +B _i )

C′ _t =tanh(W _xC x _t +W _hC h _t-1 +B _C )

C _t =f _t ×C _t-1 +i _t ×C′ _t

Where: _it is the output of the input gate; _Wxi and _Whi are the network parameters that need to be learned by the input gate; _Bi is the bias of the input gate; _Ct ′ is the input after the tanh function is updated; _Wxc and _Whc are the network parameters that need to be learned for the cell state; _Bc is the bias of the cell state; _Ct is the cell state.

(3) Output gate. It is mainly used to calculate the output level of input information _xt . The information output in the cell state is multiplied by the sigmoid function after passing through the tanh function to obtain the final hidden unit _ht . The calculation formula is as follows:

o _t =σ(W _xo x _t +W _ho h _t-1 +B _o )

h _t = o _t × tanh(C _t )

Where: _Ot is the output of the output gate; _ht is the hidden unit of the output; _Wxo and _Who are the network parameters of the output gate that need to be learned; _Bo is the bias of the output gate.

As a preferred solution, the method of repairing the data using different methods according to the type of power data of the missing data includes the following steps:

When the type of power data with missing data is electric energy data, the stratified mean filling method is used to stratify the electric energy data according to the same month of each year, and the average value of each layer is used to replace the missing value.

When the type of power data with missing data is load data, the maximum expectation algorithm is used to perform maximum likelihood estimation on the missing data, and iterates continuously to finally output the value to replace the missing value.

When the type of power data with missing data is current data, voltage data or frequency data, the mean filling method is used to replace the missing value with the average of the two previous and next observations or the mean of the non-missing data.

Beneficial effect: The present invention provides a method for repairing power data. This method takes into account the complex nonlinearity of power data, utilizes the strong learning ability of neural networks, and can handle the characteristics of nonlinear problems to achieve power data repair, which can effectively improve classification efficiency and accuracy.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a schematic flow chart of the method of the present invention.

FIG2 is a performance accuracy comparison chart of the SOM-LSTM neural network classification model provided by an embodiment of the present invention.

FIG3 is a performance accuracy comparison chart of a SOM-LSTM neural network classification model provided by another embodiment of the present invention.

Detailed ways

The present invention will be further described below in conjunction with specific embodiments.

The following is a further detailed description of a power data repair method proposed in the present invention in combination with the accompanying drawings and specific embodiments. According to the following description, the advantages and features of the present invention will be clearer. It should be noted that the drawings are in a very simplified form and use non-precise proportions, which are only used to conveniently and clearly assist in explaining the purpose of the embodiments of the present invention. In order to make the purposes, features and advantages of the present invention more obvious and easy to understand, please refer to the accompanying drawings. It should be noted that the structure, proportion, size, etc. illustrated in the drawings of this specification are only used to match the contents disclosed in the specification for people familiar with this technology to understand and read, and are not used to limit the limiting conditions for the implementation of the present invention, so it has no technical substantive significance. Any modification of the structure, change in the proportional relationship or adjustment of the size, without affecting the effects that the present invention can produce and the purposes that can be achieved, should still fall within the scope of the technical content disclosed in the present invention.

It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

As shown in FIG. 1 , the present embodiment provides a method for repairing power data, comprising the following steps:

Step S1, obtaining a historical power data set, and using a SOM neural network to classify the power data in the historical power data set to obtain a power data type.

Step S2: Use the Pearson correlation coefficient theory to perform correlation analysis on the power data type and the influencing factors, obtain the influencing factors of the power data type that meet the correlation threshold, and use the influencing factors corresponding to the power data type as feature values.

Step S3: Use the power data type and the corresponding characteristic value as training samples to train the LSTM neural network to obtain a trained LSTM neural network.

Step S4: Input the influencing factors of the missing data into the trained LSTM neural network to obtain the power data type of the missing data.

Step S5: Use different methods to repair the data according to the type of power data of the missing data.

The SOM neural network consists of two layers, the input layer and the output layer, and the input layer and the output layer are fully connected. When classifying the input power data, the neurons in the input layer work together to compete for the response opportunities of the input power data, thereby obtaining the output neurons. By updating the weights, the weights around the neurons in the output layer will be adjusted accordingly. After the power data is adjusted, each neuron in the input layer learns a specific category, thereby realizing the learning and classification of the input power data in the output layer. The weight value range is usually [0,1].

The Pearson correlation coefficient theory, the calculation formula is as follows:

Where: ρ _XY is the correlation coefficient, ranging from [0,1]; X, Y are continuous variables; n is the number of continuous variable samples; and are the average values of continuous variables respectively; ρ _XY = 0, indicating that the variables are independent of each other; ρ _XY > 0, indicating that the variables are positively correlated; ρ _XY < 0, indicating that the variables are negatively correlated; the larger the |ρ _X,Y |, the stronger the correlation between the variables, and vice versa.

The LSTM neural network is constructed as follows:

The LSTM neural network includes three gate structures to realize the prediction of power data type, namely forget gate, input gate and output gate.

(1) Forget gate. It is mainly used to filter input data and calculate the degree of data retention. After processing through the sigmoid neural layer, it outputs a value in the range of [0,1]. The larger the value, the more components are retained, and vice versa. The calculation formula of the forget gate is as follows:

f _t =σ(W _xf x _t +W _hf h _t-1 +B _f )

Where: _ft is the output of the forget gate; _Wxf and _Whf are the network parameters that the forget gate needs to learn; _Bf is the bias of the forget gate; σ is the sigmoid function:

(2) Input gate. It is mainly used to update the data state. It mainly includes two parts of information. One part is the data to be saved selected by the sigmoid function; the other part is the new information generated by the tanh function for the current input _xt . _C′t combines the two parts of information to generate a new memory state. The calculation formula is as follows:

i _t =δ(W _xi x _t +W _hi h _t-1 +B _i )

C′ _t =tanh(W _xC x _t +W _hC h _t-1 +B _C )

C _t =f _t ×C _t-1 +i _t ×C′ _t

Where: _it is the output of the input gate; _Wxi and _Whi are the network parameters that need to be learned by the input gate; _Bi is the bias of the input gate; _Ct ′ is the input after the tanh function is updated; _Wxc and _Whc are the network parameters that need to be learned for the cell state; _Bc is the bias of the cell state; _Ct is the cell state.

(3) Output gate. It is mainly used to calculate the output level of input information _xt . The information output in the cell state is multiplied by the sigmoid function after passing through the tanh function to obtain the final hidden unit _ht . The calculation formula is as follows:

o _t =σ(W _xo x _t +W _ho h _t-1 +B _o )

h _t = o _t × tanh(C _t )

Where: _Ot is the output of the output gate; _ht is the hidden unit of the output; _Wxo and _Who are the network parameters of the output gate that need to be learned; _Bo is the bias of the output gate.

The method of repairing the data using different methods according to the type of power data of the missing data includes the following steps:

When the type of power data with missing data is electric energy data, the stratified mean filling method is used to stratify the electric energy data according to the same month of each year, and the average value of each layer is used to replace the missing value.

When the type of power data with missing data is load data, the maximum expectation algorithm is used to perform maximum likelihood estimation on the missing data, and iterates continuously to finally output the value to replace the missing value.

When the type of power data with missing data is current data, voltage data or frequency data, the mean filling method is used to replace the missing value with the average of the two previous and next observations or the mean of the non-missing data.

As shown in Figures 2 and 3, in order to verify the effectiveness of the proposed SOM-LSMT-based power data repair method, the computer configuration of the experimental simulation in this paper is Inter Intel(R)Core(TM)i5-8300H CPU@2.30GHz2.30GHz and NVIDIAGTX 1050Ti 4G video memory, and the simulation calculation is carried out on the MATLAB R2019a platform.

This paper selects household electricity data from a certain area in China in August 2018 and temperature data collected by the corresponding meteorological station, with a data collection interval of 15 minutes. The mean absolute error (MAE) and root mean square error (RMSE) evaluation indicators are used to analyze the power data repair results.

In order to analyze the accuracy of the proposed SOM-LSTM model, this paper draws curves to compare the repair results of the SOM-LSTM model, the Extreme Learning Machine (ELM) model, and the LSTM model for continuous missing power data on typical working days and non-working days. The results are shown in Figures 2 and 3.

As shown in Figures 2 and 3, when repairing the continuous missing voltage data of users on typical working days and non-working days, the LSTM neural network has the largest error, and as the data missing increases, the accuracy decreases significantly, indicating that the fitting and repair effect of the neural network is insufficient. Compared with the LSTM model and the ELM model, the SOM-LSTM neural network that predicts and repairs the data after classification by the SOM neural network can effectively improve the problem of low repair accuracy of continuous missing data, and still maintain a high level of repair rate when there is a lot of data missing.

Although the content of the present invention has been described in detail through the above preferred embodiments, it should be appreciated that the above description should not be considered as a limitation of the present invention. After reading the above content, it will be apparent to those skilled in the art that various modifications and substitutions of the present invention will occur. Therefore, the protection scope of the present invention should be limited by the appended claims.

Claims (7)

1. A method for repairing power data, characterized in that it comprises the following steps: Step S1, obtaining a historical power data set, and using a SOM neural network to classify the power data in the historical power data set to obtain a power data type; Step S2: using the Pearson correlation coefficient theory, perform a correlation analysis on the power data type and the influencing factors, obtain the influencing factors of the power data type that meet the correlation threshold, and use the influencing factors corresponding to the power data type as feature values; Step S3: Use the power data type and the corresponding characteristic value as training samples to train the LSTM neural network to obtain a trained LSTM neural network; Step S4, input the influencing factors of the missing data into the trained LSTM neural network to obtain the power data type of the missing data; Step S5: Use different methods to repair the data according to the type of power data of the missing data. 2. A power data repair method according to claim 1, characterized in that: the SOM neural network consists of two layers, an input layer and an output layer, and the input layer and the output layer are fully connected; when classifying the input power data, the neurons in the input layer work together to compete for the response opportunities of the input power data, thereby obtaining output neurons; by updating the weights, the weights around the neurons in the output layer will be adjusted accordingly, and after the power data is adjusted, each neuron in the input layer learns a specific category, thereby realizing the learning and classification of the input power data by the output layer. 3. A power data repair method according to claim 2, characterized in that: the weight value range is [0, 1]. 4. A method for repairing power data according to claim 1, characterized in that: the Pearson correlation coefficient theory is calculated by the following formula: Where: ρ _{X, Y} are correlation coefficients, ranging from [0,1]; X, Y are continuous variables; n is the number of continuous variable samples; and are the average values of continuous variables respectively; ρ _{X, Y} = 0, indicating that the variables are independent; ρ _{X, Y} > 0, indicating that the variables are positively correlated; ρ _{X, Y} < 0, indicating that the variables are negatively correlated; the larger the |ρ _{X, Y} |, the stronger the correlation between the variables, and vice versa. 5. A power data repair method according to claim 1, characterized in that: the LSTM neural network includes three gate structures to realize the prediction of power data type, namely a forget gate, an input gate and an output gate. 6. A method for repairing power data according to claim 5, characterized in that: the calculation formula of the forget gate is as follows: f _t =σ(W _xf x _t +W _hf h _t-1 +B _f ) Where: _ft is the output of the forget gate; _Wxf and _Whf are the network parameters that the forget gate needs to learn; _Bf is the bias of the forget gate; σ is the sigmoid function: The input gate calculation formula is as follows: i _t =δ(W _xi x _t +W _hi h _t-1 +B _i ) C _t ^′ =tanh(W _xC x _t +W _hC h _t-1 +B _C ) C _t =f _t ×C _t-1 +i _t ×C _t ^′ Where: _it is the output of the input gate; _Wxi and _Whi are the network parameters that need to be learned by the input gate; _Bi is the bias of the input gate; _Ct ′ is the input after the tanh function is updated; _WxC and _WhC are the network parameters that need to be learned for the cell state; _Bc is the bias of the cell state; _Ct is the cell state; The output gate calculation formula is as follows: o _t =σ(W _xo x _t +W _ho h _t-1 +B _o ) h _t = o _t × tanh(C _t ) Where: o _t is the output of the output gate; h _t is the hidden unit of the output; W _xo and _Who are the network parameters of the output gate that need to be learned; B _o is the bias of the output gate. 7. A method for repairing power data according to claim 1, characterized in that: the method for repairing data using different methods according to the type of power data of the missing data comprises the following steps: When the type of power data with missing data is electric energy data, the stratified mean filling method is used to stratify the electric energy data according to the same month of each year, and the average value of each layer is used to replace the missing value; When the type of power data with missing data is load data, the maximum expectation algorithm is used to perform maximum likelihood estimation on the missing data, and the algorithm is iterated continuously to finally output the value to replace the missing value. When the type of power data with missing data is current data, voltage data or frequency data, the mean filling method is used to replace the missing value with the average of the two previous and next observations or the mean of the non-missing data.