User electricity consumption prediction method and system based on SSA-CCNN

Abstract

The invention belongs to the technical field of electricity consumption prediction, and provides an SSA-CCNN-based user electricity consumption prediction method and system, and the method comprises the steps: obtaining user historical electricity consumption data, and carrying out the preprocessing; decomposing the preprocessed historical electricity consumption data of the user into time sequences representing different characteristic components by adopting a singular spectrum analysis method; inputting the time sequences of the different feature components into a trained deep causal convolutional neural network model, and then obtaining predicted values of the different components; and superposing different predicted values to obtain a predicted value of the electricity consumption of the final user. The singular spectrum decomposition can effectively process the nonlinear relationship between the electricity consumption data and external factors, so that the feature extraction and analysis of the historical electricity consumption data of the user become more intuitive and easy to understand. The deep causal convolutional neural network is used to predict the power consumption of the user, the prediction error is smaller, the prediction result is more accurate, and the power market can make an appropriate power generation plan according to the prediction result to meet the normal power consumption demand of the user.

Description

Technical Field

[0001] The present invention belongs to the field of electricity consumption prediction, and in particular relates to a method and system for predicting user electricity consumption based on SSA-CCNN. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] By predicting user electricity consumption, it is not only convenient for the power market to formulate power generation plans and improve energy utilization, but also can reflect the power load situation and ensure the safe operation of the power system, which plays a vital role in the stable operation of the power system.

[0004] However, the user's electricity consumption data has certain volatility and is easily affected by external factors, such as holidays and temperature. Existing prediction models often cannot capture this mutation information. For example, the patent number CN116777049A provides a user electricity consumption prediction method and system, which builds a multi-scale electricity consumption prediction model based on the long short-term memory network LSTM and the mixed density network MDN. However, the long short-term memory network LSTM requires more parameters to train the model in large quantities, and it cannot effectively handle the nonlinear relationship between user electricity consumption data and external factors. The patent number CN114169645A discloses a smart grid short-term load prediction method. In the process of data processing, the local mean decomposition LMD is used to process the data, but the local mean decomposition LMD method is extremely sensitive to noise, and even very small noise may cause deviations in the decomposition results. There is often a lot of noise information in the user's electricity consumption data, and LMD cannot effectively separate the noise information in the user's electricity consumption data. Summary of the invention

[0005] In order to overcome the shortcomings of the above-mentioned prior art, the present invention provides a user electricity consumption prediction method and system based on SSA-CCNN, which uses singular spectrum analysis (SSA) to decompose the user's historical electricity consumption data, and inputs the decomposed different electricity consumption data feature components into a deep causal convolutional neural network (CCNN) to establish a prediction model to obtain the predicted values ​​of different data feature components, and obtain the predicted value of the user's final electricity consumption by superimposing the predicted values.

[0006] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions: The first aspect of the present invention provides a user power consumption prediction method based on SSA-CCNN; A user power consumption prediction method based on SSA-CCNN, comprising: Acquire historical electricity consumption data of users, and pre-process the historical electricity consumption data of users; The singular spectrum analysis method is used to decompose the preprocessed user historical electricity consumption data into time series of periodic components, trend components and noise components that characterize the characteristics of different users' historical electricity consumption data. The time series of the period component, trend component and noise component are input into the trained deep causal convolutional neural network model to obtain the period component prediction value, trend classification prediction value and noise component prediction value; The predicted value of the end-user electricity consumption is obtained by superimposing the periodic component prediction value, the trend classification prediction value and the noise component prediction value.

[0007] As a further technical solution, the process of preprocessing the user's historical electricity consumption data includes: data correction and normalization processing, using the missing value interpolation method to correct the abnormal data in the user's historical electricity consumption data, and using the min-miax method to normalize the corrected data. The min-miax formula is as follows:

[0008] In the formula, is the normalized data; is the original data; are the maximum and minimum values ​​respectively.

[0009] As a further technical solution, the specific process of using singular spectrum analysis to decompose the pre-processed user historical power consumption data into a time series of periodic components, trend components and noise components that characterize the characteristics of different user historical power consumption data is as follows: Embed the user's historical electricity consumption data into a trajectory matrix X, perform singular value solving on the trajectory matrix, and obtain the singular value and eigenvector of the trajectory matrix of the user's historical electricity consumption data; The eigenvectors after singular value decomposition are grouped according to the size of the singular values ​​to obtain the period component, trend component and noise component of the user's historical power consumption data trajectory matrix; The diagonal averaging method is used to reconstruct the periodic component, trend component and noise component of the trajectory matrix to obtain the time series of the periodic component, trend component and noise component.

[0010] As a further technical solution, the trajectory matrix X is,

[0011] Where M represents the window length, which ranges from 2 to N / 2. J is the number of columns of the trajectory matrix X; , N express The user's historical electricity consumption data at the moment.

[0012] As a further technical solution, the specific process of solving the singular value of the trajectory matrix to obtain the singular value and eigenvector of the user's historical power consumption data trajectory matrix is ​​as follows: Get the covariance matrix G of the trajectory matrix, G= ,in is the transposed matrix of X, and M singular values ​​and eigenvectors corresponding to the singular values ​​are obtained from the covariance matrix. Then the trajectory matrix X is:

[0013] In the formula, represents the principal components of singular value decomposition; Represented as the singular values ​​of the trajectory matrix; represents the left singular vector; represents the right singular vector; express Eigenvalues ​​in descending order.

[0014] As a further technical solution, the diagonal averaging method is as follows:

[0015] In the formula, , is the transformed one-dimensional time series.

[0016] As a further technical solution, the process of inputting the time series of the period component, trend component and noise component into the trained deep causal convolutional neural network model to obtain the period component prediction value, trend classification prediction value and noise component prediction value is as follows: The time series of periodic component, trend component and noise component are used as input to the deep causal convolutional neural network model; Capturing the long-term dependency of the time series of the periodic component, the trend component and the noise component through the first causal convolution layer and the second causal convolution layer in the deep causal convolutional neural network model; Using a pooling layer to reduce the length of the time series of the periodic component, the trend component and the noise component; The predicted values ​​of the time series of the periodic component, trend component and noise component are output through the fully connected layer.

[0017] The second aspect of the present invention provides a user power consumption prediction system based on SSA-CCNN; comprising: The data preprocessing module is configured to: obtain the user's historical electricity consumption data and preprocess the user's historical electricity consumption data; The data feature decomposition module is configured to: use a singular spectrum analysis method to decompose the pre-processed user historical power consumption data into a time series of a period component, a trend component and a noise component that characterize the characteristics of different user historical power consumption data; The user power consumption prediction module is configured to: input the time series of the period component, the trend component and the noise component into the trained deep causal convolutional neural network model, and then obtain the period component prediction value, the trend classification prediction value and the noise component prediction value; The predicted value of the end-user electricity consumption is obtained by superimposing the periodic component prediction value, the trend classification prediction value and the noise component prediction value.

[0018] The third aspect of the present invention provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the steps in a method for predicting user electricity consumption based on SSA-CCNN as described in the first aspect of the present invention.

[0019] The fourth aspect of the present invention provides an electronic device, including a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, the steps in a method for predicting user electricity consumption based on SSA-CCNN as described in the first aspect of the present invention are implemented.

[0020] One or more of the above technical solutions have the following beneficial effects: The present invention provides a user electricity consumption prediction method and system based on SSA-CCNN. By using singular spectrum decomposition to process the user's historical electricity consumption data, the nonlinear relationship between electricity consumption data and external factors can be effectively processed. By decomposing the user's historical electricity consumption data into multiple components, each component has a clear physical meaning, such as trend, periodicity and noise. The feature extraction and analysis of the user's historical electricity consumption data becomes more intuitive and easy to understand. By selecting representative components for analysis, the user's historical electricity consumption data can be better understood and processed.

[0021] A deep causal convolutional neural network model is used to predict the trend, cycle and noise components of users' historical electricity consumption data. The deep causal convolutional neural network contains two causal convolutional layers that can effectively capture the time dependency in the above data components, so as to better understand and predict the dynamic changes of the data. Its prediction error is smaller and the results are more accurate, which makes it easier for the power market to formulate appropriate power generation plans based on the above prediction results to meet users' normal electricity needs.

[0022] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0024] Figure 1 This is a flow chart of the method of the first embodiment.

[0025] Figure 2 It is a system structure diagram of the second embodiment. DETAILED DESCRIPTION

[0026] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0027] It should be noted that the terms used herein are for describing specific embodiments only and are not intended to be limiting of exemplary embodiments according to the present invention.

[0028] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0029] Embodiment 1 Time series data refers to a data sequence in which the observations of the same variable are arranged in chronological order. Electricity consumption data, such as customer energy consumption measured by smart meters, is usually recorded at a higher time resolution (such as every 5 minutes, 15 minutes, etc.), forming a large number of data sets. These data are arranged in chronological order and have characteristics such as periodicity, trend, seasonality, and randomness, which make electricity consumption data meet the definition and characteristics of time series.

[0030] By predicting the power consumption of users, it is convenient for power supply companies to formulate corresponding power supply plans. On the basis of meeting the power demand of users, it can also reasonably plan the power in the distribution network, ensure the safe operation of the power system, and further improve the utilization rate of energy. However, the power consumption data of users has certain volatility and is easily affected by factors such as holidays and temperature. Therefore, in this embodiment, in order to achieve accurate prediction of power consumption of users, combined with Figure 1 This embodiment discloses a method for predicting user power consumption based on SSA-CCNN, including: Step S1, obtaining historical electricity consumption data of a user, and preprocessing the historical electricity consumption data of the user; Step S2, using a singular spectrum analysis method to decompose the pre-processed user historical power consumption data into a time series of a period component, a trend component and a noise component that characterize the characteristics of different user historical power consumption data; Step S3, using a deep causal convolutional neural network model to process the time series of the period component, trend component and noise component respectively, and then obtain a period component prediction value, a trend classification prediction value and a noise component prediction value; Step S4, superimposing the period component prediction value, the trend classification prediction value and the noise component prediction value to obtain the prediction value of the final user's electricity consumption.

[0031] In step S1, the user's historical electricity consumption data is obtained through the smart meter, and the obtained user's historical electricity consumption data is preprocessed. The preprocessing process includes data correction and normalization. Due to machine failure and other reasons, the acquired user's historical electricity consumption data is prone to data missing and data anomalies. In this embodiment, the missing value interpolation method is used to correct the data for both data missing and data anomalies. The data correction is completed by averaging the two data before and after the abnormal data, and using the mean as a correction value to replace the abnormal data. Compared with the method of directly deleting abnormal data, the missing value interpolation method can maintain the reliability of the data to a large extent.

[0032] The corrected user historical electricity consumption data is normalized using the min-miax method. The min-miax formula is as follows:

[0033] In the formula, is the normalized data; is the original data; are the maximum and minimum values ​​respectively.

[0034] By normalizing the corrected historical electricity consumption data of users, historical electricity consumption data of users in different units and orders of magnitude are unified into the same scale, which facilitates comparison and analysis between data.

[0035] In step S2, the singular spectrum analysis method SSA can effectively decompose the time series to obtain the periodic component, trend component and noise component that characterize the characteristics of different sequences. The singular spectrum analysis method is divided into four steps: embedding, singular value decomposition, grouping and reconstruction. The decomposition stage uses the embedding and singular value decomposition methods to extract singular values ​​from the independent information contained in the original time series. In the reconstruction stage, the diagonal averaging method is used to process the matrix obtained in the previous stage, and it is restored to a one-dimensional reconstructed sequence and grouped. In this embodiment, the user's historical electricity consumption data is first represented in the form of a data set. ,The user historical electricity consumption data preprocessed by singular spectrum analysis method is decomposed into time series of period component, trend component and noise component that ,characterize the characteristics of different user historical electricity consumption data. ,The specific process is as follows; S21, embed the user's historical electricity consumption dataset into a trajectory matrix X,

[0036] Where M represents the window length, which ranges from 2 to N / 2. J is the number of columns of the trajectory matrix X; , N express The user's historical electricity consumption data at the moment; S22, obtain the covariance matrix G of the trajectory matrix X, where G= , is the transposed matrix of X, and M singular values ​​and eigenvectors corresponding to the singular values ​​are obtained from the covariance matrix. Then the trajectory matrix X can be further expressed as:

[0037] In the formula, represents the principal components of singular value decomposition; It is represented as the singular value of the user's historical electricity consumption data trajectory matrix; represents the left singular vector; represents the right singular vector; express Eigenvalues ​​in descending order; S23, grouping the eigenvectors after singular value decomposition according to the size of the singular value, and obtaining the period component, trend component and noise component of the trajectory matrix of the user's historical power consumption data; the specific process of grouping the trajectory matrix after singular value decomposition is: decomposing the principal components of the singular value Subscript Divide p Disjoint subsets ,set up ,but , calculate the set For each composite matrix of In the grouping process, it is necessary to accurately distinguish the component type represented by each singular value, that is, whether it belongs to a trend component, a periodic component, or a noise component. Since the size of the singular value represents the contribution of the eigenvector corresponding to the singular value to the user's historical electricity consumption data. That is, the larger the singular value, the higher the information contained in the corresponding eigenvector, that is, the larger the singular value, the more important the corresponding eigenvector is, and the less likely it is to be discarded.

[0038] In this embodiment, the changing trend of the user's historical electricity consumption data represented by the first singular value is taken as the trend component, and the remaining singular values ​​contain the periodic component and noise component of the user's historical electricity consumption data. Since many singular values ​​are not much different in size, it is difficult to effectively distinguish the periodic component from the noise component. Here, the periodic component of the user's historical electricity consumption data can be extracted through Fourier transform or synchronous compression transform, and the remaining component can be used as the noise component of the user's historical electricity consumption data.

[0039] S24, reconstructing the periodic component, trend component and noise component of the trajectory matrix by using the diagonal averaging method to obtain the time series of the periodic component, trend component and noise component, wherein the diagonal averaging is to average the diagonals of each sub-matrix, and the periodic component, trend component and noise component obtained in the previous step are restored to a new time series with a length of N.

[0040] The diagonal average method formula is:

[0041] In the formula, The diagonal averaging formula can be used to convert the periodic component, trend component and noise component into one-dimensional time series. .

[0042] Step S3, using the deep causal convolutional neural network model to process the time series of the period component, trend component and noise component respectively, and then obtain the period component prediction value, trend classification prediction value and noise component prediction value; wherein the deep causal convolutional neural network includes an input layer, a first causal convolutional layer, a pooling layer, a second causal convolutional layer, a maximum pooling layer, a Droprout layer and a fully connected layer. The time series of the period component, trend component and noise component are respectively input into the deep causal convolutional neural network model, and the first causal convolutional layer is implemented by a convolutional layer with a ReLU activation function. Specifically,

[0043] in, is the output of the first layer at time t, is the weight matrix of the first layer; For input at time The value of is the bias term of the first layer; is the time step considered; is the activation function.

[0044] The first causal convolutional layer can capture the long-term dependencies of the time series of periodic components, trend components and noise components.

[0045] After being processed by the first causal convolutional layer, the relevant data is transmitted to the pooling layer, which is used to reduce the length of the time series of the periodic component, trend component and noise component, thereby simplifying the complexity of the model and helping to learn higher-level features.

[0046] After processing by the pooling layer, the above data is transmitted to the second causal convolutional layer. The second causal convolutional layer is also implemented by a convolutional layer with a ReLU activation function, but uses a smaller convolution kernel to capture patterns in a shorter range. Its expression is:

[0047] in, is the output of the second layer at time t, is the weight matrix of the second layer; For the first time in time The value of is the bias term of the second layer; is the time step considered.

[0048] Similarly, the data processed by the second causal convolution layer is transmitted to the maximum pooling layer. Through another pooling process, the length of the time series data can be reduced again, and higher-level features can be further extracted. Then, the Dropout layer is used to randomly close some neurons during the training process to prevent the problem of overfitting of the model.

[0049] Finally, the predicted values ​​of the time series of periodic component, trend component and noise component are output through the fully connected layer. In step S4, the period component prediction value, trend component prediction value and noise component prediction value obtained by using the deep causal convolutional neural network model are superimposed to obtain the prediction value of the final user's electricity consumption.

[0050] Embodiment 2 This embodiment discloses a system for predicting user electricity consumption during holidays; like Figure 2 As shown, a user power consumption prediction system based on SSA-CCNN; comprising: The data preprocessing module is configured to: obtain the user's historical electricity consumption data and preprocess the user's historical electricity consumption data; The data feature decomposition module is configured to: use a singular spectrum analysis method to decompose the pre-processed user historical power consumption data into a time series of a period component, a trend component and a noise component that characterize the characteristics of different user historical power consumption data; The user power consumption prediction module is configured to: use a deep causal convolutional neural network model to process the time series of the period component, the trend component and the noise component respectively, and then obtain the period component prediction value, the trend classification prediction value and the noise component prediction value; The predicted value of the end-user electricity consumption is obtained by superimposing the periodic component prediction value, the trend classification prediction value and the noise component prediction value.

[0051] Embodiment 3 The purpose of this embodiment is to provide a computer-readable storage medium.

[0052] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps in a method for predicting user electricity consumption based on SSA-CCNN as described in Example 1 of the present disclosure.

[0053] Embodiment 4 The purpose of this embodiment is to provide an electronic device.

[0054] An electronic device includes a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, the steps in a method for predicting user electricity consumption based on SSA-CCNN as described in Example 1 of the present disclosure are implemented.

[0055] The steps involved in the apparatuses of the above embodiments 2, 3 and 4 correspond to the method embodiment 1, and the specific implementation methods can refer to the relevant description part of embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood to include any medium that can store, encode or carry an instruction set for execution by a processor and enable the processor to execute any method in the present invention.

[0056] Those skilled in the art should understand that the modules or steps of the present invention described above can be implemented by a general-purpose computer device, or alternatively, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0057] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A user electricity consumption prediction method based on SSA-CCNN, characterized in that: include: Acquire historical electricity consumption data of users, and pre-process the historical electricity consumption data of users; The singular spectrum analysis method is used to decompose the preprocessed user historical electricity consumption data into time series of periodic components, trend components and noise components that characterize the characteristics of different users' historical electricity consumption data. The time series of the period component, trend component and noise component are input into the trained deep causal convolutional neural network model to obtain the period component prediction value, trend classification prediction value and noise component prediction value; The predicted value of the end-user electricity consumption is obtained by superimposing the periodic component prediction value, the trend classification prediction value and the noise component prediction value.

2. A user power consumption prediction method based on SSA-CCNN as claimed in claim 1, characterized in that: The process of preprocessing the user's historical electricity consumption data includes: data correction and normalization. The missing value interpolation method is used to correct the abnormal data in the user's historical electricity consumption data. The corrected data is normalized using the min-miax method. The min-miax formula is as follows: In the formula, is the normalized data; is the original data; are the maximum and minimum values ​​respectively.

3. A user power consumption prediction method based on SSA-CCNN as claimed in claim 1, characterized in that: The specific process of using the singular spectrum analysis method to decompose the pre-processed user historical power consumption data into a time series of periodic components, trend components and noise components that characterize the characteristics of different user historical power consumption data is as follows: Embed the user's historical electricity consumption data into a trajectory matrix X, perform singular value solving on the trajectory matrix, and obtain the singular value and eigenvector of the trajectory matrix of the user's historical electricity consumption data; The eigenvectors after singular value decomposition are grouped according to the size of the singular values ​​to obtain the period component, trend component and noise component of the user's historical power consumption data trajectory matrix; The diagonal averaging method is used to reconstruct the periodic component, trend component and noise component of the trajectory matrix to obtain the time series of the periodic component, trend component and noise component.

4. A method for predicting user power consumption based on SSA-CCNN as claimed in claim 3, characterized in that: The trajectory matrix X is, Where M represents the window length, which ranges from 2 to N / 2. J is the number of columns of the trajectory matrix X; , N express The user's historical electricity consumption data at the moment.

5. A method for predicting user power consumption based on SSA-CCNN as claimed in claim 3, characterized in that: The specific process of solving the singular value of the trajectory matrix and obtaining the singular value and eigenvector of the trajectory matrix of the user's historical power consumption data is as follows: Get the covariance matrix G of the trajectory matrix, G= ,in is the transposed matrix of X, and M singular values ​​and eigenvectors corresponding to the singular values ​​are obtained from the covariance matrix. Then the trajectory matrix X is expressed as: In the formula, represents the principal components of singular value decomposition; Represented as the singular values ​​of the trajectory matrix; represents the left singular vector; represents the right singular vector; express Eigenvalues ​​in descending order.

6. A method for predicting user power consumption based on SSA-CCNN as claimed in claim 3, characterized in that: The diagonal averaging method is as follows: In the formula, , is the transformed one-dimensional time series.

7. A method for predicting user power consumption based on SSA-CCNN as claimed in claim 1, characterized in that: The process of inputting the time series of the period component, trend component and noise component into the trained deep causal convolutional neural network model to obtain the period component prediction value, trend classification prediction value and noise component prediction value is as follows: The time series of periodic component, trend component and noise component are used as input to the deep causal convolutional neural network model; Capturing the long-term dependency of the time series of the periodic component, trend component and noise component through the first causal convolution layer and the second causal convolution layer in the deep causal convolutional neural network model; Using a pooling layer to reduce the length of the time series of the periodic component, the trend component and the noise component; The predicted values ​​of the time series of the periodic component, trend component and noise component are output through the fully connected layer.

8. A user electricity consumption prediction system based on SSA-CCNN, characterized in that: include: The data preprocessing module is configured to: obtain the user's historical electricity consumption data and preprocess the user's historical electricity consumption data; The data feature decomposition module is configured to: use a singular spectrum analysis method to decompose the pre-processed user historical power consumption data into a time series of a period component, a trend component and a noise component that characterize the characteristics of different user historical power consumption data; The user power consumption prediction module is configured to: input the time series of the period component, the trend component and the noise component into the trained deep causal convolutional neural network model, and then obtain the period component prediction value, the trend classification prediction value and the noise component prediction value; The predicted value of the end-user electricity consumption is obtained by superimposing the periodic component prediction value, the trend classification prediction value and the noise component prediction value.

9. A computer-readable storage medium having a program stored thereon, characterized in that: When the program is executed by a processor, the steps in the user power consumption prediction method based on SSA-CCNN as described in any one of claims 1 to 7 are implemented.

10. An electronic device comprising a memory, a processor, and a program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps in the user power consumption prediction method based on SSA-CCNN as described in any one of claims 1 to 7 are implemented.

Images