Park short-term power load prediction method based on VMD and SRU network

Abstract

The invention relates to the technical field of power load prediction, and provides a park short-term power load prediction method based on a VMD and SRU network, and the method comprises the following specific steps: 1, data acquisition and preprocessing; 2, decomposing data through VMD; step 3, carrying out data normalization; step 4, constructing and training an SRU network; and 5, model prediction and result superposition are carried out. The power load prediction method has the advantages that original power load data are decomposed by using a VMD-SRU-based combined prediction model, then training prediction is performed, a simple loop unit extracts time sequence features of the decomposed data, and through a unique network structure, the training speed can be increased while the power load prediction precision is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power load forecasting, and in particular to a method for forecasting a park's short-term power load based on a VMD and SRU network. Background Art

[0002] With the booming and continuous deepening of the smart industrial park economy, especially the widespread application and practice of integrated energy management and energy storage technology in industrial parks, industrial parks are steadily transforming towards a more low-carbon, environmentally friendly and efficient economic model. In this transformation process, power load management in the industrial production link is particularly important. Through scientific and reasonable adjustment and scheduling of power load, not only can the utilization rate of new energy be effectively improved, but also its potential in energy conservation, emission reduction and operating cost reduction can be maximized.

[0003] However, to achieve reasonable dispatch and optimization of power load, the premise is to make highly accurate predictions of future power load. This kind of prediction requires not only a high level of prediction accuracy, but also requires the training and prediction process to respond quickly, so as to timely and effectively guide the adjustment and dispatch of power load and ensure the continuity and stability of industrial production processes.

[0004] Traditionally, power load forecasting mostly uses physical model methods or statistical model methods. Although these methods can reflect the changing trend of power load to a certain extent, they generally face the problems of high time complexity, high data quality requirements, and the gradual decrease in prediction accuracy over time. What is particularly critical is that they often find it difficult to accurately capture the complex nonlinear relationship between power load and various influencing factors, as well as the temporal characteristics of the data itself, which makes it difficult for traditional methods to meet the high-precision requirements of industrial parks for power load forecasting.

[0005] In recent years, with the rapid development and widespread penetration of artificial intelligence technology, all walks of life have benefited from this technological innovation, achieving significant improvements in production efficiency and steady growth in economic benefits. In the field of power load forecasting, the continuous progress of machine learning and deep learning algorithms has provided new solutions to this problem. However, in early application research, machine learning or deep learning algorithms mainly relied on a single prediction model for power load forecasting, which to a certain extent limited the further improvement of prediction accuracy. Summary of the invention

[0006] Aiming at the deficiencies of the existing technology, the problem of low accuracy and slow prediction speed of the existing power load prediction is solved.

[0007] The present invention provides a method for predicting short-term power load in a park based on VMD and SRU network, and the specific steps are as follows: Step 1: Data acquisition and preprocessing; The original data of power load of a certain park is obtained from a data server or a data acquisition system, the original data is preprocessed, and the processed data is used as a data set.

[0008] Step 2: VMD decomposition data; The variational mode decomposition technique is used to decompose the power load data obtained in step one.

[0009] Step 3: Data normalization; Perform maximum and minimum normalization processing on the data decomposed in step 2.

[0010] Step 4: SRU network construction and training; According to the normalized data set in step 3, different SRU networks are constructed for the K IMF components for training to obtain K prediction models. The recurrent unit structure in the SRU network mainly includes the following calculation formula: Forget Gate Calculation formula: Internal State Calculation formula: Reset Gate Calculation formula: Hidden State Calculation formula: in, is the Sigmoid function; , and Represents the weight matrix that needs to be learned; is the input data at time t; , , and is the parameter vector to be learned; represents the unit state at the previous moment; ⊙ represents the Hadamard product.

[0011] Step 5: Model prediction and result superposition; Use the model built and trained in step 4 to predict the power load in the park. Denormalize the prediction results of each prediction model and superimpose them to obtain the final prediction result.

[0012] As a preferred solution, the specific decomposition process of VMD in step 2 includes: S1. Get K modal functions from the original signal And the corresponding center frequency , and introduce the Lagrange multiplication operator and the quadratic penalty factor ; S2, Initialization modal function set , center frequency set , Lagrange multiplication operator set And the number of iterations n=0; S3, according to the formula and renew and K times; S4. Update Lagrange multipliers , And the cumulative number of iterations n; S5. Repeat S2 to S4 until the termination condition is met: in, is the discrimination accuracy and .

[0013] As a preferred solution, the normalized formula is: , in, : Original power load data; : Normalized power load data; : The maximum value in the data set; : The minimum value in the data set.

[0014] As a preferred solution, the preprocessing method includes using interpolation to fill missing values ​​and using a Z-Score method to detect and correct outliers.

[0015] As a preferred solution, the initial value of K is set to 7, and the value of K is adjusted by observing whether there is modal aliasing phenomenon. When modal aliasing phenomenon exists, the K value is reduced, and when modal aliasing phenomenon does not exist, the K value is increased, adjusted to the critical point of the modal aliasing phenomenon, and rounded to obtain the K value.

[0016] The beneficial effects of the present invention are: The present invention uses a combined prediction model based on VMD-SRU. The original power load data is first decomposed, and then training prediction is performed. The simple recurrent unit (SRU) extracts the time series characteristics of the decomposed data. Through the unique network structure, the training speed can be improved while improving the power load prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to make the content of the present invention more clearly understood, the present invention is further described in detail below according to specific embodiments and in conjunction with the accompanying drawings, wherein Figure 1 It is a flow chart of the present invention.

[0018] Figure 2 This is a structural diagram of the SRU network unit in the present invention.

[0019] Figure 3 This is a flow chart of VMD decomposition in the present invention. DETAILED DESCRIPTION

[0020] To illustrate the features of the present invention, the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0021] Example: See also Figures 1 to 3 The embodiment of the present invention provides a method for predicting short-term power load in a park based on VMD and SRU network. The specific steps of the method are as follows: Step 1: Data acquisition and preprocessing Obtain the original data of power load of a certain park from the data server or data acquisition system. Use interpolation to fill missing values ​​and use the Z-Score method to detect and correct outliers. The processed data is used as a data set.

[0022] Step 2: VMD decomposition data The VMD algorithm is used to decompose the data set to obtain multiple IMFs. Before using the VMD algorithm to decompose the original power load data, the VMD parameters are first set. For the modal number K, it is generally adjusted according to the decomposition result. The initial value of K is set to 7, and the value of K is adjusted to observe whether there is a modal aliasing phenomenon. When there is a modal aliasing phenomenon, the K value is reduced. When there is no modal aliasing phenomenon, the K value is increased. It is adjusted to the critical point of the modal aliasing phenomenon and the K value is rounded to obtain the integer; the quadratic penalty factor α is 2700; the noise tolerance tau is 1; the synthetic signal DC is 0. After setting the parameters, the VMD decomposition operation is performed to obtain multiple IMF components.

[0023] The specific decomposition process is: S1. Get K modal functions from the original signal And the corresponding center frequency , and introduce the Lagrange multiplication operator and the quadratic penalty factor ; S2, Initialization modal function set , center frequency set , Lagrange multiplication operator set And the number of iterations n=0; S3, according to the formula and renew and K times; S4. Update Lagrange multipliers , , and accumulate the number of iterations n; S5. Repeat S2 to S4 until the termination condition is met: in, is the discrimination accuracy and .

[0024] Step 3: Data Normalization Perform maximum and minimum normalization on each IMF. Calculate the maximum and minimum values ​​of each IMF and use the normalization formula The normalized data set is divided into training set, validation set and test set according to a certain ratio.

[0025] Step 4: SRU network construction and training Since each IMF component is different, it is necessary to build an SRU network for each IMF component for training. The network structure and hyperparameters of each SRU network are different. The Adam optimization algorithm is used uniformly during training, and the mean square error loss (MSELoss) is used as the loss function. Use normalized IMF data for training, and continuously optimize network parameters through manual parameter adjustment. Observe the change curve of the loss function during training to ensure model convergence.

[0026] The cyclic unit structure in the SRU network mainly includes the following calculation formula: Forget Gate Calculation formula: Internal State Calculation formula: Reset Gate Calculation formula: Hidden State Calculation formula: in, is the Sigmoid function; , and Represents the weight matrix that needs to be learned; is the input data at time t; , , and is the parameter vector to be learned; represents the unit state at the previous moment; ⊙ represents the Hadamard product.

[0027] Step 5: Model prediction and result superposition Use the trained SRU network to predict each IMF. Denormalize the prediction results to get the actual predicted load value.

[0028] The above embodiments and drawings are only used to illustrate the technical solutions of the present invention, and are not intended to limit the present invention. The present invention is described in detail with reference to the preferred embodiments. It should be understood by those skilled in the art that any changes, modifications, additions or substitutions made by those skilled in the art within the essential scope of the present invention do not deviate from the purpose of the present invention and should also fall within the scope of protection of the claims of the present invention. Other related technical structures not fully disclosed in the present invention are prior art in the art.

Claims

1. A method for short-term power load forecasting in a park based on VMD and SRU network, characterized in that: The specific steps are as follows: Step 1: Data acquisition and preprocessing; Obtain the original data of power load of a certain park from the data server or data acquisition system, pre-process the original data, and use the processed data as a data set; Step 2: VMD decomposition data; Decomposing the power load data obtained in step 1 using variational mode decomposition technology; Step 3: Data normalization; Perform maximum and minimum normalization processing on the data decomposed in step 2; Step 4: SRU network construction and training; According to the normalized data set in step 3, different SRU networks are constructed for the K IMF components for training to obtain K prediction models. The recurrent unit structure in the SRU network mainly includes the following calculation formula: Forget Gate Calculation formula: Internal State Calculation formula: Reset Gate Calculation formula: Hidden State Calculation formula: in, is the Sigmoid function; , and Represents the weight matrix that needs to be learned; is the input data at time t; , , and is the parameter vector to be learned; represents the unit state at the previous moment; ⊙ represents the Hadamard product; Step 5: Model prediction and result superposition; Use the model built and trained in step 4 to predict the power load in the park. Denormalize the prediction results of each prediction model and superimpose them to obtain the final prediction result.

2. The method for predicting short-term power load in a park based on VMD and SRU network according to claim 1 is characterized in that: The specific decomposition process of VMD in step 2 includes: S1. Get K modal functions from the original signal And the corresponding center frequency , and introduce the Lagrange multiplication operator and the quadratic penalty factor ; S2, Initialization modal function set , center frequency set , Lagrange multiplication operator set And the number of iterations n=0; S3, according to the formula and renew and K times; S4. Update Lagrange multipliers , And the cumulative number of iterations n; S5. Repeat S2 to S4 until the termination condition is met: in, is the discrimination accuracy and .

3. The method for predicting short-term power load in a park based on VMD and SRU network according to claim 1 is characterized in that: The normalized formula is: , in, : Original power load data; : Normalized power load data; : The maximum value in the data set; : The minimum value in the data set.

4. The method for predicting short-term power load in a park based on VMD and SRU network according to claim 1 is characterized in that: The preprocessing method includes using interpolation to fill missing values ​​and using the Z-Score method to detect and correct outliers.

5. The method for predicting short-term power load in a park based on VMD and SRU network according to claim 2 is characterized in that: The initial value of K is set to 7, and the value of K is adjusted by observing whether there is a modal aliasing phenomenon. When there is a modal aliasing phenomenon, the K value is reduced, and when there is no modal aliasing phenomenon, the K value is increased, and adjusted to the critical point of the modal aliasing phenomenon, and the K value is rounded to obtain the integer.

Images