Power load prediction method based on time sequence core fusion

Abstract

The invention provides a power load prediction method based on time sequence core fusion, and relates to the technical field of artificial intelligence, and the method comprises the steps: carrying out the normalization processing of adjustable load data, and obtaining an adjustable load data feature vector; constructing an adjustable power load prediction model by using a multi-layer sensing layer, a random pooling layer and a fusion layer; inputting the adjustable load data feature vector into an adjustable power load prediction model, and training the adjustable power load prediction model based on an accuracy index to obtain a trained adjustable power load prediction model; and performing reverse normalization calculation on the feature sequence containing the prediction group to obtain an adjustable power load prediction result, and completing prediction of the adjustable power load. The problem that the accuracy of an adjustable power load prediction result is not high is solved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method for predicting power load based on time series core fusion. Background Art

[0002] Compared with traditional power loads, adjustable loads have dual characteristics of power supply and user. In terms of load characteristics, they can be decomposed into two parts: periodic component and non-periodic nonlinear component of the same magnitude. At present, the regression prediction method, smoothing prediction method, and LSTM prediction method of traditional load prediction are still mainly used in the field of adjustable load prediction. The above methods lack universality in the field of adjustable load prediction. When the non-periodic nonlinear component increases, the prediction accuracy will decrease significantly. Although the unit time series neural network method represented by the LSTM prediction method can enhance the prediction accuracy by enhancing the network structure and overcome the problem of increased error caused by the large non-periodic nonlinear component, it will significantly increase the operation time and computing power resource requirements. Summary of the invention

[0003] In view of the above-mentioned deficiencies in the prior art, the present invention provides a power load forecasting method based on time series core fusion to solve the problem of low accuracy of adjustable power load forecasting results.

[0004] In order to achieve the above-mentioned invention object, the technical solution adopted by the present invention is: a power load forecasting method based on time series core fusion, comprising: S1: obtaining an adjustable load data feature vector by normalizing the adjustable load data; wherein the adjustable load data feature vector is a time series feature vector including a low-frequency component and a high-frequency component; S2: Using multi-layer perception layer, random pooling layer, Kalman prediction layer and fusion layer, an adjustable power load prediction model is constructed; S3: inputting the adjustable load data feature vector into the adjustable power load prediction model, training the adjustable power load prediction model based on the accuracy index, and obtaining a trained adjustable power load prediction model; wherein the trained adjustable power load prediction model is used to analyze the adjustable load data feature vector to obtain a feature sequence containing a prediction group; S4: by performing a denormalization calculation on the characteristic sequence including the prediction group, an adjustable power load prediction result is obtained, thereby completing the prediction of the adjustable power load.

[0005] Furthermore, the expression of the adjustable load data characteristic vector is: ; ; ; in, represents the adjustable load data feature vector, represents the normalized weight, Indicates adjustable load data, represents the mean value of the adjustable load data, represents the standard deviation of the adjustable load data, represents the normalized offset, Indicates the total duration, express The adjustable load data feature vector at time.

[0006] Furthermore, the adjustable power load prediction model includes: A multi-layer perception layer, used for performing linear and dimensional transformation on the adjustable load data feature vector to obtain a transformed prediction sequence; A random pooling layer for learning features of adjustable load data based on the prediction sequence to generate a feature transition matrix; The Kalman prediction layer is used to fuse the high-frequency and low-frequency sequences of the feature transition matrix to obtain the feature core matrix; The fusion layer includes a feature core matrix, which is used to fuse the feature core matrix with the adjustable load data feature vector to obtain a feature sequence containing a prediction group.

[0007] Further, the S3 includes: The adjustable load data feature vector is used as the initial variable of the adjustable power load prediction model: ; Using the initial variables, the feature core matrix is ​​obtained through multi-layer perception layer and random pooling layer: ; The characteristic core matrix is ​​merged with the adjustable load data characteristic vector to obtain a characteristic sequence containing a prediction group: ; The error between the real feature sequence and the feature sequence containing the prediction group is calculated to obtain an accuracy index, and the adjustable power load prediction model is trained to obtain a trained adjustable power load prediction model: ; ; in, represents the initial variables of the adjustable power load forecasting model, represents the adjustable load data feature vector, represents the characteristic core matrix, express Wheel feature transition matrix, and represent discrete time points, represents the low-frequency component extraction function, express Wheel feature transition matrix, represents the high frequency component extraction function, represents the number of iterations, Indicates the total duration, Indicates the additional time length, represents the feature sequence containing the prediction group, represents the feature sequence of the prediction group in round i-1, represents the intermediate parameters for calculating the characteristic core matrix, represents the feature sequence containing the prediction group, express The moment feature core matrix, represents the mean square error, represents the number of feature sequences, represents the real feature sequence, represents the mean absolute error, Indicates the iteration round.

[0008] Furthermore, the expression of the adjustable power load prediction result is: ; in, represents the forecast result of adjustable power load, represents the standard deviation of the adjustable load data, represents the feature sequence containing the prediction group, represents the normalized offset, represents the normalized weight, Indicates the mean value of the adjustable load data.

[0009] The beneficial effects of the present invention are as follows: normalizing the adjustable load data to obtain the adjustable load data feature vector, and using the adjustable power load prediction model to analyze the adjustable load data feature vector to obtain the adjustable power load prediction result. (1) Based on the sequence mean and standard deviation, a learnable linear parameter is introduced at the same time, so that the entire normalization and denormalization process can be continuously iterated and optimized according to the network training process, thereby obtaining the best normalization effect, so that the network operation stability can be improved, and the prediction accuracy can be improved; (2) The adjustable power load prediction model is used to aggregate information from different sequence features to obtain the global core representation, and the local sequence representation is merged with the core representation to realize the indirect interaction between the sequence features. This centralized interaction not only reduces the comparison complexity, but also utilizes the aggregated representation, which helps to improve the local representation and improve the speed and accuracy of network calculation; (3) Using the multi-layer perception layer and the random pooling layer, the core features of each feature sequence are learned while ensuring the calculation speed, and the fused features are assigned to each feature sequence, so that the generated data has a very high correlation with the input data and can better fit the actual value in the future. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] This specification will be further described in the form of exemplary embodiments, which will be described in detail by the accompanying drawings. These embodiments are not restrictive, and in these embodiments, the same number represents the same structure, wherein: Figure 1 is an exemplary flow chart of a method for power load forecasting based on time series core fusion according to some embodiments of this specification; Figure 2 is an exemplary schematic diagram of adjustable load data with two characteristic components according to some embodiments of the present specification; Figure 3 is an exemplary schematic diagram of an adjustable load data feature vector according to some embodiments of the present specification; FIG4( a) is an exemplary schematic diagram of adjustable power load prediction results according to some embodiments of the present specification; FIG4( b ) is an exemplary schematic diagram of adjustable power load prediction results according to some embodiments of this specification; Figure 5 It is an exemplary schematic diagram of the error between the first characteristic component and the second characteristic component according to some embodiments of the present specification. DETAILED DESCRIPTION

[0011] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0012] Example Figure 1 is an exemplary flow chart of a method for power load forecasting based on time series core fusion according to some embodiments of this specification. Figure 1 As shown, the process includes the following steps. In some embodiments, the process can be executed by a processor.

[0013] S1: Obtain an adjustable load data feature vector by normalizing the adjustable load data.

[0014] Adjustable load data refers to the data on the relationship between power system load and time.

[0015] In some embodiments, the processor may obtain adjustable load data based on historical load data recorded by the power system.

[0016] The adjustable load data feature vector is a normalized adjustable load data vector form. For example, the adjustable load data feature vector may be a time series power load feature vector; the adjustable load data feature vector may include a low frequency component and a high frequency component.

[0017] In some embodiments, the expression of the adjustable load data feature vector may be: ; ; ; in, represents the adjustable load data feature vector, represents the normalized weight, Indicates adjustable load data, represents the mean value of the adjustable load data, represents the standard deviation of the adjustable load data, represents the normalized offset, Indicates the total duration, express The adjustable load data feature vector at time.

[0018] In some embodiments, the processor may perform wavelet transform on the adjustable load data feature vector, extract the feature sequence through frequency, and obtain a low-frequency component and a high-frequency component.

[0019] The low frequency component is a low frequency component in the adjustable load data feature vector that reflects the long-term trend.

[0020] The high-frequency component is a high-frequency component in the adjustable load data feature vector that reflects short-term fluctuations.

[0021] In some embodiments, the expressions of the low frequency component and the high frequency component may be: ; ; ; ; in, represents the low-frequency component, represents the low-frequency component extraction function, and represent discrete time points, Indicates the total duration, represents the high frequency component, Represents the high frequency component extraction function.

[0022] S2: Build an adjustable power load forecasting model using multi-layer perception layers, random pooling layers, and feature core matrices.

[0023] The adjustable power load prediction model is used to predict the load of the power system to obtain the power load prediction result. The adjustable power load prediction model can be of various types. For example, the type of the adjustable power load prediction model can include a deep neural network model.

[0024] In some embodiments, the input of the adjustable power load prediction model may be an adjustable load data feature vector, and the output of the adjustable power load prediction model may be a feature sequence comprising a prediction group.

[0025] In some embodiments, the structure of the adjustable power load prediction model is as follows: The adjustable power load forecasting model includes a multi-layer perception layer, a random pooling layer, a Kalman prediction layer and a fusion layer. The output of the multi-layer perception layer is used as the input of the random pooling layer, the output of the random pooling layer is used as the input of the Kalman prediction layer, the output of the multi-layer perception layer and the output of the Kalman prediction layer are used as the input of the fusion layer, and the output of the fusion layer is used as the final output of the adjustable power load forecasting model.

[0026] The multi-layer perception layer is used to perform linear and dimensional transformation on the adjustable load data feature vector to obtain a transformed prediction sequence. The input of the multi-layer perception layer may include the adjustable load data feature vector, and the output may include the prediction sequence.

[0027] The prediction sequence is the sequence of adjustable load data feature vectors after linear dimension transformation.

[0028] In some embodiments, the multi-layer perception layer may include an input module, a first fully connected module, a second fully connected module, and an output module.

[0029] In some embodiments, the input module can be used to obtain an adjustable load data feature vector, the first fully connected module and the second fully connected module can be used to analyze the adjustable load data feature vector to obtain an analysis result, and the output module can be used to output the analysis result.

[0030] In some embodiments, the activation function used by the first fully connected module and the second fully connected module may be a GELU activation function: ; in, represents the output of the activation function, Represents the feature value after the network activation layer, represents the distribution rule of feature sequence, Represents the cumulative distribution function of the normal distribution.

[0031] The random pooling layer is used to learn the features of the adjustable load data based on the prediction sequence and generate a feature core matrix. The input of the random pooling layer may include the prediction sequence, and the output may include the feature transition matrix.

[0032] The characteristic transition matrix is ​​a transition matrix that reflects the characteristic information of the high-frequency component sequence and the characteristic information of the low-frequency component sequence.

[0033] In some embodiments, the expression of the characteristic transition matrix may be: ; ; in, represents the random pooling layer function, represents the multi-layer perception layer function, represents the feature sequence of the prediction group numbered i-1, represents the intermediate parameters for calculating the characteristic core matrix, represents the coefficients of the characteristic transition matrix, represents the value of the element in row i and column j in A, represents the number of features, Indicates the The value of the element in the row and column j, Indicates OK, represents the distributive associative function, represents the characteristic transition matrix. .

[0034] In some embodiments, The shape can be , including high-frequency component sequence feature information and low-frequency component sequence feature information; among them, represents the field of real numbers, represents the number of sequence features, Indicates the total duration.

[0035] The Kalman prediction layer is used to fuse the high-frequency and low-frequency sequences of the feature transition matrix to obtain the feature core matrix. The input of the Kalman prediction layer may include the feature transition matrix, and the output may include the feature core matrix.

[0036] In some embodiments, the processor may input the prediction sequence into the Kalman prediction layer and obtain the characteristic core matrix through inverse wavelet transform. For example, the processor may obtain the state transfer matrix through calculation based on the prediction sequence; obtain the prediction result through calculation using the state transfer matrix; and obtain the characteristic core matrix by fusing the prediction result with high and low frequencies.

[0037] The state transfer matrix is ​​a matrix that reflects the state transfer of the characteristic core matrix.

[0038] In some embodiments, the expression of the state transition matrix may be: ; in, represents the state transfer matrix, express The characteristic core matrix of express The characteristic core matrix of express The characteristic core matrix of .

[0039] The prediction result reflects the transformation of the characteristic core matrix over a period of time in the future.

[0040] In some embodiments, the expression of the prediction result may be: ; in, Represents the prediction result, represents the t-th power of the state transfer matrix, Represents the data with subscript T in the feature core matrix.

[0041] The fusion layer is used to fuse the feature core matrix with the adjustable load data feature vector to obtain a feature sequence containing a prediction group. The input of the fusion layer may include the feature core matrix and the adjustable load data feature vector, and the output may include a feature sequence containing a prediction group.

[0042] The characteristic core matrix is ​​a matrix that reflects the characteristic information of the aggregated sequence.

[0043] In some embodiments, the processor may perform core fusion on the adjustable load data feature vector using the feature core matrix to obtain a feature sequence including the prediction group. For example, the processor may assign the Nth row of the feature core matrix to the Mth adjustable load data feature vector, and the combination is to directly add the assigned sequence to the corresponding normalized feature sequence to obtain a feature sequence including the prediction group.

[0044] S3: Input the adjustable load data feature vector into the adjustable power load prediction model, train the adjustable power load prediction model based on the accuracy index, and obtain a trained adjustable power load prediction model; wherein the trained adjustable power load prediction model is used to analyze the adjustable load data feature vector to obtain a feature sequence containing a prediction group.

[0045] In some embodiments, the adjustable power load prediction model can be obtained by training multiple labeled training samples. For example, multiple labeled training samples can be input into the initial adjustable power load prediction model, and the accuracy index can be constructed by the labels and the results of the initial adjustable power load prediction model. The parameters of the initial adjustable power load prediction model are iteratively updated by gradient descent or other methods based on the accuracy index. When the preset conditions are met, the model training is completed and the trained adjustable power load prediction model is obtained. The preset conditions can be that the accuracy index converges, the number of iterations reaches a threshold, etc.

[0046] In some embodiments, the training sample may include a historical adjustable load data feature vector. The label may be a corresponding real feature sequence. The label may be manually annotated.

[0047] The real characteristic sequence is a sequence constructed from the adjustable power load results at the corresponding time points actually monitored by the power system.

[0048] In some embodiments, the processor can implement S3 based on the following steps: using the adjustable load data feature vector as the initial variable of the adjustable power load prediction model; using the initial variable to obtain a feature core matrix through a multi-layer perception layer and a random pooling layer; fusing the feature core matrix with the adjustable load data feature vector to obtain a feature sequence containing a prediction group; performing error calculation between the real feature sequence and the feature sequence containing the prediction group to obtain an accuracy index, training the adjustable power load prediction model, and obtaining a trained adjustable power load prediction model.

[0049] Initial variables are the input variables of the adjustable power load forecasting model before training.

[0050] In some embodiments, the expression of the initial variable may be: .

[0051] In some embodiments, the expression of the feature core matrix may be: .

[0052] In some embodiments, the expression containing the feature sequence of the prediction group may be: .

[0053] The accuracy index is an index used to judge the performance of the adjustable power load forecasting model. For example, the accuracy index may include mean square error and mean absolute error.

[0054] In some embodiments, the expression of the accuracy indicator may be: ; ; in, represents the initial variables of the adjustable power load forecasting model, represents the adjustable load data feature vector, represents the characteristic core matrix, express Wheel feature transition matrix, and represent discrete time points, represents the low-frequency component extraction function, express Wheel feature transition matrix, represents the high frequency component extraction function, represents the number of iterations, Indicates the total duration, Indicates the additional time length, represents the feature sequence containing the prediction group, represents the feature sequence of the prediction group in round i-1, represents the intermediate parameters for calculating the characteristic core matrix, represents the feature sequence containing the prediction group, express The moment feature core matrix, represents the mean square error, represents the number of feature sequences, represents the real feature sequence, represents the mean absolute error, represents the iteration round; where .

[0055] S4: by performing a denormalization calculation on the characteristic sequence including the prediction group, an adjustable power load prediction result is obtained, thereby completing the prediction of the adjustable power load.

[0056] The adjustable power load forecast result reflects the adjustable power load result of the power system in the future period of time.

[0057] In some embodiments, the expression of the adjustable power load prediction result may be: ; in, represents the forecast result of adjustable power load, represents the standard deviation of the adjustable load data, represents the feature sequence containing the prediction group, represents the normalized offset, represents the normalized weight, Indicates the mean value of the adjustable load data.

[0058] In some embodiments, the processor may use the first 90 time points of the adjustable load data with two characteristic components as the prediction data points, and normalize the data to obtain the following: Figure 3 The original data for prediction (i.e., the adjustable load data feature vector) shown in FIG4 is used to generate a feature core matrix through a multi-layer perceptron and a random pooling layer. Finally, the feature core matrix is ​​assigned to each feature sequence and then denormalized to obtain the adjustable power load prediction results shown in FIG4(a) and FIG4(b). Figure 2 As shown, the adjustable load data with two characteristic components may include a first characteristic component (periodic load), a second characteristic component (non-periodic oscillating load), a total load time series length (100 units), and each unit corresponds to 1 hour.

[0059] In some embodiments, Figure 5As shown, the sum of the errors of the first characteristic component and the second characteristic component is distributed on time series 91 to time series 100, showing an average error of 0.0015 and an accuracy of 99.25%. The data used in this method is predicted by other methods, and the accuracy comparison results are shown in Table 1.

[0060] Table 1

[0061] In some embodiments of the present specification, the processor can normalize the adjustable load data to obtain the adjustable load data feature vector, and use the adjustable power load prediction model to analyze the adjustable load data feature vector to obtain the adjustable power load prediction result. (1) Based on the sequence mean and standard deviation, a learnable linear parameter is introduced at the same time, so that the entire normalization and denormalization process can be continuously iterated and optimized according to the network training process, so as to obtain the best normalization effect, so that the operation stability of the network can be improved, and the prediction accuracy can be improved; (2) The adjustable power load prediction model is used to aggregate information from different sequence features to obtain the global core representation, and the local sequence representation is merged with the core representation to realize the indirect interaction between the sequence features. This centralized interaction not only reduces the comparison complexity, but also utilizes the aggregated representation, which helps to improve the local representation and improve the speed and accuracy of network calculation; (3) Using a multi-layer perception layer and a random pooling layer, the core features of each feature sequence are learned while ensuring the calculation speed, and the fused features are assigned to each feature sequence, so that the generated data has a high correlation with the input data and can better fit the actual value in the future.

Claims

1. An adjustable power load forecasting method based on time series core fusion, characterized in that: include: S1: obtaining an adjustable load data feature vector by normalizing the adjustable load data; wherein the adjustable load data feature vector is a time series feature vector including a low-frequency component and a high-frequency component; S2: Using multi-layer perception layer, random pooling layer, Kalman prediction layer and fusion layer, an adjustable power load prediction model is constructed; S3: inputting the adjustable load data feature vector into the adjustable power load prediction model, training the adjustable power load prediction model based on the accuracy index, and obtaining a trained adjustable power load prediction model; wherein the trained adjustable power load prediction model is used to analyze the adjustable load data feature vector to obtain a feature sequence containing a prediction group; S4: by performing a denormalization calculation on the characteristic sequence including the prediction group, an adjustable power load prediction result is obtained, thereby completing the prediction of the adjustable power load.

2. The power load forecasting method based on time series core fusion according to claim 1 is characterized in that: The expression of the adjustable load data characteristic vector is: ； ； ； in, represents the adjustable load data feature vector, represents the normalized weight, Indicates adjustable load data, represents the mean value of the adjustable load data, represents the standard deviation of the adjustable load data, represents the normalized offset, Indicates the total duration, express The adjustable load data feature vector at time.

3. The power load forecasting method based on time series core fusion according to claim 1 is characterized in that: The adjustable power load prediction model includes: A multi-layer perception layer, used for performing linear and dimensional transformation on the adjustable load data feature vector to obtain a transformed prediction sequence; A random pooling layer for learning features of adjustable load data based on the prediction sequence to generate a feature transition matrix; The Kalman prediction layer is used to fuse the high-frequency and low-frequency sequences of the feature transition matrix to obtain the feature core matrix; The fusion layer includes a feature core matrix, which is used to fuse the feature core matrix with the adjustable load data feature vector to obtain a feature sequence containing a prediction group.

4. The power load forecasting method based on time series core fusion according to claim 1 is characterized in that: The S3 includes: The adjustable load data feature vector is used as the initial variable of the adjustable power load prediction model: ； Using the initial variables, the feature core matrix is ​​obtained through multi-layer perception layer and random pooling layer: ； The characteristic core matrix is ​​merged with the adjustable load data characteristic vector to obtain a characteristic sequence containing a prediction group: ； The error between the real feature sequence and the feature sequence containing the prediction group is calculated to obtain an accuracy index, and the adjustable power load prediction model is trained to obtain a trained adjustable power load prediction model: ； ； in, represents the initial variables of the adjustable power load forecasting model, represents the adjustable load data feature vector, represents the characteristic core matrix, express Wheel feature transition matrix, and represent discrete time points, represents the low-frequency component extraction function, express Wheel feature transition matrix, represents the high frequency component extraction function, represents the number of iterations, Indicates the total duration, Indicates the additional time length, represents the feature sequence containing the prediction group, represents the feature sequence of the prediction group in round i-1, represents the intermediate parameters for calculating the characteristic core matrix, represents the feature sequence containing the prediction group, express The moment feature core matrix, represents the mean square error, represents the number of feature sequences, represents the real feature sequence, represents the mean absolute error, Indicates the iteration round.

5. The method for power load forecasting based on time series core fusion according to claim 1 is characterized in that: The expression of the adjustable power load prediction result is: ； in, represents the forecast result of adjustable power load, represents the standard deviation of the adjustable load data, represents the feature sequence containing the prediction group, represents the normalized offset, represents the normalized weight, Indicates the mean value of the adjustable load data.

Images