Park power load prediction method and system and computer readable storage medium
Through the CEEMD decomposition and XLSTM-Transformer combination model, the problem that the existing technology is difficult to adapt to complex power load changes is solved, and a higher accuracy and robust campus power load prediction is achieved.
Patent Information
- Application Number
- CN202411851651.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-16
- Publication Date
- 2025-05-16
AI Technical Summary
The existing campus power load prediction methods are difficult to adapt to complex and variable power load changes, and a single deep learning model cannot fully perform feature mining under limited data conditions, resulting in limited prediction accuracy.
The CEEMD decomposition technology is used to reduce the complexity of the original data, and combined with the XLSTM-Transformer combination model for feature extraction and prediction, multiple eigenmodal components and residual sequences are obtained through complementary set empirical modal decomposition, and data feature extraction and prediction processing is further used to use the XLSTM and Transformer models for data feature extraction and prediction processing.
It improves the accuracy and robustness of the power load prediction in the park, and can more accurately capture the main feature information in the data, thereby meeting the needs of power system allocation in the park.
Smart Images
Figure CN120016432A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of park power load forecasting, and in particular to an XLSTM-Transformer park power load forecasting method, system and computer-readable storage medium based on CEEMD decomposition. Background Art
[0002] In today's rapidly developing society, efficient use and management of energy is of vital importance. As an important carrier of industrial agglomeration and economic development, the power load forecast of the industrial park has great practical significance. With the continuous expansion of industrial production and the acceleration of the intelligent process, the demand for electricity by enterprises in the industrial park is growing and showing complex and changeable characteristics. On the one hand, different types of enterprises have different electricity consumption patterns. For example, the startup of high-power equipment in manufacturing enterprises will bring instantaneous high loads, while technology-based enterprises may have relatively stable power demand due to the continuous operation of servers. On the other hand, natural factors such as seasonal changes and temperature fluctuations will also have a significant impact on the power load of the industrial park. In the high temperature in summer, the large-scale use of refrigeration equipment such as air conditioners will cause a sharp increase in power load; in winter, heating equipment in some specific industries will also increase power consumption.
[0003] At the same time, with the continuous development and application of renewable energy, more and more parks have begun to introduce clean energy such as solar energy and wind energy. The intermittent and uncertain nature of these energy sources further increases the difficulty of predicting the power load in the park. Accurately predicting the power load in the park can not only help power companies to rationally plan grid construction and dispatch power resources to ensure stable and reliable power supply in the park, but also provide an important basis for park managers to formulate energy-saving and emission reduction strategies and optimize energy allocation. Through accurate prediction of power load, the park can achieve efficient use of energy, reduce operating costs, and improve economic and environmental benefits. In addition, with the continuous advancement of smart grid technology, the improvement of real-time monitoring and data analysis capabilities has also made it possible to more accurately predict power load. Using advanced sensor technology and big data analysis methods, a large amount of power data can be collected and processed, and the potential laws therein can be mined, thereby improving the accuracy and reliability of load forecasting.
[0004] At present, the following methods are mainly used in the field of power load forecasting in the park: Time series analysis method predicts based on the law of historical power load data changing over time, such as the simple moving average method predicts the future by calculating the average value of load values at several time points in the past, but this method is difficult to adapt to complex load changes. Regression analysis method considers the relationship between factors such as temperature, date type, economic growth and power load by establishing a linear or nonlinear regression model between power load and multiple influencing factors, but it is difficult to accurately determine the influencing factors and establish a suitable model. Artificial intelligence methods are playing an increasingly important role in this field. Artificial neural networks, especially multilayer perceptrons, imitate the working mode of neurons in the human brain, establish a nonlinear mapping relationship between input and output by learning a large amount of historical data, and have strong nonlinear fitting ability, but require a large amount of training data and poor interpretability. Support vector machines are based on statistical learning theory, have good prediction performance in small sample conditions, and have strong generalization ability, but have high computational complexity and large influence of parameter selection. The long short-term memory network (LSTM) in the deep learning method is specialized in processing time series data, which can effectively capture the long-term dependence and short-term fluctuation characteristics of power load; the convolutional neural network (CNN) can extract spatial and temporal features by performing convolution operations on load data to improve prediction accuracy. However, a single deep learning model cannot fully mine features in limited data, thus limiting the prediction accuracy. Summary of the invention
[0005] The present invention aims to solve the existing technical problems and proposes a park power load forecasting method and system. The CEEMD decomposition technology is used to reduce the complexity of the original park power load data, and the XLSTM-Transformer combined model is used to achieve more sufficient feature extraction and more accurate prediction to meet the needs of park power system deployment, so as to further improve the accuracy and robustness of park power load forecasting.
[0006] The technical solution of the present invention is as follows:
[0007] A method for predicting power load in a park, comprising:
[0008] The park power load data is collected and preliminarily interpolated to obtain an initial park power load data set;
[0009] The initial park power load data set is normalized, and then complementary set empirical mode decomposition is used to obtain multiple intrinsic mode components and a residual sequence;
[0010] A feature extraction model of the extended long short-term memory neural network XLSTM is constructed, and feature extraction is performed on the decomposition obtained by the complementary set empirical mode decomposition. The transformer model Transformer is selected for prediction processing, and the prediction results of each component are integrated to obtain the final prediction value of the future park power load.
[0011] Furthermore, the collecting of the park power load data and performing preliminary interpolation processing to obtain an initial park power load data set includes:
[0012] First, determine the known time data point set as {(t 1 ,y 1 ),(t 2 ,y 2 ),.....,(t n ,y n )}, where t 1 ,t 2 ,.....,t n is a known time point, y 1 ,y 2 ,.....,y n is the corresponding time data value;
[0013] For the target time point t 0 , calculate its distance d from each known time data point i =|t 0 -t i |, where i = 1, 2, ..., n;
[0014] Find the time point t at which the distance is minimum k , that is, satisfying d k =min(d 1 ,d 2 ,.....,d n );
[0015] Set the time point t k The corresponding time data value y k As the target time point t 0 The interpolation result of .
[0016] Furthermore, the park power load data set is normalized, and then complementary set empirical mode decomposition (CEEMD) is used to obtain multiple intrinsic mode components and a residual sequence, including:
[0017] 1) The initial park power load data is normalized, scaled in the interval [0, 1], converted into a dimensionless pure value, and the load is mapped to the interval [0, 1] using the min-max normalization method. The calculation formula is as follows:
[0018]
[0019] Where: x max is the maximum value of the data, x min is the minimum value of the data;
[0020] 2) Add a certain amount of white noise n(t) to the normalized signal x(t) to form a new signal:
[0021] x n (t) = x(t) + n(t)
[0022] Among them, n(t) has a mean of 0 and a variance of σ 2 White noise;
[0023] 3) Add white noise to the signal x n (t) Perform EMD decomposition, adding different white noise each time to obtain the signal decomposition amount:
[0024]
[0025] Among them, r N (t) is the residual component, N is the number of IMFs;
[0026] 4) Repeat steps 2) and 3) multiple times to obtain multiple new IMFs;
[0027] 5) Average all IMFs to get the final IMF:
[0028]
[0029] Among them, K is the number of repetitions, which is set to 100 times based on experience;
[0030] 6) Reconstruct the original signal using the obtained IMF and residual components:
[0031]
[0032] Furthermore, XLSTM is used to extract features from each component, specifically:
[0033] First, prepare the input time series data. Suppose there is a time series data set x = {x 1 ,x 2 ,.....,x T}, where x t represents the input vector at time step t, where T is the length of the time series;
[0034] Initialize various parameters of XLSTM, including forget gate, input gate, output gate, cell state and expansion state;
[0035] Input vector x t , the hidden state h of the previous time step t-1 , cell state c t-1 and extended state t-1 Perform linear transformations through different weight matrices respectively, then add the results and use the sigmoid activation function to get the value of the input gate i t , the input gate determines how much new information can be stored in the cell state at the current time step;
[0036] The input vector, the hidden state of the previous time step, the cell state and the extended state are linearly transformed and activated by the sigmoid function to obtain the value of the forget gate f. t ; The forget gate determines how much information in the cell state at the previous time step needs to be forgotten.
[0037] The output gate o is calculated in the same way t , which determines how much information about the cell state at the current time step can be output as a hidden state;
[0038] The input vector and the hidden state of the previous time step are linearly transformed and activated by the tanh function to obtain the candidate cell state c~ t ;
[0039] The input gate and forget gate are used to control the inflow and outflow of information, and the cell state of the previous time step and the candidate cell state are weighted summed to obtain the cell state c of the current time step. t ;
[0040] The input vector, the hidden state of the previous time step, and the cell state are linearly transformed and activated by a specific function to obtain the extended state e. t ;
[0041] The output gate controls the output of the cell state, and adds the extended state to obtain the hidden state h of the current time step. t , the hidden state contains the characteristic information of the time series at the current time step;
[0042] Repeat the above calculation process, processing each time step in turn until the entire time series is processed;
[0043] After multiple time steps of calculation, the hidden state h of XLSTM t Contains the characteristic information of the time series.
[0044] Furthermore, Transformer is used for prediction processing, specifically:
[0045] Convert the input sequence into a format that the model can process. If the input sequence is long, consider truncating or splitting it to fit the input length limit of the model.
[0046] Add position encoding so that the model can capture the order information of the input sequence. The position encoding is either absolute position encoding or relative position encoding.
[0047] Input the encoded input sequence into the Transformer model;
[0048] The model goes through the multi-head attention layer and the feedforward neural network layer in sequence. In the multi-head attention layer, the model pays attention to information at different positions at the same time, calculates the attention weights and performs weighted summation on the inputs. The feedforward neural network layer further processes the attention output.
[0049] After learning and training of multiple Transformer layers, the trained Transformer model is finally used to output the prediction results of each component, and the predicted value of the park power load is obtained through sequence reconstruction.
[0050] A park power load forecasting system, comprising:
[0051] An initial park power load data set generation module collects the park power load data and performs preliminary interpolation processing to obtain an initial park power load data set;
[0052] The initial park power load data set normalization and decomposition module normalizes the initial park power load data set, and then uses complementary set empirical mode decomposition to obtain multiple intrinsic mode components and a residual sequence;
[0053] The data training learning and prediction module builds the XLSTM feature extraction model, extracts features from the decomposition obtained by the complementary set empirical mode decomposition, uses the transformer model Transformer for prediction processing, and integrates the prediction results of each component to obtain the final prediction value of the future park power load.
[0054] Furthermore, the initial park power load data set normalization and decomposition module includes:
[0055] The normalization processing module normalizes the initial park power load data, scales it in the interval [0, 1], converts it into a dimensionless pure value, and uses the min-max normalization method to map the load to the interval [0, 1] and perform calculations;
[0056] The white noise adding module adds a certain amount of white noise n(t) to the normalized signal x(t) to form a new signal;
[0057] EMD decomposition module adds white noise to the signal x n (t) Perform EMD decomposition, adding different white noises each time to obtain the signal decomposition amount;
[0058] The IMF average processing module averages all IMFs to obtain the final IMF;
[0059] The original signal reconstruction module reconstructs the original signal using the obtained IMF and residual components.
[0060] A computer-readable storage medium storing a computer program, which implements the method described above when the program is executed by a processor.
[0061] The beneficial effects of the present invention are as follows:
[0062] The present invention uses complementary set empirical mode decomposition (CEEMD) to decompose the original park power load data into multiple modal components and a residual component, reducing the complexity of the original data and further extracting the main characteristic components in the data; the combined model XLSTM-Transformer (Extended Long Short-Term Memory Neural Network-Encoder Decoder Model) is used for prediction, wherein the XLSTM model further extracts and learns the features of the decomposed data, retains the main feature information in the original data as much as possible and provides it to the Transformer model for complete prediction. Compared with traditional prediction methods, the present invention has better data feature processing and capture effects, thereby improving the accuracy of park power load prediction and providing help for the park to reasonably dispatch power resources. BRIEF DESCRIPTION OF THE DRAWINGS
[0063] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.
[0064] Figure 1 This is a flow chart of an XLSTM-Transformer park power load forecasting method based on CEEMD decomposition in the present invention.
[0065] Figure 2 This is the XLSTM network structure diagram in the present invention.
[0066] Figure 3 This is the Transformer network structure diagram in the present invention.
[0067] Figure 4 This is the CEEMD decomposition result diagram of the present invention.
[0068] Figure 5 This is the fitting diagram of the prediction results.
[0069] Figure 6 is a system block diagram of the present invention;
[0070] Figure 7 It is a block diagram of the normalization and decomposition module of the initial park power load data set. DETAILED DESCRIPTION
[0071] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0072] like Figure 1 As shown, this embodiment provides an XLSTM-Transformer campus power load forecasting method based on CEEMD decomposition, and the specific steps are as follows:
[0073] The power load data of the park is collected and preliminarily interpolated to obtain the initial power load data set of the park. The nearest neighbor interpolation is used to process missing data. The specific process is as follows:
[0074] The present invention uses power load data of a region in South China for example demonstration, which includes data for the whole year of 2016, and the data sampling frequency is 15 minutes.
[0075] First, determine the known time data point set as {(t 1 ,y 1 ),(t 2 ,y 2 ),.....,(t n ,y n )} to perform interpolation processing, where t 1 ,t 2 ,.....,t n is a known time point, y 1 ,y 2 ,.....,y n is the corresponding time data value.
[0076] For the target time point t 0 , calculate its distance d from each known time point i =|t 0 -t i |, where i=1,2,....,n.
[0077] Find the time point t at which the distance is minimum k , that is, satisfying d k =min(d 1 ,d2 ,.....,d n ).
[0078] Set the time point t k The corresponding time data value y k As the target time point t 0 The interpolation result of .
[0079] Furthermore, the park power load data set is normalized, and then complementary set empirical mode decomposition (CEEMD) is used to obtain multiple intrinsic mode components and a residual sequence, which specifically includes the following process:
[0080] Normalization: The initial park power load data is scaled in the interval [0,1] and converted into a dimensionless pure value. The load is mapped to the interval [0,1] using the min-max normalization method. The calculation formula is as follows:
[0081]
[0082] Where: x max is the maximum value of the data, x min is the minimum value of the data;
[0083] 1) Add white noise: Add a certain amount of white noise n(t) to the normalized signal x(t) to form a new signal:
[0084] x n (t) = x(t) + n(t)
[0085] Among them, n(t) has a mean of 0 and a variance of σ 2 White noise;
[0086] 2) Perform EMD decomposition: Add white noise to the signal x n (t) Perform EMD decomposition, adding different white noise each time, and obtain the signal decomposition process:
[0087]
[0088] Among them, r N (t) is the residual component, N is the number of IMFs;
[0089] 3) Repeating process: Repeat steps 1) and 2) multiple times to obtain multiple new IMFs;
[0090] 4) Average IMF: Average all IMFs to get the final IMF:
[0091]
[0092] Where K is the number of repetitions, which can be set to 100 times based on experience;
[0093] 5) Reconstruct the original signal using the obtained IMF and residual components:
[0094]
[0095] Through the above steps, we can get Figure 4 The power load data shown in the figure is decomposed into components, and each component data is divided into a training set, a validation set, and a test set in a ratio of 8:1:1.
[0096] Further, if Figure 2 Use XLSTM to extract features from each component, set the number of XLSTM hidden layers to 3, the learning rate to 0.001, the Dropout rate to 0.2, and the batch size to 32. The specific process is as follows:
[0097] First, prepare the input time series data. Assume that we have a time series data set x = {x 1 ,x 2 ,.....,x T}, where x t represents the input vector at time step t, where T is the length of the time series;
[0098] Initialize various parameters of XLSTM, including forget gate, input gate, output gate, cell state and expansion state;
[0099] For time step t:
[0100] Calculate input gate i t : Input vector x t , the hidden state h of the previous time step t-1 , cell state c t-1 and extended state t-1 Perform linear transformations through different weight matrices respectively, then add the results and use the sigmoid activation function to get the value of the input gate i t The input gate determines how much new information can be stored in the cell state at the current time step;
[0101] Calculate the forget gate f t : Similarly, the input vector, the hidden state of the previous time step, the cell state, and the extended state are linearly transformed and activated by the sigmoid function to obtain the value of the forget gate f t , the forget gate determines how much information in the cell state of the previous time step needs to be forgotten;
[0102] Calculate the output gate o t :Calculate the output gate o in the same way t, which determines how much information of the cell state at the current time step can be output as a hidden state;
[0103] Calculate candidate cell states c~ t : The input vector and the hidden state of the previous time step are linearly transformed and activated by the tanh function to obtain the candidate cell state;
[0104] Update cell state c t : The inflow and outflow of information are controlled by the input gate and the forget gate, and the cell state of the previous time step and the candidate cell state are weighted summed to obtain the cell state of the current time step;
[0105] Calculate the extended state e t :The input vector, the hidden state of the previous time step, and the cell state are linearly transformed and activated by a specific function to obtain an extended state. This extended state provides the model with additional information processing and feature extraction capabilities;
[0106] Calculate the hidden state h t : The output gate controls the output of the cell state, and adds the extended state to obtain the hidden state of the current time step. This hidden state contains the characteristic information of the time series at the current time step;
[0107] Repeat this process, processing each time step in turn, until the entire time series is processed;
[0108] After multiple time steps of calculation, the hidden state h of XLSTM t Contains the characteristic information of the time series. So as to make the next prediction;
[0109] Further, if Figure 3 Use Transformer for prediction processing, set the number of Transformer multi-head attention heads to 8, the learning rate to 0.001, the Dropout rate to 0.2, and the batch size to 32. The details are as follows:
[0110] Convert the input sequence into a format that the model can process. If the input sequence is long, consider truncating or splitting it to fit the input length limit of the model;
[0111] Add position encoding so that the model can capture the order information of the input sequence. The position encoding can be absolute position encoding or relative position encoding, etc.
[0112] Input the encoded input sequence into the Transformer model;
[0113] The model goes through the multi-head attention layer, feedforward neural network layer, etc. In the multi-head attention layer, the model pays attention to information at different positions at the same time, calculates the attention weights and performs weighted summation on the inputs. The feedforward neural network layer further processes the attention output;
[0114] After learning and training of multiple Transformer layers, the trained Transformer model is finally used to output the prediction results of each component, and the predicted value of the park power load is obtained through sequence reconstruction.
[0115] An XLSTM-Transformer campus power load forecasting system based on CEEMD decomposition, such as Figure 6 As shown, including:
[0116] An initial park power load data set generation module collects the park power load data and performs preliminary interpolation processing to obtain an initial park power load data set;
[0117] The initial park power load data set normalization and decomposition module normalizes the initial park power load data set and then uses complementary set empirical mode decomposition to obtain multiple intrinsic mode components;
[0118] The data training learning and prediction module builds the XLSTM feature extraction model, extracts features from the decomposition obtained by the complementary set empirical mode decomposition, uses the transformer model Transformer for prediction processing, and integrates the prediction results of each component to obtain the final prediction value of the future park power load.
[0119] like Figure 7 As shown, the initial park power load data set normalization and decomposition module includes:
[0120] The normalization processing module normalizes the initial park power load data, scales it in the interval [0, 1], converts it into a dimensionless pure value, and uses the min-max normalization method to map the load to the interval [0, 1] and perform calculations;
[0121] The white noise adding module adds a certain amount of white noise n(t) to the normalized signal x(t) to form a new signal;
[0122] EMD decomposition module adds white noise to the signal x n (t) Perform EMD decomposition, adding different white noises each time to obtain the signal decomposition amount;
[0123] The IMF average processing module averages all IMFs to obtain the final IMF;
[0124] The original signal reconstruction module reconstructs the original signal using the obtained IMF and residual components.
[0125] By comparing the method of the present invention with the commonly used model LSTM, the following is obtained: Figure 5 From the fitting diagram of the method of the present invention and the LSTM model with the true value, it can be seen that the prediction accuracy of the method proposed in the present invention is more accurate, which is much higher than the prediction accuracy of the traditional model.
[0126] Those skilled in the art will readily appreciate other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses or adaptations of the present invention that follow the general principles of the present invention and include common knowledge or customary techniques in the art that are not disclosed by the present invention. The specification and examples are to be considered exemplary only, and the true scope and spirit of the present invention are indicated by the claims.
Claims
1. A method for predicting power load in a park, characterized in that: include: The park power load data is collected and preliminarily interpolated to obtain an initial park power load data set; The initial park power load data set is normalized, and then complementary set empirical mode decomposition is used to obtain multiple intrinsic mode components and a residual sequence; A feature extraction model of the extended long short-term memory neural network XLSTM is constructed, and feature extraction is performed on the decomposition obtained by the complementary set empirical mode decomposition. The transformer model Transformer is selected for prediction processing, and the prediction results of each component are integrated to obtain the final prediction value of the future park power load.
2. A method for predicting park power load according to claim 1, characterized in that: The collecting of the park power load data and performing preliminary interpolation processing to obtain an initial park power load data set includes: First, determine the known time data point set as {(t1,y1),(t2,y2),.....,(t n ,y n )}, where t1,t2,.....,t n are known time points, y1,y2,.....,y n is the corresponding time data value; For the target time point t0, calculate its distance d from each known time data point i =|t0-t i |, where i = 1, 2, ..., n; Find the time point t at which the distance is minimum k , that is, satisfying d k =min(d1,d2,.....,d n ); Set the time point t k The corresponding time data value y k As the interpolation result of the target time point t0.
3. A method for predicting park power load according to claim 1, characterized in that: The park power load data set is normalized, and then complementary set empirical mode decomposition (CEEMD) is used to obtain multiple intrinsic mode components and a residual sequence, including: 1) The initial park power load data is normalized, scaled in the interval [0, 1], converted into a dimensionless pure value, and the load is mapped to the interval [0, 1] using the min-max normalization method. The calculation formula is as follows: Where: x max is the maximum value of the data, x min is the minimum value of the data; 2) Add a certain amount of white noise n(t) to the normalized signal x(t) to form a new signal: x n (t)=x(t)+n(t) Among them, n(t) has a mean of 0 and a variance of σ 2 White noise; 3) Add white noise to the signal x n (t) Perform EMD decomposition, adding different white noise each time to obtain the signal decomposition amount: Among them, r N (t) is the residual component, N is the number of IMFs; 4) Repeat steps 2) and 3) multiple times to obtain multiple new IMFs; 5) Average all IMFs to get the final IMF: Among them, K is the number of repetitions, which is set to 100 times based on experience; 6) Reconstruct the original signal using the obtained IMF and residual components:
4. A method for predicting park power load according to claim 1, characterized in that: Use XLSTM to extract features from each component, specifically: First, prepare the input time series data. Suppose there is a time series data set x = {x1, x2, ....., x T }, where x t represents the input vector at time step t, where T is the length of the time series; Initialize various parameters of XLSTM, including forget gate, input gate, output gate, cell state and expansion state; Input vector x t , the hidden state h of the previous time step t-1 , cell state c t-1 and extended state t-1 Perform linear transformations through different weight matrices respectively, then add the results and use the sigmoid activation function to get the value of the input gate i t , the input gate determines how much new information can be stored in the cell state at the current time step; The input vector, the hidden state of the previous time step, the cell state and the extended state are linearly transformed and activated by the sigmoid function to obtain the value of the forget gate f. t ; The forget gate determines how much information in the cell state at the previous time step needs to be forgotten. The output gate o is calculated in the same way t , which determines how much information about the cell state at the current time step can be output as a hidden state; The input vector and the hidden state of the previous time step are linearly transformed and activated by the tanh function to obtain the candidate cell state c~ t ; The input gate and forget gate are used to control the inflow and outflow of information, and the cell state of the previous time step and the candidate cell state are weighted summed to obtain the cell state c of the current time step. t ; The input vector, the hidden state of the previous time step, and the cell state are linearly transformed and activated by a specific function to obtain the extended state e. t ; The output gate controls the output of the cell state, and adds the extended state to obtain the hidden state h of the current time step. t , the hidden state contains the characteristic information of the time series at the current time step; Repeat the above calculation process, processing each time step in turn until the entire time series is processed; After multiple time steps of calculation, the hidden state h of XLSTM t Contains the characteristic information of the time series.
5. A method for predicting park power load according to claim 1, characterized in that: Use Transformer for prediction processing, specifically: Convert the input sequence into a format that the model can process. If the input sequence is long, consider truncating or splitting it to fit the input length limit of the model. Add position encoding so that the model can capture the order information of the input sequence. The position encoding is either absolute position encoding or relative position encoding. Input the encoded input sequence into the Transformer model; The model goes through the multi-head attention layer and the feedforward neural network layer in sequence. In the multi-head attention layer, the model pays attention to information at different positions at the same time, calculates the attention weights and performs weighted summation on the inputs. The feedforward neural network layer further processes the attention output. After learning and training of multiple Transformer layers, the trained Transformer model is finally used to output the prediction results of each component, and the predicted value of the park power load is obtained through sequence reconstruction.
6. A park power load forecasting system, characterized in that: include; An initial park power load data set generation module collects the park power load data and performs preliminary interpolation processing to obtain an initial park power load data set; The initial park power load data set normalization and decomposition module normalizes the initial park power load data set and then uses complementary set empirical mode decomposition to obtain multiple intrinsic mode components; The data training learning and prediction module builds the XLSTM feature extraction model, extracts features from the decomposition obtained by the complementary set empirical mode decomposition, uses the transformer model Transformer for prediction processing, and integrates the prediction results of each component to obtain the final prediction value of the future park power load.
7. A park power load forecasting system according to claim 6, characterized in that: The initial park power load data set normalization and decomposition modules include; The normalization processing module normalizes the initial park power load data, scales it in the interval [0, 1], converts it into a dimensionless pure value, and uses the min-max normalization method to map the load to the interval [0, 1] and perform calculations; The white noise adding module adds a certain amount of white noise n(t) to the normalized signal x(t) to form a new signal; The EMD decomposition module adds white noise to the signal x n (t) Perform EMD decomposition, adding different white noises each time to obtain the signal decomposition amount; The IMF average processing module averages all IMFs to obtain the final IMF; The original signal reconstruction module reconstructs the original signal using the obtained IMF and residual components.
8. A computer-readable storage medium storing a computer program, characterized in that: When the program is executed by a processor, the method according to claim 1 to claim 5 is implemented.
Citation Information
Cited By
Fault parameter data prediction method
CN121765470A