Power load prediction method
By using the LSTM-MF-MSSA model to train historical power load data, the problem of insufficient accuracy in the face of complex data is solved, and higher prediction accuracy and accuracy are achieved.
Patent Information
- Application Number
- CN202411568068.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-05
- Publication Date
- 2025-02-25
AI Technical Summary
When traditional power load prediction technology faces massive and complex power load data, prediction accuracy and accuracy are difficult to meet current needs.
The LSTM-MF-MSSA model is used to predict power loads. By obtaining historical power load data, an initial model is constructed, and the target model is obtained for prediction through training.
The prediction accuracy and accuracy of the prediction model are improved, effectively solving the problem of insufficient prediction accuracy and accuracy in the prior art.
Smart Images
Figure CN119518725A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of power systems, and in particular to a power load forecasting method. Background Art
[0002] Power load forecasting plays a vital role in power grid planning and dispatching. Its core lies in building a scientific and efficient forecasting model based on the historical power load data obtained, so as to accurately analyze the operation status of the power grid. If the forecast is inaccurate or inadequate, it will lead to a significant increase in operating and maintenance costs during the power generation cycle. With the deepening of smart grid informatization construction, power load data has become massive and complex. Although traditional forecasting technology has fast modeling and calculation speed, it has high requirements on the quality of power load data. Faced with the massive and complex power load data today, the prediction accuracy and precision of its forecasting model can no longer meet current needs. Summary of the invention
[0003] Based on this, it is necessary to propose a power load forecasting method to address the above problems. By adopting the LSTM-MF-MSSA model to predict the power load, the prediction accuracy and precision of the prediction model can be improved, thereby effectively solving the technical problems existing in the existing technology.
[0004] To achieve the above object, the present invention provides a method for predicting power load in a first aspect, the method comprising:
[0005] Acquire a plurality of historical power load values and a plurality of corresponding historical power load data, wherein each historical power load data includes a plurality of power load prediction factors;
[0006] Build the initial LSTM-MF-MSSA model;
[0007] Multiple historical power load values and corresponding multiple historical power load data are sequentially input into the initial LSTM-MF-MSSA model for training to obtain a target LSTM-MF-MSSA model, which is used to predict the first power load value.
[0008] Optionally, the expression of the MF-MSSA mechanism in the initial LSTM-MF-MSSA model is:
[0009]
[0010] Among them, SSA[LN(Z l )|U l,k ] is the feature weight value of the feature extracted from the lth load factor in the kth subspace, LN(Z l ) is the lth load factor, Z lThe feature extracted from the lth load factor conforms to the random variable of mixed Gaussian distribution, U l,k is the kth subspace of the features extracted from the lth load factor, softmax{} is the Sigmoid function, MSSA[LN(Z l )|U l,[1,...,k,...,K] ] is the feature weight value of the feature extracted by the lth load factor in all subspaces, U l,[1,...,k,...,K] All subspaces of features extracted for the lth load factor.
[0011] Optionally, the method further comprises:
[0012] Using the CR algorithm, the position of each cancer cell in the cancer cell population is iteratively updated to obtain the optimal LSTM hyperparameters corresponding to the global optimal fitness value of all cancer cells in each iteration, where the position of each cancer cell includes the LSTM hyperparameters;
[0013] A target CR-LSTM-MF-MSSA model is determined based on the optimal LSTM hyperparameters of each iteration, the target LSTM-MF-MSSA model, multiple historical power load values and corresponding multiple historical power load data, and the target CR-LSTM-MF-MSSA model is used to predict the second power load value.
[0014] Optionally, determining the target CR-LSTM-MF-MSSA model according to the optimal LSTM hyperparameters of each iteration, the target LSTM-MF-MSSA model, multiple historical power load values and corresponding multiple historical power load data includes:
[0015] According to the optimal LSTM hyperparameters of the first iteration, the LSTM hyperparameters in the target LSTM-MF-MSSA model are adjusted to obtain a first initial CR-LSTM-MF-MSSA model; a plurality of historical power load values and corresponding plurality of historical power load data are sequentially input into the first initial CR-LSTM-MF-MSSA model for training to obtain a first intermediate CR-LSTM-MF-MSSA model;
[0016] According to the optimal LSTM hyperparameters of the nth iteration, the LSTM hyperparameters in the n-1th intermediate CR-LSTM-MF-MSSA model are adjusted to obtain the nth initial CR-LSTM-MF-MSSA model; multiple historical power load values and corresponding multiple historical power load data are sequentially input into the nth initial CR-LSTM-MF-MSSA model for training to obtain the nth intermediate CR-LSTM-MF-MSSA model; wherein n successively takes integers greater than 1 until n is equal to the maximum number of iterations;
[0017] The nth intermediate CR-LSTM-MF-MSSA model corresponding to n being equal to the maximum number of iterations is used as the target CR-LSTM-MF-MSSA model.
[0018] Optionally, the CR algorithm is used to iteratively update the position of each cancer cell in the cancer cell population to obtain the optimal LSTM hyperparameters corresponding to the global optimal fitness value of all cancer cells in each iteration, including:
[0019] The position of each cancer cell in the cancer cell population is randomly initialized to obtain the target initial position of each cancer cell;
[0020] Determine the local optimal fitness value of each cancer cell according to the target initial position of each cancer cell in the cancer cell population, and use the local optimal fitness value of each cancer cell as the local poor fitness value of each cancer cell;
[0021] Determine the global optimal fitness value of all cancer cells according to the local optimal fitness value of each cancer cell;
[0022] According to the target initial position, the local optimal fitness value and the local poor fitness value of each cancer cell, and the global optimal fitness value of all cancer cells, the position of each cancer cell is iteratively updated until the number of iterations of the iterative update is equal to the maximum number of iterations, and the iterative update ends;
[0023] In the process of iterative updating,
[0024] For each cancer cell, determine the fitness value of the cancer cell at the t-th iteration, compare the fitness value of the cancer cell at the t-th iteration with the local poor fitness value of the cancer cell, if the fitness value of the t-th iteration is greater than the local poor fitness value, then use the fitness value of the t-th iteration as the local poor fitness value of the cancer cell;
[0025] For each cancer cell, determine the fitness value of the cancer cell at the t-th iteration, compare the fitness value of the cancer cell at the t-th iteration with the local optimal fitness value of the cancer cell, if the fitness value of the t-th iteration is less than the local optimal fitness value, then use the fitness value of the t-th iteration as the local optimal fitness value of the cancer cell;
[0026] For each cancer cell, determine the fitness value of the cancer cell at the t-th iteration, compare the fitness value of the cancer cell at the t-th iteration with the local optimal fitness value of the cancer cell. If the fitness value at the t-th iteration is not equal to the local optimal fitness value, use the cancer cell diffusion formula to update the position of the cancer cell at the t+1 iteration according to the position of the cancer cell at the t-1 iteration and the position corresponding to the local poor fitness value of the cancer cell; if the fitness value of the t-th iteration is equal to the local optimal fitness value, then the cancer cell at the t-th iteration is placed at the t-1 iteration. The fitness value of the t-1 iteration is compared with the local optimal fitness value of the cancer cell. When the fitness value of the t-1 iteration is not equal to the local optimal fitness value, the cancer cell competition formula is used to update the position of the cancer cell at the t+1 iteration according to the position of the cancer cell at the t iteration and the position corresponding to the local optimal fitness value of the cancer cell; when the fitness value of the t-1 iteration is equal to the local optimal fitness value, the cancer cell random mutation formula is used to update the position of the cancer cell at the t+1 iteration according to the position of the cancer cell at the t iteration;
[0027] For each cancer cell, determine the fitness value of the cancer cell at the t-th iteration, compare the fitness value of the cancer cell at the t-th iteration with the global optimal fitness value of all cancer cells, if the fitness value of the t-th iteration is less than the global optimal fitness value, then take the fitness value of the t-th iteration as the global optimal fitness value of all cancer cells, and take the fitness value of the t-th iteration as the global optimal fitness value of all cancer cells at the t-th iteration, and take the position of the cancer cell corresponding to the global optimal fitness value of the t-th iteration as the optimal LSTM hyperparameter corresponding to the global optimal fitness value of the t-th iteration.
[0028] Optionally, the cancer cell proliferation formula is:
[0029]
[0030] in, is the position of the i-th cancer cell in the cancer cell population at the t+1th iteration, is the position of the ith cancer cell in the cancer cell population at the tth iteration, r is the growth coefficient of the cancer cells in the cancer cell population, is the position of the ith cancer cell in the cancer cell population at the t-1th iteration, K is the position of the ith cancer cell in the cancer cell population at the tth iteration, b is the maximum number of cells that the cancer cell population can bear, b is the first random number, x i,worst is the position corresponding to the local poor fitness value of the i-th cancer cell in the cancer cell population.
[0031] Optionally, the cancer cell competition formula is:
[0032]
[0033] in, is the position of the ith cancer cell in the cancer cell population at the t+1th iteration, x i,optimal is the position corresponding to the local optimal fitness value of the ith cancer cell in the cancer cell population, r is the growth coefficient of the cancer cells in the cancer cell population, is the position of the ith cancer cell in the cancer cell population at the tth iteration, β is the competition coefficient of the cancer cells in the cancer cell population, max[] is the maximum value function, min[] is the minimum value function, Lb is the upper limit position of the cancer cells in the cancer cell population, Ub is the lower limit position of the cancer cells in the cancer cell population, T max is the maximum number of iterations.
[0034] Optionally, the cancer cell random mutation formula is:
[0035]
[0036] in, is the position of the i-th cancer cell in the cancer cell population at the t+1th iteration, is the position of the ith cancer cell in the cancer cell population at the tth iteration, sin() is the sine function, a is the second random number, π is the circumference of a circle, and x a The position of any cell in the preset normal cell population.
[0037] Optionally, the randomly initializing the position of each cancer cell in the cancer cell population to obtain a target initial position of each cancer cell includes:
[0038] The position of each cancer cell in the cancer cell population is randomly initialized to obtain the initial position of each cancer cell;
[0039] The initial position of each cancer cell is reinitialized using the Piecewise chaotic map to obtain the target initial position of each cancer cell.
[0040] Optionally, the using of the Piecewise chaotic map to reinitialize the initial position of each cancer cell to obtain the target initial position of each cancer cell includes:
[0041] Using the formula Reinitialize the initial position of each cancer cell to obtain the target initial position of each cancer cell;
[0042] in, is the target initial position of the i-th cancer cell in the cancer cell population, is the initial position of the i-th cancer cell in the cancer cell population, and P is the third random number.
[0043] To achieve the above object, the present invention provides a power load forecasting device in a second aspect, the device comprising:
[0044] An acquisition module, used to acquire a plurality of historical power load values and a plurality of corresponding historical power load data, wherein each historical power load data includes a plurality of power load prediction factors;
[0045] Construction module, used to build the initial LSTM-MF-MSSA model;
[0046] A training prediction module is used to input multiple historical power load values and corresponding multiple historical power load data into the initial LSTM-MF-MSSA model in sequence for training to obtain a target LSTM-MF-MSSA model, and the target LSTM-MF-MSSA model is used to predict the first power load value.
[0047] To achieve the above-mentioned object, the present invention provides in a third aspect a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the method as described in any one of the first aspects.
[0048] To achieve the above-mentioned purpose, the present invention provides a computer device in a fourth aspect, comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the method as described in any one of the first aspects.
[0049] The embodiment of the present invention has the following beneficial effects: the above method obtains multiple historical power load values and corresponding multiple historical power load data, each of which includes multiple power load prediction factors, and then constructs an initial LSTM-MF-MSSA model. Finally, the multiple historical power load values and the corresponding multiple historical power load data are sequentially input into the initial LSTM-MF-MSSA model for training to obtain a target LSTM-MF-MSSA model, which is used to predict the first power load value; that is, by using the LSTM-MF-MSSA model to predict the power load, the prediction accuracy and precision of the prediction model can be improved, thereby effectively solving the technical problems existing in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS
[0050] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0051] in:
[0052] Figure 1 A schematic diagram of a power load prediction method in an embodiment of the present application;
[0053] Figure 2 A schematic diagram showing comparison of iterative updates between the CR algorithm and the PLO algorithm in an embodiment of the present application;
[0054] Figure 3 This is a schematic diagram of the optimization of the CR algorithm in the embodiment of the present application;
[0055] Figure 4 A schematic diagram showing a comparison between an actual value and a predicted value of power load prediction in an embodiment of the present application;
[0056] Figure 5 A schematic diagram of a power load prediction device in an embodiment of the present application;
[0057] Figure 6 The internal structure of a computer device in some embodiments is shown. DETAILED DESCRIPTION
[0058] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions 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.
[0059] Power load forecasting plays a vital role in power grid planning and dispatching. Its core lies in building a scientific and efficient forecasting model based on the historical power load data obtained, so as to accurately analyze the operation status of the power grid. If the forecast is inaccurate or inadequate, it will lead to a significant increase in operating and maintenance costs during the power generation cycle. With the deepening of smart grid informatization construction, power load data has become massive and complex. Although traditional forecasting technology has fast modeling and calculation speed, it has high requirements on the quality of power load data. Faced with the massive and complex power load data today, the prediction accuracy and precision of its forecasting model can no longer meet current needs.
[0060] In response to the above problems, the present application proposes a power load forecasting method. By adopting the LSTM-MF-MSSA model to predict the power load, the prediction accuracy and precision of the prediction model can be improved, thereby effectively solving the technical problems existing in the prior art. The specific implementation principle will be described in detail in the following embodiments.
[0061] In a first aspect, the present application provides a method for predicting electric load.
[0062] See also Figure 1 , is a schematic diagram of a power load forecasting method in an embodiment of the present application, the method comprising:
[0063] Step 110: Acquire a plurality of historical power load values and a corresponding plurality of historical power load data, wherein each piece of historical power load data includes a plurality of power load prediction factors.
[0064] It should be noted that the historical power load values and the historical power load data are both historical data, and there is a one-to-one correspondence between the historical power load values and the historical power load data, so as to serve as the data set for the model training of this application.
[0065] In some embodiments, the multiple power load prediction factors may include, but are not limited to, humidity, wind speed, pressure, precipitation, visibility, water vapor pressure, and perceived temperature.
[0066] Step 120: Build an initial LSTM-MF-MSSA model.
[0067] Among them, the initial LSTM-MF-MSSA model here refers to the LSTM-MF-MSSA model that has not been trained; LSTM stands for Long Short-Term Memory; MF stands for Multi-Feature; MSSA stands for Mechanism Sparse Subspace Attention.
[0068] In some embodiments, an initial LSTM-MF-MSSA model may be constructed by introducing the MF-MSSA mechanism into the LSTM model.
[0069] In this application, by introducing the MF-MSSA mechanism into the LSTM model, the model prediction effect can be improved and made interpretable.
[0070] It is understandable that the MF-MSSA mechanism implements approximate gradient descent on the encoding rate of the features, and the subsequent multi-layer perceptron sparsifies the features, which produces a series of deep network architectures similar to white boxes, which are mathematically completely explainable; and in the present application, the MF-MSSA mechanism assigns different weights to the input features of the LSTM-MF-MSSA model and is explainable, thereby highlighting more critical influencing factors and helping the LSTM-MF-MSSA model to make more accurate judgments. Therefore, by introducing the MF-MSSA mechanism into the LSTM model, the model prediction effect can be improved and made explainable.
[0071] Step 130: Input multiple historical power load values and corresponding multiple historical power load data into the initial LSTM-MF-MSSA model in sequence for training to obtain a target LSTM-MF-MSSA model, which is used to predict the first power load value.
[0072] Among them, the target LSTM-MF-MSSA model here refers to the trained LSTM-MF-MSSA model, which can be directly used to predict the output first power load value based on the input power load data. Its power load data is similar to the historical power load data in the above step 110, and will not be repeated here.
[0073] It should be noted that after multiple historical power load values and corresponding multiple historical power load data are successively input into the initial LSTM-MF-MSSA model for training, after training to a certain extent, the trained target LSTM-MF-MSSA model is obtained; wherein, each historical power load value can be used as the true value of the initial BERT model, that is, by comparing the true value with each historical power load value output during the training process one by one, it can be determined whether the initial LSTM-MF-MSSA model is well trained and has met the expected requirements.
[0074] It should be further explained that after obtaining the target LSTM-MF-MSSA model, the power load data only needs to be input into the target LSTM-MF-MSSA model to automatically perform predictions and obtain the first power load value of the predicted output.
[0075] In an embodiment of the present application, by obtaining multiple historical power load values and corresponding multiple historical power load data, each historical power load data includes multiple power load prediction factors, and then constructing an initial LSTM-MF-MSSA model, and finally inputting the multiple historical power load values and the corresponding multiple historical power load data into the initial LSTM-MF-MSSA model in sequence for training, a target LSTM-MF-MSSA model is obtained, and the target LSTM-MF-MSSA model is used to predict the first power load value; that is, by using the LSTM-MF-MSSA model to predict the power load, the prediction accuracy and precision of the prediction model can be improved, thereby effectively solving the technical problems existing in the prior art.
[0076] In a feasible implementation, in step 110 of the above embodiment, the expression of the MF-MSSA mechanism in the initial LSTM-MF-MSSA model is:
[0077]
[0078] Among them, SSA[LN(Z l )|U l,k ] is the feature weight value of the feature extracted from the lth load factor in the kth subspace, LN(Z l ) is the lth load factor, Z l The feature extracted from the lth load factor conforms to the random variable of mixed Gaussian distribution, U l,k is the kth subspace of the features extracted from the lth load factor, softmax{} is the Sigmoid function, MSSA[LN(Z l )|U l,[1,...,k,...,K] ] is the feature weight value of the feature extracted by the lth load factor in all subspaces, U l,[1,...,k,...,K] All subspaces of features extracted for the lth load factor.
[0079] It should be noted that U l,[1,...,k,...,K] The K in refers to both the Kth subspace of the features extracted by the lth load factor and the total number of subspaces of the features extracted by the lth load factor.
[0080] It should be noted that in order to obtain the optimal feature weight value of the features extracted by each load factor, the method used in this application is different from other models that use the same sparse basis (ie, MSSA), that is, the sparse basis of the features extracted by each load factor is independent.
[0081] In the embodiments of the present application, a rigorous expression of the MF-MSSA mechanism in the LSTM-MF-MSSA model is provided from a mathematical perspective. The accuracy of the expression of the MF-MSSA mechanism can be ensured from the rigor of mathematical logic, and the above expression is preferably shown to provide reference and understanding for technical personnel; in addition, the MF-MSSA mechanism can effectively improve the prediction accuracy, enhance the interpretability of the model, adapt to complex data, optimize the calculation efficiency and support multi-feature fusion by assigning different weights to the input features of the LSTM-MF-MSSA model using the above expression.
[0082] It can be understood that improving prediction accuracy: the MF-MSSA mechanism can more carefully capture the characteristics of different load factors in different subspaces by introducing subspaces and attention mechanisms, thereby improving the prediction accuracy of the model; enhancing the interpretability of the model: by assigning different feature weight values to the characteristics of each load factor, the MF-MSSA mechanism enables the model to more clearly show which load factors have an important impact on the prediction results, thereby enhancing the interpretability of the model; adapting to complex data: in the face of massive and complex power load data, the MF-MSSA mechanism can effectively handle high-dimensional and complex data structures by sparsifying features and subspace partitioning, thereby improving the adaptability and robustness of the model; optimizing computational efficiency: by sparse feature weight values and subspace partitioning, the MF-MSSA mechanism may reduce the computational amount of the model to some extent, thereby improving computational efficiency; supporting multi-feature fusion: the MF-MSSA mechanism can handle multiple load factors and realize the fusion of multiple features through the attention mechanism, thereby making full use of multi-source information in the data.
[0083] In a feasible implementation, in step 110 of the above embodiment, the hidden layer of the LSTM model in the initial LSTM-MF-MSSA model consists of a forget gate, an input gate and an output gate, and the expression of the forget gate, the input gate and the output gate is the LSTM algorithm.
[0084] In the embodiment of the present application, the LSTM model can effectively handle the long-term dependency problem in sequence data through the collaborative work of the forget gate, input gate and output gate. The forget gate can decide which information needs to be forgotten, thereby avoiding the accumulation of irrelevant information. The input gate controls the addition of new information to ensure that only important information is stored. The output gate decides which information to output based on the current state, so that the LSTM can remember and process long-distance information.
[0085] In addition, the LSTM model effectively solves the gradient vanishing and gradient exploding problems in traditional RNN (recurrent neural network) through its three control gates, which improves the stability and training efficiency of the model and enables the model to learn and generalize better.
[0086] In a feasible implementation, the method in the above embodiment also includes: using the CR algorithm to iteratively update the position of each cancer cell in the cancer cell population, and obtain the optimal LSTM hyperparameters corresponding to the global optimal fitness value of all cancer cells in each iteration, wherein the position of each cancer cell includes the LSTM hyperparameters; determining the target CR-LSTM-MF-MSSA model according to the optimal LSTM hyperparameters of each iteration, the target LSTM-MF-MSSA model, multiple historical power load values and corresponding multiple historical power load data, and the target CR-LSTM-MF-MSSA model is used to predict the second power load value.
[0087] Among them, CR stands for Complete Remission; the target CR-LSTM-MF-MSSA model here refers to the model used to predict the output second power load value based on the input power load data. Its power load data is similar to the historical power load data in the above step 110, and will not be repeated here.
[0088] It should be noted that the CR algorithm refers to an algorithmic process in which the proliferation mode of tumor cancer cells is completely alleviated. In the present application, by using the CR algorithm, the position of each cancer cell in the cancer cell population is iteratively calculated, and the position is continuously adjusted along with the iterative calculation, so that it eventually tends to a stable state of global search for the optimal solution.
[0089] In some embodiments, a cancer cell population may be first formed by a group of randomly initialized cancer cells, each cancer cell having a position feature, and the position feature of each cancer cell is set as an LSTM hyperparameter here, that is, the position feature of each cancer cell is a multidimensional position feature, and its dimension is equal to the number of parameters in the LSTM hyperparameter. The position of each cancer cell is adjusted through each iterative operation to adjust the LSTM hyperparameter of each cancer cell, and then the fitness value of each cancer cell is calculated, and then the global optimal fitness value of all cancer cells in each iteration is determined, thereby determining the optimal LSTM hyperparameter corresponding to the global optimal fitness value of all cancer cells in each iteration.
[0090] In some embodiments, LSTM hyperparameters include but are not limited to learning rate, number of hidden layer neurons, and training rounds, etc.; of course, in other embodiments, LSTM hyperparameters may also only include learning rate and number of hidden layer neurons.
[0091] In some embodiments, when the LSTM hyperparameters include the learning rate and the number of neurons in the hidden layer, the position feature of each cancer cell is a two-dimensional position feature, that is, the position of each cancer cell includes the learning rate and the number of neurons in the hidden layer.
[0092] In some embodiments, the LSTM hyperparameters in the target LSTM-MF-MSSA model can be adjusted according to the optimal LSTM hyperparameters of the first iteration, and then multiple historical power load values and corresponding multiple historical power load data are input into the adjusted target LSTM-MF-MSSA model for training, and then the trained and adjusted target LSTM-MF-MSSA model is adjusted and trained again according to the optimal LSTM hyperparameters of the next iteration, that is, the target CR-LSTM-MF-MSSA model is obtained by repeated adjustment and training.
[0093] It should be further explained that after obtaining the target CR-LSTM-MF-MSSA model, the power load data only needs to be input into the target CR-LSTM-MF-MSSA model to automatically perform predictions and obtain the second power load value of the predicted output.
[0094] It can be understood that the predicted effect of the second power load value of the present application is much better than the predicted effect of the first power load value. This is because the use of the CR algorithm for iterative updating can obtain the optimal LSTM hyperparameters to adjust the target LSTM-MF-MSSA model, and then obtain the target CR-LSTM-MF-MSSA model, so that the power load is predicted through the target CR-LSTM-MF-MSSA model to further improve the prediction precision and accuracy.
[0095] In an embodiment of the present application, the CR algorithm is introduced to optimize the LSTM hyperparameters, so as to adjust the target LSTM-MF-MSSA model according to the LSTM hyperparameters, and then obtain the target CR-LSTM-MF-MSSA model, so as to predict the power load through the target CR-LSTM-MF-MSSA model, which can improve the prediction accuracy and precision, enhance the model generalization ability and improve the prediction efficiency.
[0096] It can be understood that improving prediction accuracy and precision: by iteratively updating the position of each cancer cell in the cancer cell population (i.e., LSTM hyperparameters) through the CR algorithm, it can globally search and tend to the stable state of the optimal solution, thereby finding the optimal LSTM hyperparameter combination. The optimal LSTM hyperparameters can significantly improve the performance of the model, thereby improving the prediction accuracy and accuracy of the power load; enhancing the generalization ability of the model: by optimizing the LSTM hyperparameters, the model can better adapt to different power load data and enhance the generalization ability of the model, which means that even in the face of new power load data that has never appeared in the data set, the model can give more accurate prediction results; improving prediction efficiency: because the CR algorithm can efficiently search for the optimal solution, the time to adjust to obtain the target CR-LSTM-MF-MSSA model is relatively short, which enables the model to give prediction results more quickly and in a timely manner, providing strong support for power scheduling and planning.
[0097] In a feasible implementation method, the above embodiment determines the target CR-LSTM-MF-MSSA model according to the optimal LSTM hyperparameters of each iteration, the target LSTM-MF-MSSA model, multiple historical power load values and corresponding multiple historical power load data, including: adjusting the LSTM hyperparameters in the target LSTM-MF-MSSA model according to the optimal LSTM hyperparameters of the first iteration to obtain a first initial CR-LSTM-MF-MSSA model; inputting the multiple historical power load values and the corresponding multiple historical power load data into the first initial CR-LSTM-MF-MSSA model in sequence for training to obtain a first intermediate CR-LSTM-MF-MS SA model; according to the optimal LSTM hyperparameters of the nth iteration, the LSTM hyperparameters in the n-1th intermediate CR-LSTM-MF-MSSA model are adjusted to obtain the nth initial CR-LSTM-MF-MSSA model; multiple historical power load values and corresponding multiple historical power load data are successively input into the nth initial CR-LSTM-MF-MSSA model for training to obtain the nth intermediate CR-LSTM-MF-MSSA model; wherein n successively takes integers greater than 1 until n is equal to the maximum number of iterations; the nth intermediate CR-LSTM-MF-MSSA model corresponding to n being equal to the maximum number of iterations is taken as the target CR-LSTM-MF-MSSA model.
[0098] The maximum number of iterations may be obtained by the operator based on a large amount of experience, experiments or statistics. Of course, the maximum number of iterations may also be set by the operator based on actual needs and is not limited here.
[0099] It should be noted that the training process of inputting multiple historical power load values and the corresponding multiple historical power load data into the nth initial CR-LSTM-MF-MSSA model for training to obtain the nth intermediate CR-LSTM-MF-MSSA model is similar to the above-mentioned training process of inputting multiple historical power load values and the corresponding multiple historical power load data into the initial LSTM-MF-MSSA model for training to obtain the target LSTM-MF-MSSA model, and will not be repeated here.
[0100] In the embodiment of the present application, by iteratively adjusting the LSTM hyperparameters to repeatedly adjust and train the CR-LSTM-MF-MSSA model performance to obtain the target CR-LSTM-MF-MSSA model, the model prediction accuracy can be improved, the model stability can be enhanced, the model training process can be optimized, and the development of power load forecasting technology can be promoted.
[0101] It can be understood that improving the prediction accuracy of the model: by adjusting the model according to the optimal LSTM hyperparameters of each iteration in an iterative manner, the model can be better adapted to the data characteristics, thereby improving the prediction accuracy, and inputting multiple historical power load values and corresponding multiple historical power load data into the model for training in sequence can help the model learn the potential laws and patterns in the data, thereby improving the accuracy of the prediction; enhancing the stability of the model: through multiple iterations and adjusting hyperparameters, the model can be prevented from overfitting the training data during the training process, thereby enhancing the generalization ability and stability of the model, and because the model is constantly learning and adjusting during the training process, it can better cope with noise and outliers in the data and improve the robustness of the model; Optimize the model training process: Automatically adjust the hyperparameters through iteration, which can reduce the workload of manual adjustment and improve the efficiency and automation of model training. In the iterative process, the hyperparameters can be continuously adjusted according to the performance of the model until the optimal model configuration is found, which facilitates the tuning and improvement of the model. Promote the development of power load forecasting technology: By combining the characteristics of LSTM and power load forecasting, the model performance is optimized by iteratively adjusting the hyperparameters, which helps to improve the overall level of power load forecasting technology. The optimized CR-LSTM-MF-MSSA model can be applied to more complex power load forecasting scenarios, providing more accurate and reliable forecasting support for the planning, scheduling and operation of the power system.
[0102] In a feasible implementation, the CR algorithm is used in the above embodiment to iteratively update the position of each cancer cell in the cancer cell population to obtain the optimal LSTM hyperparameters corresponding to the global optimal fitness value of all cancer cells in each iteration, including: randomly initializing the position of each cancer cell in the cancer cell population to obtain the target initial position of each cancer cell; determining the local optimal fitness value of each cancer cell according to the target initial position of each cancer cell in the cancer cell population, and using the local optimal fitness value of each cancer cell as the local poor fitness value of each cancer cell; determining the global optimal fitness value of all cancer cells according to the local optimal fitness value of each cancer cell; and determining the global optimal fitness value of all cancer cells according to the local optimal fitness value of each cancer cell. The target initial position of the cell, the local better fitness value and the local worse fitness value, as well as the global better fitness value of all cancer cells, start to iteratively update the position of each cancer cell until the number of iterations of the iterative update is equal to the maximum number of iterations, and the iterative update ends; in the process of iterative update, for each cancer cell, determine the fitness value of the cancer cell at the t-th iteration, compare the fitness value of the cancer cell at the t-th iteration with the local worse fitness value of the cancer cell, if the fitness value of the t-th iteration is greater than the local worse fitness value, then use the fitness value of the t-th iteration as the local worse fitness value of the cancer cell; for each cancer cell, determine the fitness value of the cancer cell at the t-th iteration, and compare the fitness value of the cancer cell at the t-th iteration with the local worse fitness value of the cancer cell. The fitness value of the cancer cell at the t-th iteration is compared with the local optimal fitness value of the cancer cell. If the fitness value of the t-th iteration is less than the local optimal fitness value, the fitness value of the t-th iteration is used as the local optimal fitness value of the cancer cell. For each cancer cell, the fitness value of the cancer cell at the t-th iteration is determined, and the fitness value of the cancer cell at the t-th iteration is compared with the local optimal fitness value of the cancer cell. If the fitness value of the t-th iteration is not equal to the local optimal fitness value, the cancer cell diffusion formula is used to update the cancer cell at the t-th iteration according to the position of the cancer cell at the t-1 iteration and the position corresponding to the local poor fitness value of the cancer cell. The position of the cancer cell at the t+1th iteration; if the fitness value of the tth iteration is equal to the local optimal fitness value, then the fitness value of the cancer cell at the t-1th iteration is compared with the local optimal fitness value of the cancer cell. When the fitness value of the t-1th iteration is not equal to the local optimal fitness value, the cancer cell competition formula is used to update the position of the cancer cell at the t+1th iteration according to the position of the cancer cell at the tth iteration and the position corresponding to the local optimal fitness value of the cancer cell; when the fitness value of the t-1th iteration is equal to the local optimal fitness value, the cancer cell random mutation formula is used to update the position of the cancer cell at the t+1th iteration according to the position of the cancer cell at the tth iteration;For each cancer cell, determine the fitness value of the cancer cell at the t-th iteration, compare the fitness value of the cancer cell at the t-th iteration with the global optimal fitness value of all cancer cells, if the fitness value of the t-th iteration is less than the global optimal fitness value, then take the fitness value of the t-th iteration as the global optimal fitness value of all cancer cells, and take the fitness value of the t-th iteration as the global optimal fitness value of all cancer cells at the t-th iteration, and take the position of the cancer cell corresponding to the global optimal fitness value of the t-th iteration as the optimal LSTM hyperparameter corresponding to the global optimal fitness value of the t-th iteration. ;
[0103] In an embodiment of the present application, the position of each cancer cell in the cancer cell population is iteratively updated by using the CR algorithm to obtain the optimal LSTM hyperparameters, which can optimize the LSTM hyperparameter search, improve model performance, enhance algorithm robustness, and promote algorithm innovation.
[0104] It is understandable that optimizing the LSTM hyperparameter search: the CR algorithm can effectively conduct a global search in the entire search space by simulating biological behaviors such as the diffusion, competition, and random mutation of cancer cells, avoiding falling into the local optimal solution, thereby finding a better LSTM hyperparameter combination, and in the iterative process, the CR algorithm dynamically adjusts the position of each cancer cell according to its fitness value, making the search process more flexible and adaptive. This adaptive adjustment mechanism helps the algorithm find the optimal solution in a complex and changeable data environment; improve model performance: by optimizing the LSTM hyperparameters, the prediction accuracy and generalization ability of the model can be improved, making it better adapted to different power load forecasts Scenarios and reasonable hyperparameter settings can accelerate the LSTM training process, reduce training time and computing resource consumption; enhance algorithm robustness: the CR algorithm enhances the algorithm's anti-noise ability by simulating the competition and mutation mechanisms in the biological evolution process, enabling it to find the optimal solution in power load data containing noise and outliers. As the CR algorithm has the ability of global search and adaptive adjustment, it has strong adaptability to power load data sets of different scales and complexities; promote algorithm innovation: combining evolutionary thinking in biology with optimization algorithms, it provides new ideas and methods for LSTM hyperparameter optimization, promoting algorithm innovation and technological progress.
[0105] In a feasible implementation, the cancer cell diffusion formula in the above embodiment is:
[0106]
[0107] in, is the position of the i-th cancer cell in the cancer cell population at the t+1th iteration, is the position of the ith cancer cell in the cancer cell population at the tth iteration, r is the growth coefficient of the cancer cells in the cancer cell population, is the position of the ith cancer cell in the cancer cell population at the t-1th iteration, K is the position of the ith cancer cell in the cancer cell population at the tth iteration, b is the maximum number of cells that the cancer cell population can bear, b is the first random number, x i,worst is the position corresponding to the local poor fitness value of the i-th cancer cell in the cancer cell population.
[0108] In some embodiments, the growth coefficient r of the cancer cells in the cancer cell population is 0.3, and the value range of the first random number b is [0, 1].
[0109] In the embodiments of the present application, a rigorous cancer cell diffusion formula for the iterative position of cancer cells in a cancer cell population is provided from a mathematical perspective. The rigor of mathematical logic can ensure the accuracy of the iterative position of cancer cells in the cancer cell population, and the above cancer cell diffusion formula is preferably shown to provide reference, understanding and calculation for technical personnel; in addition, by adopting the above cancer cell diffusion formula to calculate the position of cancer cells in a cancer cell population, a better LSTM hyperparameter combination can be found in a complex search space, thereby improving the prediction accuracy and performance of the model.
[0110] In a feasible implementation, the cancer cell competition formula in the above embodiment is:
[0111]
[0112] in, is the position of the ith cancer cell in the cancer cell population at the t+1th iteration, x i,optimal is the position corresponding to the local optimal fitness value of the ith cancer cell in the cancer cell population, r is the growth coefficient of the cancer cells in the cancer cell population, is the position of the ith cancer cell in the cancer cell population at the tth iteration, β is the competition coefficient of the cancer cells in the cancer cell population, max[] is the maximum value function, min[] is the minimum value function, Lb is the upper limit position of the cancer cells in the cancer cell population, Ub is the lower limit position of the cancer cells in the cancer cell population, T max is the maximum number of iterations.
[0113] In some embodiments, the competition coefficient β of cancer cells in the cancer cell population ranges from (0.1, 0.5).
[0114] In the embodiments of the present application, a rigorous cancer cell competition formula for the iterative position of cancer cells in a cancer cell population is provided from a mathematical perspective. The rigor of mathematical logic can ensure the accuracy of the iterative position of cancer cells in the cancer cell population, and the above-mentioned cancer cell competition formula is preferably shown to provide reference, understanding and calculation for technical personnel; in addition, by adopting the above-mentioned cancer cell competition formula to calculate the position of cancer cells in a cancer cell population, a better LSTM hyperparameter combination can be found in a complex search space, thereby improving the prediction accuracy and performance of the model.
[0115] In a feasible implementation, the random mutation formula of cancer cells in the above embodiment is:
[0116]
[0117] in, is the position of the i-th cancer cell in the cancer cell population at the t+1th iteration, is the position of the ith cancer cell in the cancer cell population at the tth iteration, sin() is the sine function, a is the second random number, π is the circumference of a circle, and x a The position of any cell in the preset normal cell population.
[0118] In some embodiments, the value range of the second random number a is [0, 1].
[0119] In the embodiments of the present application, a rigorous random variation formula of cancer cells in the iteration of the position of cancer cells in the cancer cell population is provided from a mathematical perspective. The rigor of mathematical logic can ensure the accuracy of the iteration of the position of cancer cells in the cancer cell population, and the above-mentioned random variation formula of cancer cells is preferably shown to provide reference, understanding and calculation for technical personnel; in addition, by adopting the above-mentioned random variation formula of cancer cells to calculate the position of cancer cells in the cancer cell population, a better LSTM hyperparameter combination can be found in a complex search space, thereby improving the prediction accuracy and performance of the model.
[0120] In a feasible implementation, the above embodiment randomly initializes the position of each cancer cell in the cancer cell population to obtain the target initial position of each cancer cell, including: randomly initializing the position of each cancer cell in the cancer cell population to obtain the initial position of each cancer cell; using Piecewise chaotic mapping to re-initialize the initial position of each cancer cell to obtain the target initial position of each cancer cell.
[0121] In an embodiment of the present application, by randomly initializing the position of each cancer cell in the cancer cell population and re-initializing it using a Piecewise chaotic map to obtain the target initial position of each cancer cell, the diversity of the initial positions can be improved, the search capability of the algorithm can be enhanced, the stability and reliability of the algorithm can be improved, and algorithm innovation and technology can be promoted.
[0122] It is understandable that the diversity of the initial positions can be improved by randomly initializing the position of each cancer cell in the cancer cell population, so that a set of initial positions with relatively uneven distribution can be obtained. The chaotic map reinitializes the initial position of each cancer cell to obtain a set of target initial positions with relatively uniform distribution, so as to further increase the complexity and diversity of the initial positions. The chaotic map has the characteristics of ergodicity and initial value sensitivity, and can generate a sequence with pseudo-randomness and unpredictability. Therefore, the initial position (i.e., the target initial position) processed by the chaotic map is more difficult to predict, which helps the algorithm to jump out of the local optimal solution in the subsequent iteration process; enhance the search ability of the algorithm: due to the increase in the diversity and complexity of the initial position, the algorithm can more comprehensively search the entire search space in the subsequent iteration process, which helps the algorithm to find a better solution and improve the search ability of the algorithm. By increasing the diversity and complexity of the initial position, the algorithm can more easily jump out of the local optimal solution during the search process, thereby avoiding the occurrence of premature convergence; improve the stability and reliability of the algorithm: due to the increase in the diversity and complexity of the initial position, the algorithm has strong adaptability to data sets of different sizes and complexities, which enables the algorithm to maintain stable performance when processing different types of data sets, and although the initial position is randomly generated, by using Pi ecewi The se chaotic map is reinitialized to ensure that the initial position sequence obtained each time the algorithm is run is consistent, which helps to improve the repeatability of the algorithm results and makes the performance of the algorithm more reliable; promote algorithm innovation and technological progress: combining chaos theory with optimization algorithms provides new ideas and methods for algorithm design. This interdisciplinary integration helps to promote algorithm innovation and technological progress and provide new solutions for research and application in related fields.
[0123] In a feasible implementation, the Piecewise chaotic mapping in the above embodiment is used to reinitialize the initial position of each cancer cell to obtain the target initial position of each cancer cell, including:
[0124] Using the formula Reinitialize the initial position of each cancer cell to obtain the target initial position of each cancer cell;
[0125] in, is the target initial position of the i-th cancer cell in the cancer cell population, is the initial position of the i-th cancer cell in the cancer cell population, and P is the third random number.
[0126] In some embodiments, the value of the third random number P is 0.4.
[0127] In the embodiments of the present application, a rigorous Piecewise chaotic mapping formula for reinitializing the positions of cancer cells in a cancer cell population is provided from a mathematical perspective. The accuracy of the reinitialization of the positions of cancer cells in a cancer cell population can be ensured from the rigor of mathematical logic, and the above-mentioned Piecewise chaotic mapping formula is preferably shown to provide reference, understanding and calculation for technical personnel. In addition, by adopting the above-mentioned Piecewise chaotic mapping formula to reinitialize the calculation of the positions of cancer cells in a cancer cell population, the diversity of initial positions can be improved, the search capability of the algorithm can be enhanced, the stability and reliability of the algorithm can be improved, and algorithm innovation and technology can be promoted.
[0128] In another feasible implementation, the present application has also studied another power load forecasting method, that is, replacing the CR algorithm of the present application with a PLO (Poly Light Optimization) algorithm, but its effect is not as good as the CR algorithm of the present application.
[0129] See also Figure 2 , is a comparative diagram of iterative updates between the CR algorithm and the PLO algorithm in the embodiment of the present application. It can be seen from the diagram that the CR algorithm can meet the global optimal output after iterative updates for a few times, while the PLO algorithm can only meet the global optimal output after iterative updates for hundreds of times, and the global optimal of the CR algorithm is smaller than the global optimal of the PLO algorithm. Therefore, the optimization effect of the PLO algorithm is not as good as that of the CR algorithm of the present application.
[0130] See also Figure 3 , which is a schematic diagram of the optimization of the CR algorithm in the embodiment of the present application. It can be seen from the schematic diagram that the CR algorithm of the present application can basically meet the global optimal output when it is iterated and updated six times, that is, in some embodiments, the maximum number of iterations can be set to 6.
[0131] See also Figure 4 , which is a schematic diagram for comparing the actual value and the predicted value of the power load forecast in the embodiment of the present application. It can be seen from the schematic diagram that the prediction precision and accuracy of the power load forecast of the present application are relatively high.
[0132] In a second aspect, the present application provides a power load prediction device.
[0133] See also Figure 5 , is a schematic diagram of a power load prediction device in an embodiment of the present application, the device 510 includes:
[0134] An acquisition module 511 is used to acquire a plurality of historical power load values and a plurality of corresponding historical power load data, wherein each historical power load data includes a plurality of power load prediction factors;
[0135] A construction module 512 is used to construct an initial LSTM-MF-MSSA model;
[0136] The training prediction module 513 is used to input multiple historical power load values and corresponding multiple historical power load data into the initial LSTM-MF-MSSA model in sequence for training to obtain a target LSTM-MF-MSSA model, and the target LSTM-MF-MSSA model is used to predict the first power load value.
[0137] In the embodiment of the present application, the relevant contents of the acquisition module 511, the construction module 512 and the training prediction module 513 can be found in Figure 1 The contents in the illustrated embodiment will not be described in detail here.
[0138] It should be noted that the device 510 of the present application also includes some other modules. It can be understood that the method of the present application and the device 510 have a one-to-one correspondence. Therefore, some other modules of the device 510 of the present application are the contents corresponding to the method of the present application in the above-mentioned embodiment.
[0139] In an embodiment of the present application, by obtaining multiple historical power load values and corresponding multiple historical power load data, each historical power load data includes multiple power load prediction factors, and then constructing an initial LSTM-MF-MSSA model, and finally inputting the multiple historical power load values and the corresponding multiple historical power load data into the initial LSTM-MF-MSSA model in sequence for training, a target LSTM-MF-MSSA model is obtained, and the target LSTM-MF-MSSA model is used to predict the first power load value; that is, by using the LSTM-MF-MSSA model to predict the power load, the prediction accuracy and precision of the prediction model can be improved, thereby effectively solving the technical problems existing in the prior art.
[0140] In a third aspect, the present application further provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor executes a power load forecasting method in the above method embodiment.
[0141] In a fourth aspect, the present application further provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes a power load forecasting method in the above method embodiment.
[0142] Figure 6 The internal structure diagram of a computer device in some embodiments is shown. The computer device may be a terminal, a server, or a gateway. Figure 6 As shown, the computer device includes a processor, a memory and a network interface connected through a system bus.
[0143] The memory includes a non-volatile storage medium and an internal memory. The non-volatile storage medium of the computer device stores an operating system and may also store a computer program. When the computer program is executed by the processor, the processor can implement each step in the above method embodiment. The internal memory may also store a computer program. When the computer program is executed by the processor, the processor can implement each step in the above method embodiment. Those skilled in the art can understand that Figure 6 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.
[0144] A person skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing related hardware through a computer program. The program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods.
[0145] Among them, any reference to memory, storage, database or other media used in the embodiments provided in the present application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synch li nk) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).
[0146] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0147] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.
Claims
1. A method for predicting power load, characterized in that: The method comprises: Acquire a plurality of historical power load values and a plurality of corresponding historical power load data, wherein each historical power load data includes a plurality of power load prediction factors; Build the initial LSTM-MF-MSSA model; Multiple historical power load values and corresponding multiple historical power load data are sequentially input into the initial LSTM-MF-MSSA model for training to obtain a target LSTM-MF-MSSA model, which is used to predict the first power load value.
2. The method according to claim 1, characterized in that: The expression of the MF-MSSA mechanism in the initial LSTM-MF-MSSA model is: Among them, SSA[LN(Z l )|U l,k ] is the feature weight value of the feature extracted from the lth load factor in the kth subspace, LN(Z l ) is the lth load factor, Z l The feature extracted from the lth load factor conforms to the random variable of mixed Gaussian distribution, U l,k is the kth subspace of the features extracted from the lth load factor, softmax{} is the Sigmoid function, MSSA[LN(Z l )|U l,[1,...,k,...,K] ] is the feature weight value of the feature extracted by the lth load factor in all subspaces, U l,[1,...,k,...,K] All subspaces of features extracted for the lth load factor.
3. The method according to claim 1, characterized in that: The method further comprises: Using the CR algorithm, the position of each cancer cell in the cancer cell population is iteratively updated to obtain the optimal LSTM hyperparameters corresponding to the global optimal fitness value of all cancer cells in each iteration, where the position of each cancer cell includes the LSTM hyperparameters; A target CR-LSTM-MF-MSSA model is determined based on the optimal LSTM hyperparameters of each iteration, the target LSTM-MF-MSSA model, multiple historical power load values and corresponding multiple historical power load data, and the target CR-LSTM-MF-MSSA model is used to predict the second power load value.
4. The method according to claim 3, characterized in that The method of determining the target CR-LSTM-MF-MSSA model according to the optimal LSTM hyperparameters of each iteration, the target LSTM-MF-MSSA model, multiple historical power load values and corresponding multiple historical power load data includes: According to the optimal LSTM hyperparameters of the first iteration, the LSTM hyperparameters in the target LSTM-MF-MSSA model are adjusted to obtain a first initial CR-LSTM-MF-MSSA model; a plurality of historical power load values and corresponding plurality of historical power load data are sequentially input into the first initial CR-LSTM-MF-MSSA model for training to obtain a first intermediate CR-LSTM-MF-MSSA model; According to the optimal LSTM hyperparameters of the nth iteration, the LSTM hyperparameters in the n-1th intermediate CR-LSTM-MF-MSSA model are adjusted to obtain the nth initial CR-LSTM-MF-MSSA model; multiple historical power load values and corresponding multiple historical power load data are sequentially input into the nth initial CR-LSTM-MF-MSSA model for training to obtain the nth intermediate CR-LSTM-MF-MSSA model; wherein n successively takes integers greater than 1 until n is equal to the maximum number of iterations; The nth intermediate CR-LSTM-MF-MSSA model corresponding to n being equal to the maximum number of iterations is used as the target CR-LSTM-MF-MSSA model.
5. The method according to claim 3, characterized in that: The CR algorithm is used to iteratively update the position of each cancer cell in the cancer cell population to obtain the optimal LSTM hyperparameters corresponding to the global optimal fitness value of all cancer cells in each iteration, including: The position of each cancer cell in the cancer cell population is randomly initialized to obtain the target initial position of each cancer cell; Determine the local optimal fitness value of each cancer cell according to the target initial position of each cancer cell in the cancer cell population, and use the local optimal fitness value of each cancer cell as the local poor fitness value of each cancer cell; Determine the global optimal fitness value of all cancer cells according to the local optimal fitness value of each cancer cell; According to the target initial position, the local optimal fitness value and the local poor fitness value of each cancer cell, and the global optimal fitness value of all cancer cells, the position of each cancer cell is iteratively updated until the number of iterations of the iterative update is equal to the maximum number of iterations, and the iterative update ends; In the process of iterative updating, For each cancer cell, determine the fitness value of the cancer cell at the t-th iteration, compare the fitness value of the cancer cell at the t-th iteration with the local poor fitness value of the cancer cell, if the fitness value of the t-th iteration is greater than the local poor fitness value, then use the fitness value of the t-th iteration as the local poor fitness value of the cancer cell; For each cancer cell, determine the fitness value of the cancer cell at the t-th iteration, compare the fitness value of the cancer cell at the t-th iteration with the local optimal fitness value of the cancer cell, if the fitness value of the t-th iteration is less than the local optimal fitness value, then use the fitness value of the t-th iteration as the local optimal fitness value of the cancer cell; For each cancer cell, determine the fitness value of the cancer cell at the t-th iteration, compare the fitness value of the cancer cell at the t-th iteration with the local optimal fitness value of the cancer cell. If the fitness value at the t-th iteration is not equal to the local optimal fitness value, use the cancer cell diffusion formula to update the position of the cancer cell at the t+1 iteration according to the position of the cancer cell at the t-1 iteration and the position corresponding to the local poor fitness value of the cancer cell; if the fitness value of the t-th iteration is equal to the local optimal fitness value, then the cancer cell at the t-th iteration is placed at the t-1 iteration. The fitness value of the t-1 iteration is compared with the local optimal fitness value of the cancer cell. When the fitness value of the t-1 iteration is not equal to the local optimal fitness value, the cancer cell competition formula is used to update the position of the cancer cell at the t+1 iteration according to the position of the cancer cell at the t iteration and the position corresponding to the local optimal fitness value of the cancer cell; when the fitness value of the t-1 iteration is equal to the local optimal fitness value, the cancer cell random mutation formula is used to update the position of the cancer cell at the t+1 iteration according to the position of the cancer cell at the t iteration; For each cancer cell, determine the fitness value of the cancer cell at the t-th iteration, compare the fitness value of the cancer cell at the t-th iteration with the global optimal fitness value of all cancer cells, if the fitness value of the t-th iteration is less than the global optimal fitness value, then take the fitness value of the t-th iteration as the global optimal fitness value of all cancer cells, and take the fitness value of the t-th iteration as the global optimal fitness value of all cancer cells at the t-th iteration, and take the position of the cancer cell corresponding to the global optimal fitness value of the t-th iteration as the optimal LSTM hyperparameter corresponding to the global optimal fitness value of the t-th iteration.
6. The method according to claim 5, characterized in that The cancer cell proliferation formula is: in, is the position of the ith cancer cell in the cancer cell population at the t+1th iteration, is the position of the ith cancer cell in the cancer cell population at the tth iteration, r is the growth coefficient of the cancer cells in the cancer cell population, is the position of the ith cancer cell in the cancer cell population at the t-1th iteration, K is the position of the ith cancer cell in the cancer cell population at the tth iteration, b is the maximum number of cells that the cancer cell population can bear, b is the first random number, x i,worst is the position corresponding to the local poor fitness value of the i-th cancer cell in the cancer cell population.
7. The method according to claim 5, characterized in that The cancer cell competition formula is: in, is the position of the ith cancer cell in the cancer cell population at the t+1th iteration, x i,optimal is the position corresponding to the local optimal fitness value of the ith cancer cell in the cancer cell population, r is the growth coefficient of the cancer cells in the cancer cell population, is the position of the ith cancer cell in the cancer cell population at the tth iteration, β is the competition coefficient of the cancer cells in the cancer cell population, max[] is the maximum value function, min[] is the minimum value function, Lb is the upper limit position of the cancer cells in the cancer cell population, Ub is the lower limit position of the cancer cells in the cancer cell population, T max is the maximum number of iterations.
8. The method according to claim 5, characterized in that The random mutation formula of cancer cells is: in, is the position of the ith cancer cell in the cancer cell population at the t+1th iteration, is the position of the ith cancer cell in the cancer cell population at the tth iteration, sin() is the sine function, a is the second random number, π is the circumference of a circle, and x a The position of any cell in the preset normal cell population.
9. The method according to claim 5, characterized in that The step of randomly initializing the position of each cancer cell in the cancer cell population to obtain a target initial position of each cancer cell includes: The position of each cancer cell in the cancer cell population is randomly initialized to obtain the initial position of each cancer cell; The initial position of each cancer cell is reinitialized using Piecewise chaotic mapping to obtain the target initial position of each cancer cell.
10. The method according to claim 9, characterized in that The Piecewise chaotic mapping is used to reinitialize the initial position of each cancer cell to obtain the target initial position of each cancer cell, including: Using the formula Reinitialize the initial position of each cancer cell to obtain the target initial position of each cancer cell; in, is the target initial position of the i-th cancer cell in the cancer cell population, is the initial position of the i-th cancer cell in the cancer cell population, and P is the third random number.