Load prediction method and load prediction system based on big data

Abstract

The invention discloses a load prediction method and a load prediction system based on big data, and relates to the technical field of power grid load prediction, and the method comprises the steps: carrying out the preprocessing of load historical data based on a least square support vector machine; decomposing the load historical data after complexity reduction to obtain a zero-crossing rate and a sample entropy, and determining a multi-frequency component of the load data; through a preset hybrid algorithm and in combination with the multi-frequency component of the load data, training to obtain a multimode factor weighted combination analysis model; on the basis of a multi-modal-factor weighted combination analysis model and according to an improved grey wolf algorithm, the weight of the load prediction result of each prediction modal factor module is determined; and performing weighted combination on the load prediction result of each prediction memetic module to obtain a final load prediction result. Through the combined prediction method, the model robustness is improved, the model output result is dynamically updated to optimize the weight ratio among the model factor modules, and the adaptive capacity of load prediction can be improved in a complex environment.

Description

Technical Field

[0001] The present invention belongs to the technical field of power grid load forecasting, and in particular relates to a load forecasting method and a load forecasting system based on big data. Background Art

[0002] With the rapid development of new power systems, electricity load scenarios within power systems are becoming increasingly diverse, which brings new challenges and opportunities to load forecasting and management. New power systems include industrial loads, commercial loads, residential loads, agricultural loads, as well as emerging electric vehicle loads, data center loads, etc. Each load has its own unique electricity consumption characteristics and changing patterns. With the access of renewable energy and the development of distributed energy, load volatility has become more complex, which puts higher requirements on the accuracy of load forecasting. At the same time, the application of smart grids and demand response technologies has made loads no longer passive recipients, but can interact with the power system. Therefore, the enhanced interactivity of loads requires load forecasting to take into account changes in user behavior and market demand. In order to meet these challenges, it is necessary to apply corresponding intelligent forecasting methods for different types of load forecasting scenarios to obtain more accurate and efficient load forecasting results, thereby significantly improving the load management capabilities of new power systems.

[0003] Patent publication number CN112100867B discloses a power grid load forecasting method, comprising: correcting the parameters of daily maximum load data and daily average temperature data according to a first correction coefficient and a second correction coefficient, respectively, to obtain a corrected GM(1,2) model; performing a cumulative subtraction operation on the corrected GM(1,2) model to obtain a power grid daily load forecasting model, and using the power grid daily load forecasting model to forecast the power grid for the next date. The above-mentioned power grid load forecasting method discloses predicting the power grid daily load forecasting model based on historical data. However, when inputting large amounts of data, traditional models have poor prediction accuracy for different types of data, resulting in high computing resource requirements.

[0004] Currently, no effective solutions have been proposed for the problems in related technologies. Summary of the Invention

[0005] In order to solve the problem in related technologies that traditional models have poor prediction accuracy for different types of data during the input of large amounts of data, resulting in high demand for computing resources, the present invention proposes a load forecasting method based on big data to overcome the above-mentioned technical problems existing in existing related technologies.

[0006] To this end, the specific technical solutions adopted in the present invention are as follows:

[0007] According to one aspect of the present invention, a load forecasting method based on big data is provided, comprising:

[0008] S1. Based on the least squares support vector machine, the load history data is nonlinearly transformed, and the Lagrangian function form is defined to obtain the vector regression function expression to reduce the complexity of the load history data.

[0009] S2. Decomposing the load history data after complexity reduction to obtain zero-crossing rate and sample entropy; using the zero-crossing rate and sample entropy, determining the multi-frequency components of the load data.

[0010] S3. Through the preset hybrid algorithm and combined with the multi-frequency components of the load data, a multi-meme weighted combination analysis model is trained.

[0011] S4. Based on the prediction performance, model complexity and confidence of each prediction meme module in the multi-meme weighted combination analysis model, and according to the improved grey wolf algorithm, the weight of the load prediction result of each prediction meme module is determined; the load prediction results of each prediction meme module are weightedly combined to obtain the final load prediction result.

[0012] Furthermore, based on the least squares support vector machine, the load history data is nonlinearly transformed, and the Lagrangian function form is defined to obtain the vector regression function expression to reduce the complexity of the load history data.

[0013] The training sample set corresponding to the load history data is collected, and the vector input in the training sample set is transformed into a nonlinear problem using nonlinear variables to obtain the function of the least squares support vector machine. Based on the function of the least squares support vector machine, the Lagrangian function is defined, and the spatial function is input after conditional optimization and solution to form the vector regression function expression:

[0014]

[0015] Where f3(x) represents the vector regression function, i represents an arbitrary non-zero constant, and x i represents the i-th input, b represents the deviation value of the adjacent uncertain regression function, K represents the spatial function, a i Represents the actual value of the sample, and x represents the new input variable to be predicted and regressed.

[0016] Furthermore, by decomposing the load history data after complexity reduction, the zero-crossing rate and sample entropy are obtained, including:

[0017] Configure the number of iterations of the noise-adaptive complete ensemble empirical mode decomposition and the added noise amplitude; decompose the historical data after complexity reduction into intrinsic mode function components and residual components through the noise-adaptive complete ensemble empirical mode; and obtain the zero-crossing rate and sample entropy from the intrinsic mode function components.

[0018] Furthermore, the zero-crossing rate and sample entropy are used to determine the multi-frequency components of the load data, including:

[0019] The high-frequency component, medium-frequency component and low-frequency component of the load data are determined by using the zero-crossing rate and sample entropy. The high-frequency component includes the influence of noise and sudden events, the medium-frequency component includes daily or weekly periodicity, and the low-frequency component includes long-term trends and seasonality.

[0020] Furthermore, through the preset hybrid algorithm and combined with the multi-frequency components of the load data, a multi-meme weighted combination analysis model is trained to include:

[0021] Through wavelet transform and convolutional neural network, combined with high-frequency components, the wavelet transform and convolutional neural network model is trained to predict high-frequency components; through empirical mode decomposition and recurrent neural network, combined with medium-frequency components, the recurrent neural network model is trained to predict medium-frequency components; through Fourier transform and random forest, combined with low-frequency components, the random forest model is trained to predict low-frequency components; the wavelet transform is combined with the convolutional neural network model, the recurrent neural network model and the random forest model to obtain a multi-meme weighted combination analysis model.

[0022] Furthermore, based on the prediction performance, model complexity and confidence of each prediction meme module in the multi-meme weighted combination analysis model, and according to the improved grey wolf algorithm, the weights of the load prediction results of each prediction meme module are determined, including:

[0023] Assign initial weights to each prediction meme module in the multi-meme weighted combination analysis model; obtain the prediction performance index, model complexity and confidence of each prediction meme module; utilize the iterative update of the improved grey wolf algorithm, combined with the fitness function including the prediction performance index, model complexity and confidence, to complete the weight optimization of the output results of each prediction meme module; wherein, the prediction meme module includes wavelet transform and convolutional neural network model, recurrent neural network model and random forest model.

[0024] Furthermore, the improved Grey Wolf Algorithm is used for iterative updates, and combined with a fitness function that includes prediction performance indicators, model complexity, and confidence, to optimize the weights of the output results of each prediction meme module. The following steps are involved:

[0025] Configure the parameters of the improved gray wolf algorithm, calculate the fitness value of each set of weight vectors based on the prediction performance indicators, model complexity and confidence of each prediction meme module, and the corresponding weight vector; iteratively update the position of the wolf pack according to the preset update rules to find the optimal weight vector; select the weight vector with the highest fitness value as the optimal weight.

[0026] Furthermore, the fitness function is:

[0027] Fitness = α·prediction performance + β·model complexity penalty + γ·confidence-adjusted fitness; where α, β, and γ represent the weight vectors corresponding to prediction performance, model complexity penalty, and confidence-adjusted fitness, respectively.

[0028] Furthermore, the prediction performance indicators of each prediction meme module are obtained including:

[0029] The prediction performance index of each prediction meme module on the test set is calculated by the mean square error.

[0030] According to another aspect of the present invention, a load forecasting system based on big data is provided, comprising:

[0031] The data preprocessing module is used to perform nonlinear transformation on the load history data based on the least squares support vector machine, define the Lagrangian function form, and solve the vector regression function expression to reduce the complexity of the load history data.

[0032] The data input matching module is used to decompose the load history data after complexity reduction to obtain the zero-crossing rate and sample entropy; and to determine the multi-frequency components of the load data using the zero-crossing rate and sample entropy.

[0033] The model training module is used to train a multi-meme weighted combination analysis model through a preset hybrid algorithm and combined with the multi-frequency components of the load data.

[0034] The weight combination module is used to determine the weight of the load forecast results of each prediction meme module based on the prediction performance, model complexity and confidence of each prediction meme module in the multi-meme weighted combination analysis model, and according to the improved grey wolf algorithm; the load forecast results of each prediction meme module are weightedly combined to obtain the final load forecast result.

[0035] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0036] In the present invention, through the combined prediction method, it is possible to utilize multiple meme modules to achieve the matching characteristics of different types of power grid load forecast data, improve the robustness of the model, and based on the rapid anomaly repair of the least squares support vector machine, utilize its linear equation solving characteristics to effectively reduce the computational complexity. By constructing a heterogeneous model cluster and accurately matching the algorithm characteristics with the data frequency domain characteristics, it can adapt to the challenges of data heterogeneity and diversified prediction needs in the big data environment. At the same time, it can dynamically update the model output results to optimize the weight ratio between each meme module, and improve the adaptability of load forecasting in complex environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. 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 any creative work.

[0038] Figure 1 is a flow chart of a load forecasting method based on big data according to an embodiment of the present invention;

[0039] Figure 2 The figure is a flow chart of the actual process proposed by the present invention. DETAILED DESCRIPTION

[0040] To further illustrate each embodiment, the present invention provides drawings, which are part of the disclosure of the present invention. They are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. By referring to these contents, ordinary technicians in this field should be able to understand other possible implementation methods and advantages of the present invention. The components in the figures are not drawn to scale, and similar component symbols are generally used to represent similar components.

[0041] According to an embodiment of the present invention, a load forecasting method and a load forecasting system based on big data are provided.

[0042] The present invention will now be further described with reference to the accompanying drawings and specific embodiments. Figure 1 According to one embodiment of the present invention, a load forecasting method based on big data is provided, comprising:

[0043] S1. Based on the least squares support vector machine, the load history data is nonlinearly transformed, and the Lagrangian function form is defined to obtain the vector regression function expression to reduce the complexity of the load history data.

[0044] S2. Decomposing the load history data after complexity reduction to obtain zero-crossing rate and sample entropy; using the zero-crossing rate and sample entropy, determining the multi-frequency components of the load data.

[0045] S3. Through the preset hybrid algorithm and combined with the multi-frequency components of the load data, a multi-meme weighted combination analysis model is trained.

[0046] S4. Based on the prediction performance, model complexity and confidence of each prediction meme module in the multi-meme weighted combination analysis model, and according to the improved grey wolf algorithm, the weight of the load prediction result of each prediction meme module is determined; the load prediction results of each prediction meme module are weightedly combined to obtain the final load prediction result.

[0047] In one embodiment, based on the least squares support vector machine, the load history data is nonlinearly transformed, and the Lagrangian function form is defined to solve the vector regression function expression to reduce the complexity of the load history data, including:

[0048] The training sample set corresponding to the load history data is collected, and the vector input in the training sample set is transformed into a nonlinear problem using nonlinear variables to obtain the function of the least squares support vector machine. Based on the function of the least squares support vector machine, the Lagrangian function is defined, and the spatial function is input after conditional optimization and solution to form the vector regression function expression:

[0049]

[0050] Where f3(x) represents the vector regression function, i represents an arbitrary non-zero constant, and x i represents the i-th input, b represents the deviation value of the adjacent uncertain regression function, K represents the spatial function, a i Represents the actual value of the sample, and x represents the new input variable to be predicted, that is, any point where the regression function value needs to be evaluated.

[0051] In one embodiment, the zero-crossing rate and sample entropy are obtained by decomposing the load history data after complexity reduction, including:

[0052] Configure the number of iterations of the noise-adaptive complete ensemble empirical mode decomposition and the added noise amplitude; decompose the historical data after complexity reduction into intrinsic mode function components and residual components through the noise-adaptive complete ensemble empirical mode; and obtain the zero-crossing rate and sample entropy from the intrinsic mode function components.

[0053] In one embodiment, determining the multi-frequency components of the load data using the zero-crossing rate and the sample entropy includes:

[0054] The high-frequency component, medium-frequency component and low-frequency component of the load data are determined by using the zero-crossing rate and sample entropy. The high-frequency component includes the influence of noise and sudden events, the medium-frequency component includes daily or weekly periodicity, and the low-frequency component includes long-term trends and seasonality.

[0055] In one embodiment, a multi-meme weighted combination analysis model is trained by a preset hybrid algorithm and combined with multi-frequency components of load data, including:

[0056] Through wavelet transform and convolutional neural network, combined with high-frequency components, the wavelet transform and convolutional neural network model is trained to predict high-frequency components; through empirical mode decomposition and recurrent neural network, combined with medium-frequency components, the recurrent neural network model is trained to predict medium-frequency components; through Fourier transform and random forest, combined with low-frequency components, the random forest model is trained to predict low-frequency components; the wavelet transform is combined with the convolutional neural network model, the recurrent neural network model and the random forest model to obtain a multi-meme weighted combination analysis model.

[0057] In one embodiment, based on the prediction performance, model complexity, and confidence of each prediction meme module in the multi-meme weighted combination analysis model, and according to the improved Grey Wolf algorithm, the weight of the load prediction result of each prediction meme module is determined, including:

[0058] Assign initial weights to each prediction meme module in the multi-meme weighted combination analysis model; obtain the prediction performance index, model complexity and confidence of each prediction meme module; utilize the iterative update of the improved grey wolf algorithm, combined with the fitness function including the prediction performance index, model complexity and confidence, to complete the weight optimization of the output results of each prediction meme module; wherein, the prediction meme module includes wavelet transform and convolutional neural network model, recurrent neural network model and random forest model.

[0059] In one embodiment, the improved grey wolf algorithm is used for iterative updating, and a fitness function including prediction performance index, model complexity, and confidence is combined to optimize the weight of the output results of each prediction meme module, including:

[0060] Configure the parameters of the improved gray wolf algorithm, calculate the fitness value of each set of weight vectors based on the prediction performance indicators, model complexity and confidence of each prediction meme module, and the corresponding weight vector; iteratively update the position of the wolf pack according to the preset update rules to find the optimal weight vector; select the weight vector with the highest fitness value as the optimal weight.

[0061] In one embodiment, the fitness function is:

[0062] Fitness = α·prediction performance + β·model complexity penalty + γ·confidence-adjusted fitness; where α, β, and γ represent the weight vectors corresponding to prediction performance, model complexity penalty, and confidence-adjusted fitness, respectively.

[0063] In one embodiment, obtaining the prediction performance indicator of each prediction meme module includes:

[0064] The prediction performance index of each prediction meme module on the test set is calculated by the mean square error.

[0065] According to another embodiment of the present invention, a load forecasting system based on big data is provided, including:

[0066] The data preprocessing module is used to perform nonlinear transformation on the load history data based on the least squares support vector machine, define the Lagrangian function form, and solve the vector regression function expression to reduce the complexity of the load history data.

[0067] The data input matching module is used to decompose the load history data after complexity reduction to obtain the zero-crossing rate and sample entropy; and to determine the multi-frequency components of the load data using the zero-crossing rate and sample entropy.

[0068] The model training module is used to train a multi-meme weighted combination analysis model through a preset hybrid algorithm and combined with the multi-frequency components of the load data.

[0069] The weight combination module is used to determine the weight of the load forecast results of each prediction meme module based on the prediction performance, model complexity and confidence of each prediction meme module in the multi-meme weighted combination analysis model, and according to the improved grey wolf algorithm; the load forecast results of each prediction meme module are weightedly combined to obtain the final load forecast result.

[0070] In order to facilitate understanding of the above technical solutions of the present invention, the working principle of the present invention in actual process is described in detail below.

[0071] like Figure 2 As shown, historical data is obtained and preprocessed. A multi-meme weighted combination analysis model is established. Multiple prediction models are trained using a preset algorithm, and prediction meme modules are established based on these multiple prediction models. The preprocessed historical data is then sequence-decomposed to obtain a historical data input sequence that matches the data processing characteristics of each prediction meme module. This input sequence is then fed into each prediction meme module. Each prediction meme module outputs a load forecast result based on the model's characteristics. Weights are assigned to the load forecast results of each prediction meme module, and the load forecast results of each prediction meme module are weighted and combined to obtain the final load forecast result.

[0072] Data preprocessing of historical data includes:

[0073] Optimization of time series data, specifically including preprocessing of power load data through least squares support vector machines:

[0074] Collect and set up the training sample set:

[0075] C={x i ,y i}|i=1,2,…,l|;

[0076] Among them, C represents the preset result, xi Indicates the input quantity, y i Indicates the output, l indicates the data length, and i indicates an arbitrary non-zero constant. Specify a training set function as f(x) = wx i +b; w represents the vector perpendicular to the hyperplane, and b represents the deviation value of the adjacent uncertain regression function.

[0077] Using nonlinear variables Convert the input vector into a nonlinear problem, then the function of the least squares support vector machine is:

[0078]

[0079] Where γ represents a positive parameter and e represents the sample deviation. Based on the function definition of the least squares support vector machine above, the Lagrangian function L(w,b,e,a) is expressed as:

[0080]

[0081]

[0082] in, represents the actual value of the sample, a and b are both real numbers, Represents a nonlinear transformation.

[0083] Optimize the conditions to get the expression:

[0084]

[0085] Finally, we get:

[0086]

[0087] Among them, E represents the weighted mean square error, Z represents the nonlinear transformation set,

[0088] Solve and get the vector regression function, the expression is:

[0089]

[0090] By inputting the spatial function K(x,x i ), the regression function expression is:

[0091]

[0092] Through the above processing of historical abnormal data, the change indicators of the non-missing data are input into the model (multi-meme weighted combination analysis model). The purpose is to reduce data complexity through data preprocessing.

[0093] The sequence decomposition of preprocessed historical data includes:

[0094] Initialize the parameters to set the number of iterations of the complete integrated empirical mode decomposition adapted to the noise and the amplitude of the added noise.

[0095] The original load sequence is decomposed into several intrinsic mode function components and a residual component through the complete integrated empirical mode decomposition of the adaptive noise. Each intrinsic mode function component represents a different frequency component in the original sequence. The residual component is an interpretation of the function component.

[0096] The intrinsic mode function components are classified by their zero-crossing rate and sample entropy. The zero-crossing rate represents the number of times a signal passes through zero per unit time, reflecting the frequency characteristics of the signal; the sample entropy is used to measure the complexity of the time series and reflects the randomness and regularity of the signal.

[0097] The load data signal characteristics (zero-crossing rate and sample entropy) are used to determine high-frequency, medium-frequency, and low-frequency components. These components are then analyzed through the characteristics of each prediction meme module. High-frequency components include the effects of noise and sudden events; medium-frequency components include daily or weekly periodicity; and low-frequency components include long-term trends and seasonality.

[0098] The multi-meme weighted combination analysis model uses a preset algorithm to train multiple prediction models, including:

[0099] Through wavelet transform-convolutional neural network, the high-frequency components are transformed by wavelet transform, the time-frequency features are extracted, a convolutional neural network model is constructed, the wavelet transformed features are input, and the wavelet transform-convolutional neural network model is trained to predict the high-frequency components.

[0100] Through empirical mode decomposition-recurrent neural network, the intermediate frequency components are subjected to empirical mode decomposition, the main modes are extracted, and a recurrent neural network model is constructed. The modes after empirical mode decomposition are input, and the recurrent neural network model is trained to predict the intermediate frequency components.

[0101] The low-frequency components are Fourier transformed through Fourier transform-random forest, the frequency domain features are extracted, and a random forest model is constructed. The Fourier transformed features are input and the random forest model is trained to predict the low-frequency components.

[0102] Assign weights to the load prediction results of each prediction meme module. The weights are determined based on the improved Grey Wolf Algorithm by taking into account the model's prediction performance on historical data, model complexity, and confidence. Specifically, the weights include:

[0103] A weight vector is defined for each model to indicate its contribution to the final prediction. The weight vector satisfies the constraint that the sum of all weights is 1.

[0104] Set parameters: including population size, maximum number of iterations, initial values ​​of control parameters and decay rate.

[0105] The historical data is divided into training set and test set, model training and prediction, the training set is trained through each prediction meme module, and prediction is made on the test set.

[0106] Performance index calculation,The prediction performance index of each model on the test set is calculated by the mean square error.

[0107] Prediction performance, the prediction performance index of the model on the test set is used as the component of the fitness function.

[0108] Based on the model complexity as the penalty term of the fitness function, according to the model confidence [confidence is calculated by selecting an appropriate method based on task requirements (such as classification, regression) and computing resources] or uncertainty estimation, adjust its weight in the fitness function, the formula is:

[0109] Fitness = α·prediction performance + β·model complexity penalty + γ·confidence-adjusted fitness.

[0110] Where α, β, and γ are preset weight coefficients, which are optimized using the Grey Wolf Algorithm. The model complexity penalty is a negative number.

[0111] Population initialization: Randomly generate a set of initial weight vectors as the initial position of the gray wolf population.

[0112] Fitness evaluation: Calculate the fitness value of each weight vector (α, β, γ).

[0113] Wolf pack update: According to the update rule of the control parameters, the wolf pack's position is iteratively updated to find the optimal weight vector.

[0114] The alpha wolf is the current optimal solution; the beta wolf and the delta wolf are suboptimal solutions; the other wolves are updated according to the positions of the alpha, beta, and delta wolves.

[0115] As the parameter update progresses through iterations, the values ​​of the control parameters are updated. After optimization, the weight vector with the highest fitness value is selected as the optimal weight to complete the weight optimization of the output results of each prediction meme module. The load forecast result is obtained by recombining the optimized weights.

[0116] A load forecasting system based on big data, comprising:

[0117] The historical data collection module is used to collect historical data of power grid load through the data collection terminal.

[0118] The data preprocessing module is used to preprocess the power load data to filter out abnormal data, reduce errors and improve prediction accuracy.

[0119] The multi-meme weighted combination analysis model has multiple prediction meme modules, which are trained by different algorithms and analyze preprocessed historical data through multiple prediction meme modules.

[0120] The data input matching module is used to process historical data into data sequences that match different prediction meme modules.

[0121] The weight combination module is used to perform weighted combination on the load forecast results output by the prediction meme module to obtain the final load forecast effect.

[0122] The plurality of prediction meme modules are respectively:

[0123] The wavelet transform-convolutional neural network prediction meme module uses the wavelet transform-convolutional neural network to perform wavelet transform on the high-frequency components, extract time-frequency features, build a convolutional neural network model, input the wavelet transformed features, and train the wavelet transform-convolutional neural network model to predict the high-frequency components.

[0124] The empirical mode decomposition-recurrent neural network prediction meme module uses empirical mode decomposition-recurrent neural network to perform empirical mode decomposition on the intermediate frequency component, extract the main mode, build a recurrent neural network model, input the mode after empirical mode decomposition, and train the recurrent neural network model to predict the intermediate frequency component.

[0125] The Fourier transform-random forest prediction meme module performs Fourier transform on the low-frequency components through Fourier transform-random forest, extracts frequency domain features, builds a random forest model, inputs the Fourier transformed features, and trains the random forest model to predict the low-frequency components.

[0126] In summary, through the combined prediction method, it is possible to utilize a variety of meme modules to achieve the matching characteristics of different types of power grid load forecast data, improve the robustness of the model, and quickly repair anomalies based on the least squares support vector machine. By utilizing its linear equation solving characteristics, the computational complexity can be effectively reduced. By constructing a heterogeneous model cluster and accurately matching the algorithm characteristics with the data frequency domain characteristics, it can adapt to the challenges of data heterogeneity and diversified prediction needs in the big data environment. At the same time, it can dynamically update the model output results to optimize the weight ratio between each meme module, and improve the adaptability of load forecasting in complex environments.

[0127] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A load forecasting method based on big data, characterized in that: include: S1. Based on the least squares support vector machine, the load history data is nonlinearly transformed, and the Lagrangian function form is defined to obtain the vector regression function expression to reduce the complexity of the load history data; S2. Decomposing the load history data after complexity reduction to obtain zero-crossing rate and sample entropy; using the zero-crossing rate and sample entropy, determining the multi-frequency components of the load data; S3. Using a preset hybrid algorithm and combining the multi-frequency components of the load data, a multi-meme weighted combination analysis model is trained; S4. Based on the prediction performance, model complexity and confidence of each prediction meme module in the multi-meme weighted combination analysis model, and according to the improved grey wolf algorithm, the weight of the load prediction result of each prediction meme module is determined; the load prediction results of each prediction meme module are weightedly combined to obtain the final load prediction result.

2. A load forecasting method based on big data according to claim 1, characterized in that: The least squares support vector machine is used to perform nonlinear transformation on the load history data, and the Lagrangian function form is defined to obtain the vector regression function expression to reduce the complexity of the load history data. Collect training sample sets corresponding to historical load data, and use nonlinear variables to transform the input vectors in the training sample sets into nonlinear problems to obtain the function of the least squares support vector machine; According to the function of the least squares support vector machine, the Lagrangian function is defined, and the spatial function is input after conditional optimization and solution to form the vector regression function expression: Where f3(x) represents the vector regression function, i represents an arbitrary non-zero constant, and x i represents the i-th input, b represents the deviation value of the adjacent uncertain regression function, K represents the spatial function, a i Represents the actual value of the sample, and x represents the new input variable to be predicted and regressed.

3. The load forecasting method based on big data according to claim 1, characterized in that: The zero-crossing rate and sample entropy obtained by decomposing the load history data after complexity reduction include: Configure the number of iterations of the complete ensemble EMD with adaptive noise and the added noise amplitude; The historical data with reduced complexity is decomposed into intrinsic mode function components and residual components through the complete integrated empirical mode of adaptive noise; Obtain the zero-crossing rate and sample entropy from the intrinsic mode function components.

4. The load forecasting method based on big data according to claim 1, characterized in that: Determining the multi-frequency components of the load data by using the zero-crossing rate and sample entropy includes: Using zero-crossing rate and sample entropy, the high-frequency component, medium-frequency component and low-frequency component of load data are determined; Among them, the high-frequency component contains the impact of noise and sudden events, the medium-frequency component contains daily or weekly periodicity, and the low-frequency component contains long-term trends and seasonality.

5. The load forecasting method based on big data according to claim 1, characterized in that: The multi-meme weighted combination analysis model obtained by training through a preset hybrid algorithm and combining the multi-frequency components of the load data includes: By combining wavelet transform and convolutional neural network with high-frequency components, wavelet transform and convolutional neural network models are trained to predict high-frequency components. By combining empirical mode decomposition and recurrent neural network with medium-frequency components, recurrent neural network models are trained to predict medium-frequency components. By combining Fourier transform and random forest with low-frequency components, random forest models are trained to predict low-frequency components. The wavelet transform is combined with the convolutional neural network model, the recurrent neural network model and the random forest model to obtain a multi-meme weighted combination analysis model.

6. The load forecasting method based on big data according to claim 1, characterized in that: The prediction performance, model complexity and confidence of each prediction meme module in the multi-meme weighted combination analysis model are determined based on the improved gray wolf algorithm to determine the weight of the load prediction result of each prediction meme module, including: Assigning initial weights to each prediction meme module in the multi-meme weighted combination analysis model; Obtain the prediction performance indicators, model complexity and confidence of each prediction meme module; The improved Grey Wolf Algorithm is used to iteratively update and combine the fitness function including prediction performance index, model complexity and confidence to optimize the weight of the output results of each prediction meme module. Among them, the meme prediction module includes wavelet transform and convolutional neural network model, recurrent neural network model and random forest model.

7. The load forecasting method based on big data according to claim 6, characterized in that: The iterative update of the improved Grey Wolf Algorithm, combined with a fitness function including prediction performance indicators, model complexity, and confidence, to complete the weight optimization of the output results of each prediction meme module includes: Configure the parameters of the improved gray wolf algorithm, and calculate the fitness value of each set of weight vectors based on the prediction performance index, model complexity and confidence of each prediction meme module, as well as the corresponding weight vector; According to the preset update rules, the positions of the wolf pack are iteratively updated to find the optimal weight vector; the weight vector with the highest fitness value is selected as the optimal weight.

8. The load forecasting method based on big data according to claim 6, characterized in that: The fitness function is: Fitness = α·prediction performance + β·model complexity penalty + γ·confidence-adjusted fitness; Where α, β, and γ represent the weight vectors corresponding to prediction performance, model complexity penalty, and confidence adjustment fitness, respectively.

9. The load forecasting method based on big data according to claim 6, characterized in that: The obtaining of the prediction performance indicators of each prediction meme module includes: The prediction performance index of each prediction meme module on the test set is calculated by the mean square error.

10. A load forecasting system based on big data, used to implement the load forecasting method based on big data according to any one of claims 1 to 9, characterized in that: include: The data preprocessing module is used to perform nonlinear transformation on the load history data based on the least squares support vector machine, define the Lagrangian function form, and solve the vector regression function expression to reduce the complexity of the load history data; The data input matching module is used to decompose the load history data after complexity reduction to obtain the zero-crossing rate and sample entropy; and to determine the multi-frequency components of the load data using the zero-crossing rate and sample entropy; The model training module is used to train a multi-meme weighted combination analysis model by combining a preset hybrid algorithm with the multi-frequency components of the load data; The weight combination module is used to determine the weight of the load forecast results of each prediction meme module based on the prediction performance, model complexity and confidence of each prediction meme module in the multi-meme weighted combination analysis model, and according to the improved grey wolf algorithm; the load forecast results of each prediction meme module are weightedly combined to obtain the final load forecast result.

Images