Wind power forecasting method based on sparrow search algorithm optimized kernel extreme learning machine
By combining the sparrow search algorithm to optimize the core limit learning machine, process abnormal data and automatically find parameters, the problem of low wind power prediction accuracy is solved, and the wind power prediction effect with high precision and rapid convergence is achieved.
Patent Information
- Application Number
- CN202111247254.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-10-26
- Publication Date
- 2024-12-13
- Estimated Expiration
- 2041-10-26
AI Technical Summary
In the wind power power prediction method, the prediction performance of the intelligent algorithm model is greatly affected by its own parameters and the parameter optimization ability is insufficient, resulting in the problem of low prediction accuracy.
Combined with the sparrow search algorithm to optimize the nuclear limit learning machine, process abnormal data through Thompson tau-interquartile method, and use the sparrow search algorithm to automatically optimize the regularization coefficients and kernel function parameters of the nuclear limit learning machine, and build a wind power prediction model .
The accuracy and convergence speed of wind power prediction are improved, the generalization ability is strong, and the prediction root mean square error is below 80kW, which significantly improves the prediction accuracy and stability.
Smart Images
Figure CN114065610B_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of wind power prediction, and relates to a wind power prediction method based on a sparrow search algorithm-optimized kernel extreme learning machine. Background Art
[0002] As energy and environmental issues become increasingly prominent, wind energy, as a pollution-free, sustainable and clean energy, has become a focus of attention in the development of renewable energy. The level of domestic wind power generation has also maintained a steady growth. In 2018, wind power generation reached 366 billion kWh, accounting for 5.2% of the total power generation. Wind power forecasting has important guiding significance for power grid dispatching. Improving wind power forecasting algorithms and improving wind power forecasting accuracy are particularly important in this process.
[0003] At present, the methods for wind power prediction mainly include traditional prediction methods: time series method, regression analysis method, etc., and new prediction methods: neural network algorithm, support vector machine, extreme learning machine, genetic algorithm, etc. At present, the prediction performance of these intelligent algorithm models is still greatly affected by their own parameters, the parameter optimization ability is insufficient, the local search ability is poor and it is easy to fall into the local optimal solution, and the prediction accuracy is not high.
[0004] In order to improve the accuracy of wind power prediction, a sparrow search algorithm-optimized kernel extreme learning machine prediction model was constructed. The intelligent optimization algorithm and the extreme learning machine model were combined, and the sparrow search algorithm was used to automatically optimize the regularization coefficient and kernel function parameters of the kernel extreme learning machine, thereby improving the accuracy and convergence speed of wind power prediction. Summary of the invention
[0005] The purpose of the present invention is to provide a wind power prediction method based on a sparrow search algorithm optimized kernel extreme learning machine, which solves the current problems of low wind power prediction accuracy and prediction performance being greatly affected by its own parameters.
[0006] The technical solution adopted by the present invention is a wind power prediction method based on a sparrow search algorithm to optimize a kernel extreme learning machine, which specifically includes the following steps:
[0007] Step 1, determine the dominant factors affecting wind power;
[0008] Step 2: Use the Thompson tau-quartile method to process the outliers of the dominant influencing factors determined in step 1 to obtain a processed sample set;
[0009] Step 3: construct a prediction model of the extreme learning machine optimized by the sparrow search algorithm, and train it through the sample set obtained in step 2 to predict the wind power.
[0010] The present invention is also characterized in that:
[0011] The specific process of step 1 is: collecting wind power data and multidimensional meteorological data from wind power plants, where the multidimensional meteorological data includes: wind speed, wind direction, humidity, cabin temperature, and blade temperature; and determining through principal component analysis that the dominant influencing factors of wind power are wind speed and wind direction.
[0012] The specific process of step 2 is: use the Thompson tau-quartile method to segment the wind speed and wind direction data obtained in step 1, eliminate abnormal data points with extremely inconsistent power and wind speed due to recording errors, measuring instrument errors or equipment shutdown, and normalize the remaining data into standard values to obtain a cleaned sample set.
[0013] The specific process of step 3 is:
[0014] Step 3.1, determine the training parameters of the kernel extreme learning machine;
[0015] Step 3.2, taking wind speed and wind direction as the input of the kernel extreme learning machine and taking wind power as the output, substituting them into the cleaned sample set obtained in step 2, dividing the sample set into training samples and prediction samples, and using the sparrow search algorithm to automatically search for the training parameters in the kernel extreme learning machine;
[0016] In step 3.3, the trained training parameters are used in the kernel extreme learning machine, and the cleaned sample set obtained in step 2 is also substituted to perform wind power prediction and obtain the predicted power.
[0017] In step 3.1, the training parameters of the kernel extreme learning machine are: activation function, kernel function matrix and regularization coefficient.
[0018] In step 3.2, the regularization coefficient and the kernel function matrix are automatically searched for the kernel function matrix and the regularization coefficient in the kernel extreme learning machine using the sparrow search algorithm.
[0019] The beneficial effects of the present invention are as follows:
[0020] 1. The correlation between various meteorological data and power fluctuations is considered, and relevant meteorological factors are introduced for wind power forecasting after principal component analysis. Multidimensional meteorological factors are conducive to improving the prediction accuracy.
[0021] 2. Use Thompson tau-quartile to eliminate abnormal data and reduce the impact of abnormal data on prediction results.
[0022] 3. Use the four-point interpolation method to reconstruct the total data after removing outliers, reducing the impact of missing data on the prediction results.
[0023] 4. Construct a prediction model based on the sparrow search algorithm to optimize the kernel extreme learning machine, which has strong generalization ability, fast convergence speed of the optimization algorithm and high prediction accuracy.
[0024] 5. The predicted root mean square error is below 80kW. BRIEF DESCRIPTION OF THE DRAWINGS
[0025] Figure 1 It is a flow chart of the wind power prediction method based on the sparrow search algorithm optimization core extreme learning machine of the present invention;
[0026] Figure 2 It is the SSA-KELM power prediction diagram in the wind power prediction method based on the sparrow search algorithm optimization kernel extreme learning machine of the present invention;
[0027] Figure 3 It is a PSO-KELM power prediction diagram in the wind power prediction method based on the sparrow search algorithm optimization kernel extreme learning machine of the present invention. DETAILED DESCRIPTION
[0028] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.
[0029] The present invention is based on the wind power prediction method of the sparrow search algorithm optimization kernel extreme learning machine, such as Figure 1 As shown in the figure, the process is: 1. Use principal component analysis to select the dominant influencing factors; 2. Use Thompson tau-quartile method to eliminate abnormal historical power data; 3. Preprocess the data and normalize the data; 4. Input the cleaned data to train the sparrow search algorithm to optimize the kernel extreme learning machine model for prediction; 5. Denormalize the data; 6. Predict the power of multi-period series, draw the graph, and calculate the root mean square error; 7. Compare the actual prediction value to conduct error evaluation and analysis on the prediction results.
[0030] Specifically:
[0031] Step 1. First, collect wind power data and multidimensional meteorological data of a domestic wind power plant, including wind speed, wind direction, humidity, cabin temperature, and blade temperature. Through principal component analysis, determine that the dominant influencing factors of wind power are wind speed and wind direction, which account for 53.9234% and 27.7168% of the total respectively. 53.9234%+27.7168%>80%, so the first two principal components of wind speed and direction are retained.
[0032] Step 2: Use the Thompson tau-quartile method to segment the wind speed and wind direction data obtained in the first step to eliminate abnormal data points where the power and wind speed are extremely inconsistent due to recording errors, measuring instrument errors, or equipment shutdown. Normalize the data and convert it into a standard value to prevent the accuracy of the prediction results from being affected by the different ranges of wind speed and wind direction data.
[0033] Step 3, construct the sparrow search algorithm to optimize the kernel extreme learning machine prediction model (ssa-kelm) and complete the parameter initialization setting. The kernel extreme learning machine needs to determine three parameters before training, namely the activation function, the kernel function matrix and the regularization coefficient. The sig activation function (with the highest fitness for this group of data) is selected in the activation function of the kernel extreme learning machine, and the regularization coefficient and the kernel function matrix use the sparrow search algorithm to automatically search for the kernel function matrix and the regularization coefficient in the kernel extreme learning machine. For the sparrow optimization algorithm, it is necessary to determine its population size, maximum number of iterations, number of optimization parameters, risk ratio and upper and lower limits. The population size, maximum number of iterations and risk ratio can be freely selected (the larger the maximum number of iterations and the population size parameters, the longer the time consumption, and the higher the training accuracy in general). For example, in this paper, the population size is selected as 60, the maximum number of iterations is 30, the number of optimization parameters is 2, the risk ratio is 0.3, the upper limit is 1000, and the lower limit is 0.1. Substitute the two optimization parameters generated in the sparrow optimization algorithm into the regularization coefficient and kernel function matrix, then use wind speed and wind direction as the input of the kernel extreme learning machine, and use wind power as its output, and substitute them into the sample set after the second step of cleaning. This paper takes 4600 groups of data as training samples and 65 groups of data as prediction samples. Carry out wind power prediction and obtain the root mean square error between the predicted power and the actual power. The minimum RMSE (root mean square error) between the predicted power and the actual power is taken as the best fitness in the sparrow optimization algorithm, and the best kernel function matrix and regularization coefficient are selected iteratively. In the process of automatic parameter search of the sparrow optimization algorithm, the model will also complete the training, and finally complete the initial value setting and model training of the overall prediction model.
[0034] Step 4, perform denormalization on the predicted power and actual power samples obtained from the final training, so as to convert the standard values in the training process into normal power data, list the predicted power and actual power in a table one by one and make a comparison chart for easy observation and analysis. Calculate the root mean square error between the predicted value and the actual value, and perform error analysis and evaluation on the results. By analyzing the root mean square error size, the accuracy of the prediction of individual power points, and whether the trend between the overall predicted value and the actual value is consistent with the wind speed and direction, find the deficiencies of the model and propose new improvement directions. At the same time, the prediction accuracy of the model is judged based on these factors, mainly the root mean square error, so as to judge the adaptability of the model to such problems.
[0035] For the kernel extreme learning machine, prediction requires inputting power samples and the corresponding wind speed and direction, and selecting activation functions, kernel function matrices, and regularization coefficients. In the selection of kernel function matrices and regularization coefficients, manual selection is often time-consuming and laborious, and the results obtained are not accurate enough. The combination of the sparrow optimization algorithm for intelligent selection will make up for this defect.
[0036] The present invention focuses on introducing multidimensional meteorological data and uses the Thompson tau-quartile method to process historical abnormal data. A prediction model based on the sparrow search algorithm to optimize the kernel extreme learning machine is constructed to predict wind power, and the sparrow search algorithm is used to automatically optimize the regularization coefficient and kernel function parameters of the kernel extreme learning machine. Thus, wind power prediction is performed with the optimal regularization coefficient and kernel function parameters.
[0037] Figure 2 To compare the wind power prediction and true value of the kernel extreme learning machine based on the sparrow algorithm search method, Figure 2 The SSA-KELM power forecast diagram shows that the root mean square error of wind power forecast based on the sparrow algorithm to optimize the nuclear extreme learning machine is 60.1556kW. The prediction effect of the nuclear extreme learning machine is greatly affected by the parameter selection. This prediction method improves the stability and prediction accuracy of the prediction algorithm, has fast prediction speed, strong generalization ability, and good optimization effect. It has good scalability.
[0038] Compared with the same type of particle swarm optimization kernel extreme learning machine, this algorithm is progressive. The training prediction samples with the same parameters, the same search upper and lower limits, the same population size, and the maximum number of iterations are set. The root mean square error predicted by the particle swarm optimization kernel extreme learning machine is 133.6366kw. The comparison between the predicted value and the true value is shown in the figure below. Figure 3 As shown ( Figure 3(Figure 2) The PSO-KELM power prediction diagram: It is obvious that the SSA-KELM prediction value is closer to the true value and the prediction effect is more accurate. In addition, this prediction system can not only be applied to wind power prediction, but its prediction method is also applicable to stock prediction, house price prediction and weather prediction.
Claims
1. Wind power prediction method based on sparrow search algorithm optimized nuclear extreme learning machine, Features: The specific steps include: Step 1, determine the dominant factors affecting wind power; The specific process of step 1 is: collecting wind power data and multi-dimensional meteorological data from the wind farm, wherein the multi-dimensional meteorological data includes: wind speed, wind direction, humidity, cabin temperature, and blade temperature; determining that the dominant influencing factors of wind power are wind speed and wind direction by principal component analysis; Step 2, use the Thompson tau-quartile method to process the dominant influencing factors determined in step 1 to obtain a processed sample set; The specific process of step 2 is: using the Thompson tau-quartile method to segmentally identify the wind speed and wind direction data obtained in step 1, eliminating abnormal data points where the power and wind speed are extremely inconsistent due to recording errors, measuring instrument errors or equipment shutdown, and normalizing the remaining data into standard values to obtain a cleaned sample set; Step 3, constructing a prediction model of an extreme learning machine optimized by a sparrow search algorithm, and training it with the sample set obtained in step 2, so as to predict the wind power; the specific process of step 3 is as follows: Step 3.1, determine the training parameters of the kernel extreme learning machine; In step 3.1, the training parameters of the kernel extreme learning machine are: activation function, kernel function matrix and regularization coefficient; Step 3.2, using the sparrow search algorithm to automatically search for training parameters in the kernel extreme learning machine; In the step 3.2, the regularization coefficient and the kernel function matrix are automatically searched for the kernel function matrix and the regularization coefficient in the kernel extreme learning machine using the sparrow search algorithm; In step 3.3, wind speed and wind direction are used as inputs of the kernel extreme learning machine, and wind power is used as output. The cleaned sample set obtained in step 2 is substituted into the sample set, and the sample set is divided into training samples and prediction samples. Wind power prediction is performed to obtain the predicted power.