Probabilistic Prediction Research Of Wind Power Based On ALO-MAM-LSTM

In recent years,the concept of "low-carbon economy,strategic highland" has been paid increasing attention,and the development of clean energy is becoming the major economic policies of various countries.Dedicated to solving the grid of energy connection conflicts and power dispatching problems for some related departments,this paper proposed an ALO-MAM-LSTM based probabilistic prediction algorithm to forecast the level of wind energy precisely.The effectiveness of the proposed methodology is verified by experiments.To reduce the interference from noisy environment for wind power forecasting,an improved isolated forest algorithm is advanced to identify and separate noise from real wind turbine data.The proposed model is derived from the algorithm of self-planning feature classification,and intends to explore various types of multi-domain feature information entropy.Inputs are re-specified to evaluate the application scope of each influencing factor,which can effectively improve the quality of original data.Finally,comparisons with some traditional methods as well as machine learning algorithms with some metrics illustrate the superiority of proposed algorithms.Focus on the exploration of wind power prediction approaches and improving accuracy,autoregressive moving average model,a classic statistical algorithm,and the long and short-term memory network,a deep learning field time series algorithm,are systematically used as the basic prediction model.Then a prediction system can be established based on the cleaned data to construct a model with the highest accuracy.These two methods are compared in multiple perspectives and a best method for wind power prediction are explored by us.Furthermore,for the purposed of improving final prediction performance,the ensemble empirical mode decomposition algorithm is utilized in the proposed method.Frequency level components of the algorithm are all predicted,and the final fusion result shows the higher accuracy than the original algorithm.Experimental results demonstrates that the combination algorithm of pre-mining information association can reduce the prediction pressure of model as well as improve the prediction performance.To enhance the learning capability for source and terminal of multiple features area model,make the model exploring connection lever of source to source as well as source to terminal automatically and mining time series information in features,the MAM-LSTM algorithm is proposed to predict wind power.In addition to achieving attention learning at different ends,the number of neurons corresponding to the minimum function value is also selected to replace experience factor.Root mean square error value is adopted to set up the original optimization goal and construct fitness function.Finally,the selected neuron number corresponding to the minimum function value is used to replace empirical factors.To decrease the uncertainly of the point prediction and promote the application of engineering background theory,a wind power probabilistic prediction evaluation model is advanced by us.Furthermore,advantages of the existing interval evaluation index is analyzed from multiple angles to explore functions and shortcomings of various metrics.We constructed a Winkle score system to effectively screen each confidence interval,and select the most suitable confidence interval according to the actual wind field demand.