基于IVMD-PSO-LSTM模型的短期风速预测

Short-term Wind Speed Prediction Based on IVMD-PSO-LSTM Model

原始的风速序列是一种非线性和不平稳的风速序列,直接针对风速序列进行建模和预测精度不高。论文提出一种基于改进变分模态分解(Improved Variational Mode Decomposition,IVMD)的粒子群优化长短期记忆模型(Particle Swarm Optimization-Long Short-Term Memory,PSO-LSTM)网络相结合的方法对短期风速序列进行预测。IVMD算法能够自适应地确定分解层数,从而将原始风速序列转化为若干个不同频率、平稳的子序列,并具有良好的完备性。论文首先通过计算不同分解层数下的各个子序列的模糊熵值来为VMD算法选取合适的分解层数,然后采用VMD算法对原始风速序列进行计算分解得到一系列的平稳子序列,再通过对LSTM模型进行PSO算法优化来寻找最优参数,对子序列建立优化后的组合模型来进行预测,最后对子序列预测结果加总得到最终的预测结果。仿真结果表明,论文提出的IVMD-PSO-LSTM混合模型相较于BP、ARMA、LSTM单一模型预测精度更高,符合现有的风速预测标准。

The original wind speed sequence is a kind of nonlinear and unsteady wind speed sequence,and the modeling and prediction of wind speed sequence directly are not very accurate.In this paper,a model based on IVMD(Improved Variational Mode Decomposition)-PSO-LSTM(Particle Swarm Optimization-Long Short-Term Memory)network is proposed to predict the short-term wind speed sequence.The IVMD algorithm can adaptively determine the number of decomposition layers,so that the original wind speed sequence can be transformed into several different frequency,stationary sub-sequences,and has good com-pleteness.In this paper,an appropriate decomposition layer is selected for VMD algorithm by calculating the fuzzy entropy value of each sub-sequence under different decomposition layers.Then,the original wind speed sequence is computed and decomposed by VMD algorithm to obtain a series of stationary sub-sequences.Then,the optimal parameters are found by PSO algorithm optimiza-tion of LSTM model.The optimized combination model is established to predict the subsequence,and the final prediction result is obtained by summation of the prediction results of the subsequence.The simulation results show that the proposed hybrid model of IVMD-PSO-LSTM is more accurate than the single model of BP,ARMA and LSTM,and meets the existing wind speed prediction standards.