基于ICEEMDAN-PSO-LSTM的短期风速预测

Short-term Wind Speed Forecasting Based on ICEEMDAN-PSO-LSTM

提出一种改进自适应噪声完备集合经验模态分解与粒子群优化长短时记忆神经网络模型的短期风速预测方法。采用ICEEMDAN算法对日风速数据进行分解并计算相应边际谱,以谱相关性为依据对历史数据进行筛选;运用PSO算法优化LSTM神经网络参数,对输入数据进行ICEEMDAN分解,将所获得的多个模态分量分别用PSO-LSTM进行预测,并通过将各分量预测值叠加的方法得到风速预测结果。使用所提方法对国内某风电场风速进行预测,通过比较分析验证所提方法的有效性。

This article proposes a short-term wind speed prediction method based on improved adaptive noise complete set empirical mode decomposition(ICEEMDAN)and particle swarm optimization(PSO)long and short term memory neural network(LSTM)models.Use ICEEMDAN algorithm to decompose daily wind speed data and calculate corresponding marginal spectra,and screen historical data based on spectral correlation;Using PSO algorithm to optimize LSTM neural network parameters,ICEEMDAN decomposition is performed on the input data,and multiple modal components obtained are predicted using PSO-LSTM.The wind speed prediction results are obtained by overlaying the predicted values of each component.Use the proposed method to predict the wind speed of a domestic wind farm,and verify the effectiveness of the proposed method through comparative analysis.