基于LMD与GA-BP神经网络组合的短期风速滚动预测方法

Short-term Wind Speed Rolling Prediction Method Based on the Hybrid LMD and GA-BP Neural Network

针对组合预测方法中经验模态分解(EMD)部分存在处理非线性和非稳态信号的不足,提出以一种基于局部均值分解(LMD)与遗传算法-BP神经网络(GA-BP)模型相结合的短时风速预测方法。首先该方法在自适应分解部分采用LMD方法对原始风速数据进行分解得到一系列乘积函数,并对乘积函数分量提取瞬时频率和瞬时幅值的特征参数,然后通过设定的频率阈值将全部的乘积函数筛选重构成高、中、低3种频段分量;其次机器学习部分采用遗传算法优化BP神经网络方式进行建模,确定模型的连接权值和阈值参数;最后将3种不同频段分量输入到确定的GA-BP神经网络模型中进行滚动预测。通过仿真实验和对实测风速预测验证了所提方法的可行性和有效性,为短时预测方法提高了一种新思路。

Aiming at the shortcomings that the empirical mode decomposition(EMD)part in the combination forecasting method is not insufficient in processing non-linear and non-stationary signal,a short-term wind speed prediction method based on hybrid local mean decomposition(LMD)and genetic algorithm-BP neural network(GA-BP)model is proposed.In the proposed method,the original wind speed data is firstly decomposed into a series of product functions(PF)by LMD method in the adaptive decomposition part,and whose characteristic parameters of instantaneous frequency and instantaneous amplitude are extracted.In addition,the full product functions are screened and restructured into three frequency range components with high frequency,intermediate frequency and low frequency by setting the frequency threshold.And then,the genetic algorithm optimizing BP neural network method is utilized to build model in the machine learning part,and the connection weight and threshold parameter of the model are determined.In the last,the different three frequency range components are input into the built GA-BP neural network model for rolling prediction.The feasibility and effectiveness of the proposed method are verified by the simulation experiments and the measured wind speed experiments,which also provides a novel ideas for the short-term prediction method.