Prediction is one of the research hotspots of applied machine learning and an important application of computer technology in practical engineering.However, due to the randomness and volatility of wind speed, the accuracy of wind speed prediction is low.In order to improve the accuracy of wind speed forecasting, we propose a new hybrid wind speed forecasting method by introducing the ensemble empirical mode decomposition (EEMD) into the composit differential evolution (CoDE) and the back propagation (BP) neural network.The original wind signal is decomposed by EEMD into several intrinsic mode functions (IMFs) with different frequencies and one residue, then a CoDE-BP neural network is used to model each of the extracted IMFs.The prediction results of all IMFs can be combined by weighted summation to obtain an aggregated output for wind speed.MATLAB simulation is carried out with wind speed datasets from a certain wind farm in Inner Mongolia at 10 min and 1 h sampling interval.Compared with the traditional Elman neural network (ENN), wavelet neural network (WNN), BP, CoDE-BP and EMD-CoDE-BP, it demonstrates that the proposed method can greatly improve the prediction accuracy and reduce prediction error.%预测问题是应用机器学习的研究热点之一, 是计算机技术领域在实际工程的重要应用, 然而由于风速具有随机性、波动性等特性, 导致风速预测存在准确率低的问题.为了提高风速预测的准确性, 将总体经验模态分解 (EEMD) 方法引入到组合差分进化算法 (CoDE) 和前馈 (BP) 神经网络中, 提出了一种新颖的混合风速预测模型 (EEMD-CoDE-BP) .利用EEMD将原始风速信号分解成一系列不同频率的子序列IMFs和残差序列r, 通过每个子序列训练CoDE-BP模型, 最终的风速预测结果由每个子序列预测结果等权求和得到.以国内某风电场每10 min、1 h采样间隔的风速数据进行MATLAB仿真, 对比包括传统的Elman神经网络 (ENN) 、小波神经网络 (WNN) 、BP、CoDE-BP和EMD-CoDE-BP等算法, 仿真结果表明所提方法能对风速进行准确有效的预测, 极大地提高了预测精度, 减小了预测误差.
展开▼
