基于离散Hopfield模式识别样本的GRNN非线性组合短期风速预测模型

GRNN nonlinear combination model for short-term wind speed forecasting based on discrete Hopfield pattern recognition

利用实时风速数据,建立基于离散Hopfield模式识别样本的广义回归神经网络(GRNN)非线性组合预测模型。在风速数据样本集经二维小波阈值去噪处理后,基于离散Hopfield识别历史数据中与待预测样本最相似的数据,并作为训练样本;将支持向量机、BP神经网络和Elman神经网络分别进行单项预测的结果作为输入向量,经GRNN进行非线性组合预测。采用某风电场的实际风速数据进行预测,结果验证了该预测模型的正确性和有效性。

With the real-time wind speed data,a GRNN （General Regression Neural Network） nonlinear combination model based on discrete Hopfield pattern recognition is built for short-term wind speed forecasting. After the wind speed data sample set is processed by the two-dimension wavelet threshold de- noising method,the historic data most similar to the sample to be forecasted are picked out by the discrete Hopfield identification and taken as the training sample. The forecasting results obtained respectively by the support vector machine,BP neural network and Elman neural network are taken as the input vectors of GRNN for the nonlinear combination forecasting. The forecasting results based on the real wind speed data of a wind farm verify the correctness and effectiveness of the proposed forecasting model.