基于改进GMDH网络的风电场短期风速预测

摘要

Based on traditional GMDH metwork and fuzzy logic theory, the wind speed forecasting in wind farm is analyzed and an improved GMDH neural network is proposed. This method introduces feedback loop to the traditional network and makes the neuron fuzzified. The low-level computational ability of GMDH and the high-level reasoning ability of fuzzy logic are combined in the improved network for predicting. Besides, exponential energy function is taken into network training as the objective error function,which improves the speed of network convergence. Comparing the forecasting result by the proposed method with those from BP neural network and traditional GMDH network, the accuracy of the improved method in short-term wind speed forecasting is proved effectively.%基于GMDH神经网络和模糊逻辑理论,对风电场风速预测进行了深入研究,提出了一种改进GMDH神经网络方法.该方法在传统网络的基础上将神经元模糊化并引入反馈环,将GMDH网络的低维计算能力和模糊逻辑的高维推理能力结合起来用于预测.在进行网络训练时,采用指数型能量函数作为目标误差函数,提高了网络收敛速度.通过与BP神经网络及传统GMDH网络的预测结果相比较,表明该改进方法能够有效地提高短期风速预测的精度.