基于ESN的光伏发电功率预测模型研究

Research of Prediction Model for Photovoltaic Power Based on ESN

针对传统神经网络学习算法复杂且稳定性差的问题,本文基于回声状态神经网络,提出了光伏发电功率预测模型。回声状态神经网络的隐含层是一种动态储备池结构,具有回声状态属性,不仅增强了网络预测的稳定性,而且只需采用线性算法即可求得网络输出权值,简化了训练过程,同时克服了传统神经网络收敛速度慢和易陷入局部极小的问题。并利用实际光伏发电站的历史数据和气象数据进行仿真验证。仿真结果表明,ESN预测模型的平均预测误差和最大预测误差分别比BP-NN预测模型提高了13.52%和102.26%,表明ESN预测模型的预测精度明显高于BP-NN预测模型;而且无论从预测精度还是稳定性,ESN预测模型都好于BP-NN预测模型,从而验证了ESN预测模型的可行性。该研究为光伏发电功率模型的实用化提供了理论基础。

In view of the complexity it and poor stability for traditional neural network learning algorithm, in this paper, the prediction model of photovoltaic power based on echo state neural network is put for- ward. The hidden layer of echo neural network is a kind of dynamic reserve pool structure, with the echo state attributes. It not only enhances the stability of network prediction, but also the weights of the net- work output can be obtained by linear algorithm. It simplifies the process of training, while solving the problem of the slow convergence speed of traditional neural network and it is easy to fall into local minima, and uses the actual historical and meteorological data of photovoltaic power station to simulate. The results show that, for the performance from the average prediction error and the maximum prediction error, the ESN prediction model is respectively improved by 13.52% and 102.26% compared with the BP-NN predic- tion model, proving that the accuracy of the ESN prediction model is obviously higher than that of the BP -NN prediction model; and both from the prediction accuracy and stability, the ESN prediction model is better than the BP- NN prediction model, which proves the feasibility of the ESN prediction model. The study provides a theoretical basis for the practical application of photovoltaic power model in the future.