基于拟牛顿法小波神经网络的光伏发电系统短期功率预测模型

光伏发电系统出力的随机性会对大电网造成冲击,需要加强光伏阵列发电功率预测的研究.为此,提出采用拟牛顿法小波神经网络建立光伏发电系统短期功率预测模型.以某光伏电站实测数据为比较对象,与基于标准梯度下降法BP神经网络以及基于附加动量和自适应学习速率结合的BP神经网络建立的2种预测模型进行对比研究,结果表明,拟牛顿法在收敛速度和预测精度上都更具有优势.此外,通过和拟牛顿法BP神经网络功率预测方法对比表明,拟牛顿法小波神经网络的预测精度更高,尤其是在一天早中晚时刻或辐照度较低情况下预测效果得到了很大的提高.

A Forecasting Method of Short-Term Power Output of Photovoltaic System Based on Wavelet Neural Network Trained by Quasi-Newton Method

The randomness of the power output of a photovohaic system has an impact on the power grid. So it is needed to strengthen the study of the power output forecasting of photovohaic systems. In this paper, the short-term forecasting model of PV station power, based on wavelet neural network which is trained by quasi-Newton method, was proposed. The comparison experiments were made for the forecasting model above and the forecasting models based on BP neural network which is trained by quasi-Newton methed, traditional BP algorithm or combining increasing momentum method with varying learning rate method. The experimental results indicate that the short-term forecasting model of PV station power, based on BP neural network with quasi-Newton method, can significantly improve the precision of PV power prediction, and the method based on wavelet neural network with quasi-Newton method did better than that based on BP neural network. Especially, it can significantly improve the precision of PV power prediction under circumstance with low irradiance, such as in daily morning and evening time, rainfalls and snow as well as fluctuating power inflection point.