基于混沌-RBF神经网络的光伏发电功率超短期预测模型

Ultra-Short Term Prediction Model of Photovoltaic Output Power Based on Chaos-RBF Neural Network

光伏出力时间序列的随机性和波动性使得光伏出力预测难以达到理想的精度,而提高预测精度是降低光伏并网不利影响的前提。因此,在揭示光伏出力波动本质的基础上,提出了混沌-径向基函数（radial basis function,RBF）预测模型。首先,应用小波去噪处理后光伏发电功率实测数据,基于C-C法重构系统相空间,运用相图法和改进最大李雅普诺夫（Lyapunov）指数法,对输出功率进行非线性动力学分析,确定光伏出力具有混沌特性;然后,分析光伏出力相空间轨迹局部变化的规律性,采用RBF神经网络对系统轨迹进行拟合,建立基于混沌-RBF神经网络的光伏发电功率超短期预测模型;最后,选取去噪后的实测数据对所建模型进行训练和预测,验证模型在不同天气状况下的预测效果。结果表明,所建模型对晴朗、多云和阴天等天气状况都有较好的预测精度,显示出良好的预测性能。

1 Randomness and fluctuation of photovoltaic（PV） output time series makes it difficult to achieve desirable prediction accuracy, while improving prediction accuracy is the premise to reduce adverse impact of PV grid integration. Therefore, on the basis of revealing the nature of PV output fluctuation, a chaotic-radial basis function（RBF） prediction model is proposed. Firstly, wavelet method is used for signal denoising of actual measured data. Then C-C method is used to reconstruct phase space, and phase space reconstruction chart method and improved maximum Lyapunov exponent method are used to analyze nonlinear kinetic characteristics of output power to determine the chaos characteristics of PV output power. Secondly, regularity of local variation of the PV output phase space trajectory is analyzed, RBF neural network is used to fit the system trajectory, and an ultra-short term prediction model of PV output power based on chaos-RBF neural network is established. Finally, the model is trained and predicted with the measured data after denoising, and prediction effect is obtained under different weather conditions. Results show that the proposed prediction model has a better accuracy with good predictive performance for different weather conditions, such as sunny, cloudy and overcast.