基于改进贝叶斯深度学习的光伏功率概率预测

Probability Prediction of Photovoltaic Power Based on Improved Bayesian Deep Learning

光伏功率点预测方法无法对功率波动进行预测,当功率出现短时波动时预测结果将会产生较大偏差。本文基于贝叶斯深度学习网络,通过堆叠自动编码器对复杂气象因素进行特征提取和自动降维,利用一维卷积神经网络学习历史运行数据的趋势特征,以此对光伏功率两类输入特性参数进行预处理,改进贝叶斯深度网络结构。最终在某光伏电厂数据上,进行了不同置信度区间的概率预测,并与其他概率预测算法进行对比验证,从仿真结果可知本文所提方法可对功率波动做出较为准确的响应,预测效果更好。

The photovoltaic power point prediction method cannot predict power fluctuations. When the power fluctuates for a short time, the prediction results will have a large deviation. Based on the Bayesian deep learning network, this paper uses stacked autoencoders to extract features of complex meteorological factors and automatically reduce dimensionality, and use onedimensional convolutional neural network to learn trend features of historical operating data to determine the two types of input characteristics of photovoltaic power. The parameters are preprocessed to improve the Bayesian deep network structure. Finally, on the data of a photovoltaic power plant, the probability prediction with different confidence intervals was carried out, and compared with other probability prediction algorithms for verification. From the simulation results, it can be seen that the method proposed in this paper can make a more accurate response to power fluctuations, and the prediction effect is better. good.