基于改进相似日和深度置信网络的光伏短期功率预测
刘永涛¹，胡冠中^1，2，张晋华^1，3，王红艳²
（1. 华北水利水电大学电气工程学院，河南 郑州 450000；2. 许昌学院电气与机械工程学院，河南 许昌 461000；
3. 新能源电力系统国家重点实验室（华北电力大学），北京 100096）
摘要： 针对基于传统灰色关联度的相似日选择算法进行光伏短期功率预测精度不高的问题，提出一种逆向云-灰色关
联度相似日选取混合算法进行光伏发电短期功率精确预测。算法充分考虑环境因素对光伏发电量的不确定性影响，通过
合理选择最优相似日作为深度信念网络的训练样本，建立基于相似日的粒子群优化深度置信网络光伏短期功率预测模
型，用于提高光伏短期功率预测精度。采用该方法，并在不同天气状况（晴天、多云） 下进行光伏短期功率预测，预测
结果分别与传统BP神经网络、相似日BP神经网络、粒子群优化深度置信网络预测结果进行了对比。结果表明，所提方
法在2 种不同天气状况下均具备较高的预测精度，证明了模型的优越性、有效性与通用性。
关键词： 光伏功率预测；灰色关联度；逆向云算法；深度置信网络
中图分类号： TM615      文献标志码： A      文章编号： 2095-0802-(2022)11-0001-06
Photovoltaic Short-term Power Prediction Based on Improved Similar Day and
Deep Belief Network
LIU Yongtao¹, HU Guanzhong^1,2, ZHANG Jinhua^1,3, WANG Hongyan²
(1. School of Electrical Engineering, North China University of Water Resources and Electric Power, Zhengzhou 450000,
Henan, China; 2. School of Electrical and Mechanical Engineering, Xuchang University, Xuchang 461000, Henan, China;
3. State Key Laboratory of New Energy Power System (North China Electric Power University), Beijing 100096, China)
Abstract: Aiming at the problem that the accuracy of photovoltaic short-term power prediction based on the traditional grey
relational degree similar day selection algorithm was not high, a hybrid algorithm of reverse cloud-grey relational degree similar
day selection was proposed to select the optimal similar day, thereafter accurate short-term power prediction for photovoltaic
power generation. The algorithm takes full account of the impact of environmental factors on the uncertainty of photovoltaic power
generation. By reasonably selecting the best similar day as the training sample of the deep belief network, a photovoltaic shortterm
power prediction model based on the particle swarm optimization deep belief network was established to improve the
precision of photovoltaic short-term power prediction. With this method, photovoltaic short-term power prediction was carried out
under different weather conditions (sunny and cloudy). The prediction results were compared with those of traditional BP neural
network, similar day BP neural network, and particle swarm optimization deep belief network. The results show that the proposed
method has high prediction accuracy under two different weather conditions, which proves the superiority, effectiveness and
universality of the model.

Key words: photovoltaic power prediction; grey relational degree; reverse cloud algorithm; deep belief network