基于CNN-GRU的光伏电站电压轨迹预测

光伏电站出力随机性易引发并网点电压大幅度波动，通过趋势预测提前调控是提高电压稳定性的有效途径。为了提升电压趋势预测精度，提出一种基于卷积神经网络（convolutional neural networks，CNN）和门控循环单元（gated recurrent unit，GRU）的电压轨迹预测方法。首先，通过采集单元提取电压数据构建时间序列；然后，计算电压时间序列的自相关系数及其与外部变量间的最大信息系数（maximal information coefficient，MIC），分析电压时间序列与外部变量在时序上的关联性；再通过CNN网络提取输入数据的高层特征；最后输入至GRU网络完成电压轨迹预测。通过某地光伏电站实测数据进行验证，结果表明：本文模型与GRU、长短期记忆网络（long short-term memory，LSTM）、CNN-LSTM、支持向量回归（support vector regression，SVR）等模型相比预测准确度更高。

Voltage Trajectory Prediction of Photovoltaic Power Station Based on CNN-GRU

The output randomness of photovoltaic power stations can easily cause large voltage fluctuations at grid-connection points. Advance regulation through trend prediction is an effective way to improve voltage stability. To improve the accuracy of voltage trend prediction, this paper proposes a voltage trajectory prediction method based on convolutional neural network (CNN) and gated recurrent unit (GRU). Specifically, a time series is constructed by extracting voltage data from the acquisition unit. Then, the autocorrelation coefficient of the voltage time series and its maximal information coefficient (MIC) relative to external variables are calculated, and the correlations of the voltage time series with external variables in timing are analyzed. Finally, the high-level features of input data are extracted through the CNN network and input into the GRU network to complete voltage trajectory prediction. The measured data of a photovoltaic power station are utilized for verification. The results show that compared with GRU, long short-term memory (LSTM), CNN-LSTM, and support vector regression (SVR) models, the proposed model has higher prediction accuracy.