考虑信息时移的分布式光伏接入配电网功率预测方法

Power Prediction Method of Distributed Photovoltaic Access Distribution Network Considering Information Time Shift

为了降低分布式光伏接入配电网功率预测值误差,提出考虑信息时移的分布式光伏接入配电网功率预测方法。采用皮尔逊相关系数描述数值天气预报(NWP)关键气象因子、光伏输出功率间的关联关系,确定最佳时移量,修正地理位置引起的气象信息偏移;采用随机森林算法处理分布式光伏出力数据,筛选高贡献度特征参数;利用双向门控循环控制单元(GRU)神经网络学习特征参数后,通过一维卷积神经网络捕捉输入时间序列的各个时间步;引入注意力机制降低气象时移量对分布式光伏出力的影响,实现功率预测结果的输出。试验结果表明:该方法可计算分布式光伏出力与气象数据间的皮尔逊相关系数,确定其最佳时移量;不同气象条件下预测值与实际值间误差均较小。

In order to reduce the error of power prediction values for distributed photovoltaic access distribution networks,a distributed photovoltaic access distribution network power prediction method considering information time shift is proposed.Using Pearson correlation coefficient to describe the correlation between key meteorological factors and photovoltaic output power of NwP,determine the optimal time shift,and correct the meteorological information offset caused by geographical location;using random forest algorithm to process distributed photovoltaic output data and screen high contribution feature parameters;after learning feature parameters using a bidirectional GRU neural network,each time step of the input time series is captured using a one-dimensional convolutional neural network;introducing attention mechanism to reduce the impact of meteorological time shift on distributed photovoltaic output and achieve the output of power prediction results.The experimental results show that this method can calculate the Pearson correlation coefficient between distributed photovoltaic output and meteorological data,and determine its optimal time shift;the error between predicted values and actual values is relatively small under different meteorological conditions.