基于多分支门控残差卷积神经网络的短期电力负荷预测

Short-Term Load Forecasting Based on Multi-branch Residual Gated Convolution Neural Network

短期电力负荷预测是电力部门进行电网规划和运行调度的重要工作之一,针对负荷数据的时序性特征,为提升电力负荷预测精度,建立了一种基于多分支门控残差卷积神经网络(residualgatedconvolutional neural network,RGCNN)的短期电力负荷预测模型。该模型首先采用多分支门控残差卷积神经网络对历史负荷的周周期特征、日周期特征、近邻特征进行深度特征提取;其次为增加模型的非线性拟合能力,采用注意力机制对权重进一步合理分配;最后通过归一化指数函数计算后输出负荷预测结果。使用2016年某电力竞赛数据进行实验,通过与4种常用模型对比,该模型预测结果的平均绝对百分误差(MAPE)评价指标下降了0.02%~0.70%,验证了该模型提高负荷预测精度的有效性。

Short-term load forecasting is one of the important tasks for power utilities to formulate grid planning and scheduling plans.Considering the temporal characteristics of the load data,in order to improve the prediction accuracy of power load,a shortterm load forecasting model is established based on multi-branch residual gated convolution neural network(RGCNN).Firstly,the multi-branch residual gated convolution neural network is used to extract the weekly cycle,daily cycle and the nearest neighbor cycle of the historical load data.Secondly,the attention mechanism is used to distribute the weight reasonably to increase the nonlinear fitting ability of the model.Finally,the load forecasting result is output after normalized exponential function calculation.Experiments are carried out with the data of a power competition in 2016.Compared with the four typical forecasting models,the proposed model provides the prediction result with the MAPE evaluation indicator decreased by 0.02%-0.70%,which verifies the effectiveness of the proposed model in improving the forecasting accuracy.