基于CNN-Bi LSTM的短期电力负荷预测

Short-term Power Load Forecasting Based on CNN-BiLSTM

短期电力负荷预测能准确评估地区整体电力负荷变化情况,为电力系统运行决策提供准确参考。电力负荷参数受多维因素影响,为充分挖掘电力负荷数据中的时序特征,提升电力负荷预测精度,该文提出一种基于特征筛选的卷积神经网络—双向长短期记忆网络组合模型的短期电力负荷预测方法。以真实电力负荷数据作为数据集,通过对多维输入参数的优化筛选,选取高相关性特征向量作为输入,构建预测模型。通过与添加注意力机制的组合模型对比验证了输入参数优化分析的可行性和优越性。最后利用实际算例将该方法与利用自动化模型构建工具构建的梯度增强基线模型及常用预测模型相比,该方法构建的组合模型可以提升多维电力负荷数据的短期预测精度。

The short-term power load forecasting accurately assesses the load changing situation of the whole area and provide precise consultation for the operation decision-making of the electric power system. Because the power load parameters are affected by multidimensional factors, in order to fully obtain the temporal characteristics of the power load and enhance the precision of the power load forecasting, this paper proposes a short-term power load forecasting method combining the convolutional neural network(CNN) with the bi-directional long short-term memory(Bi LSTM) neural network based on feature selection. Taking the real power load data as the data set, through the feature selection of the multidimensional input parameters, the predictive model is built based on the high correlation input characteristic parameters. Comparing with the model based on the attention system, the results proves the feasibility and superiority of our model with the feature selection. Finally, with the actual examples, the results show that this forecasting method may enhance the short-term forecasting accuracy of the power load data compared with the light gradient boosting machine(LightGBM) baseline models built by the automated model building tools(Automl_alex) and other commonly used forecasting models.