基于多模型融合的CNN-LSTM-XGBoost短期电力负荷预测方法

短期电力负荷的精准预测可以有效指导机组组合调度、经济调度与电力市场运营。针对输入数据特征量受限时负荷预测的低精度问题，提出一种基于多模型融合的CNN-LSTM-XGBoost短期电力负荷预测方法。通过建立融合局部特征预提取模块的LSTM（long short term memory）网络结构，并将其与XGBoost（eXtreme boosting system）预测模型并行结合，之后结合MAPE-RW（mean absolute percentage error-reciprocal weight）算法进行模型融合初始权重设置，对最佳权重进行搜索，构建最佳融合模型。通过运用电力负荷数据对所提方法进行预测实验，结果表明CNN-LSTM- XGBoost模型的MAPE（mean absolute percentage error）与RMSE（root mean square error）分别为0.377%与148.419 MW，相比于单一网络模型与融合模型结构实现了误差指标的显著降低，验证了基于多模型融合的CNN-LSTM-XGBoost短期电力负荷预测方法具有较快的模型训练速度、较高的预测准确度与较低的预测误差。

Short-Term Load Forecasting Method Based on Multi-model Fusion Using CNN-LSTM-XGBoost Framework

Accurate short-term load forecasting can provide effective guidance for unit scheduling, economic dispatch and power market operations. Concerning the low accuracy problem of load forecasting brought by the limited features of input data, a method based on multi-model fusion using CNN-LSTM-XGBoost framework is proposed. The Long Short-Term Memory network structure fused with local feature pre-extraction module is first established and then integrated with the XGBoost prediction model in parallel. Afterwards by using mean absolute percentage error-reciprocal weight algorithm to set initial model fusion weights and start searching for optimal weight, the optimal fusion model is built. From the prediction experiment of load data by virtual of the proposed method, it is discovered that the mean average percentage error and the root mean squared error of CNN-LSTM-XGBoost are 0.337% and 148.419 MW respectively, which indicates significant decrease of the error metrics compared with the outcome using single network model and multi-model structure. Therefore, it is verified that the method based on multi-model fusion using CNN-LSTM-XGBoost framework has faster training speed, higher accuracy and lower error of prediction.