华北电力大学 控制与计算机工程学院
国家自然科学基金项目（52206009）
North China Electric Power University
提升区域供热系统(District Heating Systems, DHS)的供热负荷预测精度是实现智慧供热、提升能源利用效率的重要手段。本文提出了一种基于贝叶斯后验估计，长短期记忆(Long Short Term Memory, LSTM)算法和极端梯度提升(eXtreme Gradient Boosting , XGBoost)算法的供热负荷预测组合模型。首先根据供热负荷变化的特性，确定输入的特征值，构建LSTM模型和XGBoost模型，再采用贝叶斯后验估计算法进行融合，构造出一个LSTM-XGBoost组合模型。采用石家庄某供热站2019-2020年供暖季运行数据进行仿真验证，仿真结果的评价指标表明，本文提出的基于贝叶斯后验估计的LSTM-XGBoost组合模型的平均绝对百分比误差 (Mean Absolute Percentage Error, MAPE)为0.51%，即负荷预测平均偏差为0.51%。将本文所提方法与循环神经网络(Recurrent Neural Network, RNN)模型和门循环单元(Gate Recurrent Unit, GRU)模型进行比较，验证了本文所提出的方法能有效提高供热负荷预测精度。
Improving the forecasting accuracy of heating load of district heating system (DHS) is an important method to realize intelligent heating and improve energy utilization efficiency. This paper proposes a combination heating load forecasting model based on Bayesian posterior estimation, Long Short Term Memory (LSTM) algorithm and eXtreme Gradient Boosting (XGBoost) algorithm. Initially, we confirm the input features based on the characteristics of heating load changes to construct the LSTM model and the XGBoost model respectively. Then, the Bayesian posterior estimation algorithm combined them to construct the LSTM-XGBoost combination model. The operating data of a heating station in Shijiazhuang from 2019 to 2020 was used for simulation validation, and the evaluation indicators of its results indicate that the Mean Absolute Percentage Error (MAPE) of the LSTM-XGBoost combination model based on Bayesian posterior estimation proposed is about 0.51%, which means the average deviation of heating load forecasting is about 0.51%. Comparing the proposed method with Recurrent Neural Network (RNN) model and Gate Recurrent Unit (GRU) model, and the result verified that the proposed method can effectively improve the accuracy of heating load forecasting.
