深度学习方法在能源负荷预测的应用  

Application of Deep Learning Methods in Energy Load Forecasting

作　　者：徐鹏[1,2] 张飞龙 杜景勃 XU Peng;ZHANG Feilong;DU Jingbo

机构地区：[1]北京建筑大学供热、供燃气、通风及空调工程北京市重点实验室,北京100044 [2]北京建筑大学燃气研究中心,北京100044

出　　处：《煤气与热力》2023年第4期1-14,共14页Gas & Heat

基　　金：内蒙古自治区科技重大专项(2021ZD0038)。

摘　　要：通过对6种常用的深度学习网络(循环神经网络、长短期记忆网络、门控循环单元网络、卷积神经网络、深度置信网络、生成式对抗网络)的解析,阐述深度学习的理论思想,综述深度学习方法在能源负荷预测领域的应用。深度学习在处理海量数据方面有更强的学习、提取特征的能力,有助于提高预测精度。研究文献及实例对比的结果都显示深度学习模型(LSTM模型)的性能要优于浅层机器学习模型(BP模型)以及传统统计模型(ARIMA模型)。以深度学习算法为基础的组合模型,是目前能源负荷预测领域的重点。通过将其他模型或者优化算法(PSO、EMD、GA算法等)与深度学习模型相结合,能够使得预测精度得到显著提高。实例对比中所使用的PSO-LSTM模型的预测精度高于LSTM模型,证明了组合模型能够有更高的精度。Through the analysis of six commonly used deep learning networks(Recurrent Neural Network,Long Short-Term Memory Network,Gate Recurrent Unit Network,Convolutional Neural Network,Deep Belief Network,and Generative Adversarial Network),the theoretical idea of deep learning is expounded,and the application of deep learning methods in the field of energy load forecasting is summarized.Deep learning has a stronger ability to learn and extract features in processing massive data,which helps to improve forecasting accuracy.The results of research literature and example comparison show that the performance of deep learning model(LSTM model)is better than that of shallow machine learning model(BP model)and traditional statistical model(ARIMA model).The combined model based on the deep learning algorithms is the focus of the current energy load forecasting field.By combining other models or optimization algorithms(PSO,EMD,GA algorithms,etc.)with deep learning models,the forecasting accuracy can be significantly improved.The forecasting accuracy of the PSO-LSTM model used in the example comparison is higher than that of the LSTM model,which proves that the combined model can have higher accuracy.

关 键 词：能源负荷预测 机器学习 深度学习 预测精度 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]