基于变分模态分解和多模型融合的用户级综合能源系统超短期负荷预测

Ultra Short-term Load Forecasting of User Level Integrated Energy System Based on Variational Mode Decomposition and Multi-model Fusion

针对用户级综合能源系统(integrated energy system,IES)多元负荷波动性和随机性较强、精确预测难度较大的问题,提出了一种基于变分模态分解(variational mode decomposition,VMD)和多模型融合的超短期负荷预测方法。首先采用VMD将IES各类负荷序列分解成不同的本征模态函数(intrinsic mode function,IMF);然后将各IMF结合气象信息构造不同的特征集,分别输入支持向量回归机(support vector regression,SVR)、长短期记忆(long short-term memory,LSTM)网络和一维卷积神经网络(one-dimensional convolutional neural network,1DCNN)进行预测;最后,将3个模型的预测结果输入SVR进行融合得到最终的预测值,并采用和声搜索(harmony search,HS)算法优化SVR的参数。通过某用户级IES的实际数据对所提方法的有效性进行了验证,结果表明,所提出的多模型融合方法优于单模型预测方法,对电、冷、热负荷均具有最好的预测精度。

In view of the strong volatility and randomness of the multi-energy loads and the difficulty in accurate forecasting in the user level integrated energy system(IES),an ultra-short-term load forecasting method based on the variational mode decomposition(VMD)and the multi-model fusion is proposed.Firstly,the multi-energy load sequences of IES are decomposed into different intrinsic mode functions(IMF)by the VMD.Then the IMFs are combined respectively with the meteorological information to construct different feature sets,which are input into the support vector regression(SVR),the long short-term memory(LSTM)network and the one-dimensional convolutional neural network(1DCNN)for prediction.Finally,the prediction results of the three models are input into the SVR for a fusion to obtain the final prediction values.The harmony search(HS)is used to optimize the parameters of the SVR.The effectiveness of the proposed method is verified by the actual data of a user level IES.The results show that the proposed multi-model fusion method has better prediction accuracy for electricity,cooling and heating loads,much superior to the single model prediction methods.