A short-term electricity forecasting model for new power system considering exogenous factors and multi-modal fusion
摘要
新型电力系统的能源供给侧与需求侧均发生深刻变革,电量数据日益复杂,对短期电量预测的精度提出了更高要求.提高短期电量预测精度可为新型电力系统的优化调度和电力市场营销计划提供可靠依据.为此,融合大数据处理算法、预测算法及机器学习的优势,提出考虑外生因素及多模态融合的新型电力系统短期电量预测模型.首先,将天气等外生因素纳入预测模型,针对外生因素导致原始数据复杂的问题,基于最小绝对值收敛和选择算子算法(least absolute shrinkage and selection operator,LASSO)对外生因素进行选择与分析,获得有效数据集;其次,采用互补集成经验模态分解(complementary ensemble empirical mode decomposition,CEEMD)将电量时间序列分解为高、中、低频3种模态分量,分别采用遗传算法-长短期记忆(genetic algorithm-long short-term memory,GA-LSTM)、径向基函数-自编码器(radial basis fucsion-autoencoder,RBF-AE)和粒子群优化-支持向量机(particle swarm optimization-support vector machine,PSO-SVM)进行预测,重构后得到多模态融合预测结果.结果表明,该模型的平均绝对百分比误差(mean absolute percentage error,MAPE)为1.49%,均方根误差(root mean square error,RMSE)为275.59,优于基于集成经验模态分解(ensemble empirical mode decomposition,EEMD)和变分模态分解(variational mode decomposition,VMD)的对比方法,以及反向传播神经网络(back propagation neural network,BPNN)、极限学习机(extreme learning machine,ELM)、时域卷积网络(temporal convolutional network,TCN)和门控循环单元(gated recurrent unit,GRU)等单一预测模型,有效提升了新型电力系统非平稳负荷的短期预测精度.
The energy supply side and demand side of the new power system are undergoing profound changes,and power data is becoming increasingly complex,which imposes higher requirements for the accuracy of short-term electricity forecasting.Improving the accuracy of short-term electricity forecasting can provide a reliable basis for the optimal scheduling of the new power system and the marketing plan of the power market.Therefore,this paper integrates the advantages of big data processing algorithms,forecasting algorithms,and machine learning,and proposes a short-term electricity forecasting model for the new power system considering exogenous factors and multi-modal fusion.Firstly,exogenous factors such as weather are incorporated into the forecasting model.To address the complexity of original data caused by exogenous factors,the least absolute shrinkage and selection operator(LASSO)is used to select and analyze exogenous factors,obtaining an effective dataset.Secondly,complementary ensemble empirical mode decomposition(CEEMD)is employed to decompose the electricity time series into three modal components:high-frequency,medium-frequency,and low-frequency.Genetic algorithm-long short-term memory(GA-LSTM),radial basis function autoencoder(RBF-AE),and particle swarm optimization-support vector machine(PSO-SVM)are respectively applied to forecast the three components,and the final multi-modal fusion forecasting result is obtained through reconstruction.The results show that the proposed model achieves a mean absolute percentage error(MAPE)of 1.49%and a root mean square error(RMSE)of 275.59,outperforming comparative methods based on ensemble empirical mode decomposition(EEMD)and variational mode decomposition(VMD),as well as single forecasting models including back propagation neural network(BPNN),extreme learning machine(ELM),temporal convolutional network(TCN),and gated recurrent unit(GRU),effectively improving the short-term forecasting accuracy for non-stationary loads in the new power system.