考虑电力中长期交易机制的次月典型日负荷曲线预测OA

With the rapid development of the national unified electricity market,the accurate prediction of typical curves has become a key issue for the effective connection and interaction between medium-and long-term markets and spot markets.This paper focuses on predicting the typical daily load curve for the next month and proposes a user-group typical daily load prediction framework that integrates time-of-use electricity prices and multi-scale time-series features.First,a Mamba-LWT(Mamba-Local Window Transformer)dual-channel time-series model is constructed to capture seasonal macro-trends and intra-day local fluctuations,respectively.Through adaptive aggregation in long-short time-series routing,the model accurately models the multi-scale characteristics of load curves.Meanwhile,an improved transfer learning strategy is innovatively introduced into the model.It pre-trains general time-series features using source-domain data and then supplements time-of-use electricity price features via hierarchical fine-tuning with target-domain data,effectively addressing the problem of insufficient target-domain samples under the current time-of-use electricity price policy.Finally,the effectiveness of the proposed model is verified using electricity consumption data from industrial and commercial users in a city in northern China.The proposed framework outperforms the selected mainstream comparison models in point prediction,with the errors for all prediction indicators reduced by more than 10.32％.In interval prediction,all indicators of the 90％confidence interval are improved by at least 2.21％relative to existing models.This study will provide a solid data foundation for subsequent electricity trading strategies and also serve as an important reference for the sustainable development of the electricity market.