A Power Market Nodal Price Forecasting Based on Historical Electricity Price and Unified Dispatch Load

With the progress of the reform in the electricity power system, all 8 electricity spot market pilot provinces have initiated trial runs of electricity spot market settlement and the staged progress has been achieved. As the major clearing pricing mechanism in current domestic power spot market, the forecasting of nodalprice plays an instructive role for mastering the law of nodal price fluctuations and generating bidding and offering strategies. This paper proceeds the day-ahead nodal price forecast for sampling 40 nodes of a certain province by applying and comparing 6 forecasting algorithms like LASSO regression, rigid regression, KNN, SVR, random forests, LSTM, with the historical power price and power load data generated during its real trial operation. The result of the nodal price forecasting shows that, in the case of taking historical nodal price as the only forecasting input, SVR performed the best forecasting accuracy and preferable generalization ability as well, while the LASSO regression performed the worst of which. While load data added in, the forecasting results run by SVR and LASSO regression reflected a significant improvement comparing -with other methods. In addition, SVR is also the most effective method of all. The improvement of the nodal price forecasting by adding load data partially depends on the forecasting model applied and the correlation of nodal price and the load data.