The electricity market is an important support for the construction of new power systems

and the locational marginal price (LMP) is the benchmark for transaction pricing in the electricity market

representing the market value of busbars. To further improve the prediction accuracy of LMP

this paper proposes a LMP decoupling forecasting method based on Evolutionary Learning improved Stacking. The method considers the pricing mechanism of LMP

and analyzes the characteristics of the energy component

congestion component

and marginal loss component of LMP. This paper also considers the coupling between congestion component and marginal loss component of LMP in different regions

and conducts a quantitative analysis of the feature contribution degree. Considering the differences in observation data and model training principles among basic algorithms

this paper combines multiple types of AI algorithms into a library of base learners for Stacking

including tree-based algorithms

support vector machines

neural networks

nearest neighbor algorithms

and regression-based algorithms. An evolutionary learning mechanism is deployed to optimize the selection of base learner combinations to ensure the whole model efficiency and accuracy. A case study conducted by using real data from the New England region proves its accuracy and universality

indicating its effectiveness in serving as a guide for electricity market reform in China.