Abstract

Electricity price forecasting is fundamental for energy market orientation, providing information contributing to decision making in the short term. Electricity prices are affected by the seasonal, environmental, operational, and regional factors that introduce complex nonlinearities to the prediction process. This work presents a novel hybrid approach based on artificial intelligence and global sensitivity analysis (GSA) to determine the degree to which these external factors influence the components that make up the locational marginal price (LMP) of electricity. The proposed approach is applied for the day-ahead market conditions of the Mexican southeast, a region with a tropical climate having substandard transmission capacity. The study considered a database composed of 11 independent variables (integrated by seasonal, environmental, operational, and economic factors) to forecast the energy, loss, and congestion components presented in the LMP. The data were subjected to preprocessing consisting of filters for cleaning, smoothing, and normalization. Subsequently, a multi-output model based on artificial neural networks (ANN) was trained. In the end, a global sensitivity analysis (GSA) was performed, which quantified the impact of inputs on products as well as the interactions among the inputs. The optimal model has a 50–50–50 architecture, which resulted in a forecast accuracy of over 90% for energy and losses components. GSA results show a strong relationship between the input variables and electricity prices, highlighting that the components are not affected in the same way by the exogenous inputs. The day of the year was the variable with the most significant impact on the loss and congestion components. The aforementioned is associated with the electricity transmission capacity deficit. On the other hand, fuel prices significantly impact the energy component, which is related to the generation mix profile of the peninsular region. The new hybrid methodology would improve the activities of researchers, energy companies, modelers, and electricity market forecasters, such as planning, energy management, and decision-making for policymakers and electricity traders.