Cost Optimization Incorporating Photovoltaic Power Forecasts Using Neural Networks in an Energy Management System

This paper introduces a receding horizon method designed to integrate updated Photovoltaic (PV) power generation forecasts into an effective Energy Management System (EMS). At first, three RNN PV-based forecasting algorithms are compared to find the best architecture based on evaluation metrics, and the most accurate is utilized to estimate the PV power production. Secondly, the best-performing algorithm feeds into the EMS PV power production data. In addition, the feeding process compares two approaches. In the first one, PV forecasts are updated every hour, and the EMS is run to optimize dispatch decisions for the upcoming 23 hours. In the second method, the PV forecast is received at midnight, and the EMS optimizes decisions. The proposed joint approach is verified using an accurate EMS and by comparing the results with the real PV power data benchmark. Hence, the results indicate that the dispatching of units can be accurately predicted using the proposed forecasting techniques, with the hourly update approach yielding the most precise outcomes.