METAHEURISTIC OPTIMIZATION OF SHORT-TERM HYDROTHERMAL SOLAR SCHEDULING

Abstract

Managing the electrical power system with real-time data presents difficulties, especially when there is renewable integration in the energy mix due to the intermittency of the source. With the growing share of solar photovoltaics in the energy, scenario difficulties emerge in balancing demand and generation to maintain stability in the grid as renewable sources have an intrinsic stochastic and unpredictable nature. Thus, high-quality forecasts are relied upon for generation management and scheduling. With the ever-growing increase of solar power and the world's transition into the usage of non-conventional sources, it is of paramount importance to have a highly accurate forecast of solar irradiance for the safe operation of the grid. Forecasting depends on the mathematical models derived through various approaches. Some of the recent research on machine learning can predict without a mathematical model defined explicitly. Forecasting also depends on factors like the forecast horizon and the location's geography, so empirical models replace theoretical models as no unique solution can be maintained. Depending on the forecasts, scheduling can be obtained based on constraints, such as minimizing costs for a particular power penetration level from renewables into conventional sources. Thus, proper forecasted global solar irradiance can be used to obtain solar photovoltaics output, which leads to better generation scheduling and thus cost reduction. It is also observed that much of the power system relies on non-conventional sources as fuel like coal and gas. However, incorporating hydropower, other renewables like solar, wind, and thermal power can substantially reduce the generation cost of running a thermal power plant and total pollutant emissions. Efficient fuel usage is also of primary importance due to its non-replenishable nature and increasing monetary value. Thus, it becomes a problem of hydro-thermal-solar coordination. The main idea is to minimize the generation fuel cost of a thermal unit (or several) by incorporating power from hydropower, solar power plants, and sometimes even wind for a given set of demands as hydro and renewables have practically zero fuel cost. It is essential to have an accurate forecast for renewables. It is equally important to obtain a proper generation schedule for hydro-thermal-solar because mathematically, it presents a non-linear, non-convex, highly constrained optimization problem that is quite difficult to solve and may not be solved deterministically. In this report, an in-depth discussion on solar insolation and its components, the mathematical basis for solar energy estimation considering atmospheric effects is presented. Mathematical approaches to data processing have been discussed as it essential for many algorithms. A detailed analysis for solar forecasting using state-of-the-art machine learning techniques like artificial neural networks is presented. Also, several weather scenarios are presented to test the model’s capability to capture the original points faithfully. Further, multiple models are analyzed and compared. A novel metaheuristic is proposed to solve a scheduling problem, and the model is implemented, tested, and shows the best results comparatively. The forecasted solar output values under different weather scenarios are incorporated in a hydrothermal scheduling problem, creating hydro-thermal-solar-scheduling (HTSS), and are solved using the proposed method. It is seen that a complex constrained problem of HTSS can be solved efficiently by stochastic algorithms. Finally, as a future scope, some improvements and other work have been advised.