5 - Forecasting of electricity prices, demand, and renewable resources

Abstract

In this chapter, basic concepts and techniques related to the prediction of environmental variables such as wind speed and power, solar irradiance, ambient temperature, and electricity price are discussed, including prediction techniques based on time series analysis, artificial intelligence, and more advanced approaches such as numerical weather prediction. In addition, forecasting uncertainty using probability distributions and the scenario generation and reduction approach are described. This chapter analyzes the relationship between operational research and forecasting by training a neural network (NN) for prediction purposes. To this end, the classical backpropagation (BP) algorithm is used as a starting point of a real-coded genetic algorithm (GA), implemented for NN training. Thus, the BP algorithm provides initialization, while the searching power of the GA is employed on finding a refined solution. The combined BP-GA approach is compared with an NN trained using the Levenberg–Marquardt algorithm—a widely accepted model.