Short-Term Load Forecasting Using Fuzzy Inductive Reasoning and Evolutionary Algorithms

In this paper, fuzzy inductive reasoning (FIR) is applied to the problem of short-term load forecasting (STLF) in power systems for a day in advance. The FIR model learns both past and future relations from the load and the temperature. The proposed optimization model uses an evolutionary algorithm based on a local random controlled search—simulated rebounding algorithm (SRA)—to choose the inputs to the FIR model. Using an optimization method to determine linear and nonlinear relationships between the variables, a parsimonious set of input variables can be identified improving the accuracy of the forecast. The input variables are updated when a new load pattern is happened or when relative errors are unacceptable. With this update is achieved, a better monitoring of the load trend due to the process is not strictly stationary. The FIR and SRA methodology is applied to the Ecuadorian power system as an application example. Results and comparisons with other STLF methodologies (autoregressive integrated moving average, artificial neural networks, and adaptive neuro-fuzzy inference system) are shown, and conclusions are derived.