Prediction interval modeling tuned by an improved teaching learning algorithm applied to load forecasting in microgrids

In recent years there has been growing interest in prediction models for non-conventional energy sources and demand in electrical systems because of the increasing use of renewable energy sources. The prediction interval models proposed in this paper are validated using local load data from a real-life micro grid in Huatacondo, Chile. The micro grid operates with an energy management system (EMS), which dispatches distributed generators based on unit commitment, minimizing generation costs. The relevant inputs for the EMS are predictions of the consumption and the available amount of renewable resources. In this paper a linear and a Takagi-Sugeno fuzzy model are proposed and they are used to construct a prediction interval that includes a representation of the uncertainties. The model parameters are identified such that they minimize a multi-objective cost function that not only includes the error but also the width of the prediction interval and its coverage probability. The resulting parameter identification is a complex non-convex problem. An Improved Teaching Learning Based Optimization (ITLBO) algorithm is proposed in order to solve the problem. This method is compared with a Particle Swarm Optimization procedure for a benchmark problem, showing that both algorithms find similar results. ITLBO is used to identify the load prediction models. These models are used to predict load up to two days ahead. Both models succeed in accomplish the design objectives.