Short-term load Forecast Based on Neuro-evolution Algorithm

Abstract

Load forecasting is an important tool in power system planning, operation and control. Load forecasting ensures the equilibrium between consumption and production and, so, helps in maintaining system stability, and optimal operation of the electricity market. Neuroevolution leverage the strengths of two biologically inspired areas of machine learning: artificial neural net works and evolutionary algorithms. The basic idea of Neuro-evolution algorithm is to search the space of neural network policies directly using an evolutionary algorithm, and find the best structure possible for the task at hand. Neuro-evolution can, therefore, improve the effectiveness of Neural Network by optimizing its structure in terms of complexity and efficiency using the optimization capabilities of evolutionary algorithms. The current paper presents a short-term load forecasting methodology, based on neuro-evolution algorithm. A comparative study is conducted between NE and two of the most used machine learning algorithms, artificial neural network (ANN), and Support Vector Regression (SVR).