A particle swarm optimization to identifying the ARMAX model for short-term load forecasting

In this paper, a new particle swarm optimization (PSO) approach to identifying the autoregressive moving average with exogenous variable (ARMAX) model for one-day to one-week ahead hourly load forecasts was proposed. Owing to the inherent nonlinear characteristics of power system loads, the surface of the forecasting error function possesses many local minimum points. Solutions of the gradient search-based stochastic time series (STS) technique may, therefore, stall at the local minimum points, which lead to an inadequate model. By simulating a simplified social system, the PSO algorithm offers the capability of converging toward the global minimum point of a complex error surface. The proposed PSO has been tested on the different types of Taiwan Power (Taipower) load data and compared with the evolutionary programming (EP) algorithm and the traditional STS method. Testing results indicate that the proposed PSO has high-quality solution, superior convergence characteristics, and shorter computation time.