Mid-long Term Load Forecasting Using Hidden Markov Model

This paper presents Hidden Markov Models (HMM) approach for mid-long term load forecasting. HMM has been extensively used for pattern recognition and classification problems because of its proven suitability for modeling dynamic systems. However, using HMM for predicting is not straightforward. Here we use only one HMM that is trained on the past dataset of the chosen load data. The trained HMM is used to search for the variable of interest behavioral data pattern from the past dataset. By interpolating the neighboring values of these datasets forecasts are prepared. The results obtained using HMM are encouraging and HMM offers a new paradigm for load forecasting, an area that has been of much research interest lately.

I. Introduction

Power load forecasting is the most important part of load forecasting in power system. With an accurate load forecast, several controls such as economic dispatch, energy transactions, security analysis, unit commitment and generator maintenance scheduling are all benefited. For a long time, most of the load-forecasting theory and methods are based on time series analysis, statistical models which include linear regression models, stochastic process, autoregressive and moving averages (ARMA) models and Box-Jenkins methods. With the quick development of artificial intelligence, some forecasting methods which own outstanding learn ability gain global application in short-term load forecasting, such as artificial neural networks (ANN)^[1], fuzzy logic approaches^[2] and hybrid architccturc^[3] based on self organizing (SOM), etc. Though there are a lot of forecasting models, no single one has performed well enough because each model can take just several or usually only one relevant factor into consideration. The combination forecasting method, which can fully utilizes the useful information from several models, becomes one of the most popular subjects in the field of forecasting methods ^[4]–[6]. Various combination forecasting methods have been proposed in the last few decades, which can be largely divided into two categories: linear combination forecasting and non-linear combination forecasting.