Predicting supply curve of electricity in an intra-day market using state-space models and sequential markov chain monte carlo methods

Official URL: https://risc01.sabanciuniv.edu/record=b2579564_(Table of contents)

In a free market, the price of a commodity is based on the relation between demand and supply. There is no exception in the case of the electricity market. Power companies use various load forecasting techniques to predict how much supply will be needed for a particular amount of demand. In the market equilibrium where demand and supply curves intersect, the price of a given commodity is realized. However, due to the special characteristics of the electricity market, daily and hour-by-hour prediction of the electricity price is more important. In this study, we will predict the daily supply curve of the electricity market in Turkey. These predictions are for each hour of the day. We have developed a hidden Markov model (HMM) to predict the supply curve in an intra-day market. The most popular approaches in dealing with the hidden Markov models or state-space models are sequential Monte Carlo methods (SMC) which are called particle filtering methods. However, in the case of high-dimensionality, standard particle filtering algorithms fail and are not efficient. In our article, the latent variables of the model are approximated by a sequential Markov chain Monte Carlo (SMCMC) method, which is an innovation in load forecasting, especially when dealing with a high-dimensional problem. We propose two different kernels for our algorithm to sample from the target distribution. Moreover, we use an expectation-maximization (EM) algorithm to update the hyperparameters of the model, such as the variances of latent variables and observations in our hidden Markov model.