Statistical Mechanics Of Power Demand Prediction Due To Bayesian Inference Via Expected A Posterior Estimation

We predict a time-series of power consumption on the basis of Bayesian inference using the expected a posterior (EAP) estimation which corresponds to statistical mechanics of the Ising model. In this method, we first derive an appropriate set of the power consumptions which are similar to the target one due to the performance measure via the correlation coefficient with the target data. Then, with the use of these selected datasets on the power consumptions, we predict the increase/decrease in the power consumption as an expectation value of binary variable which is averaged over the posterior probability expressed as the Boltzmann distribution of the Ising model under the random fields. We clarify that the accuracy of the present method is improved with the increase in the number of the selected datasets on the power consumptions, if we set parameters tuning fluctuations around the MAP solution appropriately. Also, we clarify that the accuracy of the present method is further improved by utilizing a timetable expressed as the Boltzmann distribution of random fields. Keywords- Power Consumption Prediction, Expected A Posterior Estimation, Multiple Data, Correlation Coefficients.