Towards power plant output modelling and optimization using parallel Regression Random Forest

In this paper, we explore the possibilities of using the Random Forest algorithm in its regression version to predict the power output of a power plant based on hourly measured data. This is a task commonly leading to a optimization problem that is, in general, best solved using a bio-inspired technique. We extend the results already published on this topic and show that Regression Random Forest can be a better alternative to solve the problem. A comparison of the method with previously published results is included. In order to implement the algorithm in a way that is as efficient as possible, a massively parallel implementation using a Graphics Processing Unit was used and is also described.