Analysis of novel random neural network controller for residential building temperature control

Random neural networks (RNN) have strong generalisation capabilities and are easy to implement on hardware as compared to Artificial Neural Networks (ANN). In this paper, a novel RNN controller is proposed to maintain a comfortable indoor environment in a single zone residential building fitted with radiators for heating. This controller is capable of maintaining a comfortable indoor environment on the basis of a predicted mean vote (PMV)-based set point. The implemented RNN controller is compared with ANN controller for energy consumption, indoor room temperature, and minimum square error. Results show that for same training data and learning algorithm parameters, RNN converges faster and it consumes less energy, results in better comfortable room temperature as compared to ANN controller.