Automated terminal unit performance analysis employing x-RBF neural network and associated energy optimisation – A case study based approach

An artificial neural network based framework, to analyse and address key energy related performance issues inside building environment is proposed. The study engages thousands of heating, ventilation, and air conditioning terminal unit data from real commercial buildings employing a new feature engineering method augmenting terminal unit time-series data to create novel features. These features are subsequently fed into the proposed neural network, named x-RBF which is a combination of x-means clustering followed by radial basic function neural networks for automatic fault detection and diagnosis. The new model has been successfully employed and investigated on different types of heating–cooling patterns concerning power demand and control strategies from actual building historical terminal unit data. The proposed x-RBF model has been trained-tested on approximately three years of building data and achieves 99.7% sensitivity, a first for real building applications. Comparison has been made with other neural networks to verify the performance of the proposed x-RBF and it is further validated through statistical measurements. The proposed model demonstrates its ability to truly predict and anticipate fault scenarios in terminal unit operation. Consequently, energy and cost calculations have been executed to realise the potential financial impact (as a consequence of performance improvements) that can be achieved by the proposed framework in the building environment.