Estimation of operative temperature in buildings using artificial neural networks

In this article, the problem how to obtain models for estimation of the operative temperature in rooms and buildings is discussed. Identification experiments have been carried out in two different buildings and different linear and non-linear estimation models have been identified based on these experiments. For the buildings studied, it is shown that the operative temperature can be estimated fairly well by using variables, which are more easily measured, such as the indoor and outdoor temperatures, the electrical power use in the room, the wall temperatures, the ventilation flow rates and the time of day. It is also shown that non-linear artificial neural network models (ANN-models), in general, give better estimations than linear ARX-models. The most accurate estimation models were obtained using feed-forward ANN-models with one hidden layer of neurons and using Levenberg–Marquardts training algorithms.In one of the buildings, it is shown that for non-linear models but not for linear, the estimations are improved much when using the time of day as an input signal. This shows that the time of day affects the operative temperature in a non-linear manner.