In our previous study (Tanimoto & Hagishima (2005), Energy and Buildings 37), a set of state transition probabilities for the Markov Chain dealing with on/off cooling schedule in dwellings was proposed. Obtained probability of turning on an air conditioning system was defined in a form of Sigmoid-function by indoor globe temperature. Obviously, a real stochastic event of shifting from the off to on state cannot be affected by only indoor environmental parameters but also by other complex factors such as presence probability of family members, time, either weekday or holiday etc. In this paper, we report an alternative model based on the Multilayered Neural Network to predict off/on cooling schedule. We gathered field measurement data on familial dwellings during summer 2008 by deploying handy type hygrothermal meters with self-recording functions to measure room air, globe and blow-off air temperature of an air conditioner. The assumed Multilayered Neural Network has 9 nodes in both input and hidden layers, and 1 single node in output layer implying either state shifting from off toon (1) or not (0). The information given to the input layer nodes consists of what time, whether weekday of holiday, presence probability of inhabitants and PPD (Predicted Percentage of Dissatisfied). PPD derived from the theory of PMV is applied as a representative parameter for the indoor environment instead of globe temperature, since it contains various influences. The field measurement data sets were divided into two parts: teaching data and data for validation. The model trained by the teaching data was confirmed to reproduce state transition characteristic of the validation period, which seems complex and is determined by various inhabitants' manners. The model performance to reproduce is observed much excellent than the previous model derived from the Markov Chain.
All Science Journal Classification (ASJC) codes
- Environmental Engineering