A Monte Carlo method for thermal building simulation

A simplified Monte Carlo method for finding an approximation of the building inside temperature distribution is given. Present simulation techniques are either over-simplified and use only a deterministic method, or are highly complex stochastic models. The new method consists of a Monte Carlo approach to find typical input distributions, used in conjunction with a more traditional deterministic building thermal simulation model. The output distribution is obtained by estimating the output distribution from a carefully selected sample of input distributions. Radiation and temperature input data are simulated separately, and then the combined effect is found with a numerical convolution integral. Because the convolution integral is only strictly valid for independent variables, a verification study is also presented, using four different buildings and five different ventilation rates. Complete experimental verification of the method requires measuring the inside temperature distribution for 5 years, with five different ventilation rates for the same four buildings. This was out of the timeframe of this study. Therefore, the method was verified by comparison of results obtained with the new technique and comprehensive results obtained by simulating every day for the same period with historical weather data. The results show that the average predicted temperature error is 0.68 degrees C, with a standard deviation of 1.37 degrees C. The verification thus shows that by using the new Monte Carlo method a good approximation can be found for the inside temperature distribution by using only 4% of the days from the 5-year period. (C) 2006 Elsevier B.V. All rights reserved.