3576 -- Modeling of Weather Data by Time Series Analysis for Air-Conditioning Load Calculations

A method for modelling weather data is proposed. Three weather elements - temperature, global solar radiation, and absolute humidity - are selected to be modelled. Each record is decomposed into deterministic or periodic and random components. The former is modelled by a Fourier series and the latter is modelled as a time series model, such as a multivariate auto-regressive moving average or vectored ARMA model. The statistical tests showed that the model can substitute the reference year data that are used for similar load calculations. The systematic treatment of climatic change or climatic differences between locations, which generates an approximate weather model where a precise weather record is not available, and data compaction is made possible by the modelling. In addition, the model, which is identified as a linear system, can provide a new load calculation method based on linear system theory with the stochastic weather data as input.