An adaptive predictive framework to online prediction of interior daylight illuminance

Aiming to solve an open problem of designing a appropriate daylighting controllers, there has been growing interest in the use of nonlinear technique to perform prediction of interior daylight illuminance. Interior illuminance modeling and prediction approach provides an objective way to predict the future value of interior daylight illuminance from time series model. The urge to consider adaptive predictive technique lies in the fact that daylight is highly dynamic and nonlinear in nature. This manuscript elucidates and evaluates the performance of three nonlinear models: Nonlinear Autoregressive (NLARX), Time Delay Neural Network (TDNN) and Adaptive Neuro Fuzzy Inference Scheme (ANFIS) for accurate real time series prediction of interior daylight illuminance from online exterior and interior sensor measurements. By adopting an online tuning of model parameters by an online RLS adaptation algorithm, error between the actual system dynamics and identified model is scaled down. The exterior and interior illuminance data set for modeling are experimentally acquired from respective illuminance sensors mounted outside and inside the test chamber at Manipal (13°13'N, 77°41'E). NLARX, TDNN and ANFIS model prediction results have been validated with the real time experimental measurements. In essence, performance index comparisons of three models indicate ANFIS as a lucrative tool for the online prediction of the dynamic interior illuminance. A practical aspect of proposed ANFIS computational prediction model elevates an opportunity to couple within computer/embedded system based algorithms to perform as a real time artificial light controllers.