Particle Swarm Optimization and Kalman Filtering for Demand Prediction of Commercial Buildings

The integration of weather forecasts and demand prediction into the energy management system of buildings is usually achieved using a model-based predictive control. The performance of such control techniques strongly depends on the accuracy of the thermal model which describes the building behavior. However, increasing the model complexity results in a reduced computational efficiency of the optimization problem which is an intrinsic part of the model predictive control. In this paper, a linear control-oriented thermal model of a commercial building is considered as the base model. Using the Particle Swarm Optimization technique, the parameters of the model are identified and the performance of the improved model is compared with the actual measurements. Afterwards, the improved model is used by a Kalman filter to predict the temperature and the heating/cooling demand of the building. The investigations are based on a commercial building located in the campus of ETH Zurich in Switzerland. Long-term measurements of temperature and power flows are used for the parameter identification. Initial parameter values are provided by the building manufacturing datasheet. The results of the case-study show that a very accurate temperature prediction can be achieved even for a four-day horizon, with a maximum absolute error of one degree Celsius.