Prediction and fault detection of building energy consumption using multi-input, single-output dynamic model

For many years degree-days methods have been used to estimate building energy consumption. If the difference between the actual and estimated energy consumption is large enough there may be good reasons to check the condition of HVAC-system. However, because degree-day method is based only on the outdoor temperature, there may be large differences between the actual and the calculated results without any noticeable failure in the system. A new method is proposed, based upon on-line measurements of weather and other characteristic data. The method is capable of being used with building automation systems for failure detection. The method uses a multi-input, single-output (MISO) dynamic model to predict the power fluctuation of the building. The model parameters are identified recursively by measuring the actual power, outdoor temperature, solar radiation, wind velocity and indoor temperatures. Other measurements may also be used as input data. The stochastic variations in power caused by occupants, equipment, lights, etc., can be included in the model. Measurements are taken once an half hour and are used to update the model parameters by a recursive extended least square algorithm (RELS). The identified model can be used to detect failures of HVAC equipment and systems. Verification of the method has been accomplished using real weather data and the TARP-computer program for the simulation of a townhouse and measurements collected from a real test building.