Physical Modelling of a Light Rail HVAC System Using Long-Term Measurements

A method to develop a physical HVAC model of a light rail using only operation data from long-term measurements is presented. Physical HVAC modelling is often based on costly wind tunnel tests or time and computationally intensive CFD calculations. This method is a new approach using only data from everyday operation of the light rail, reducing modelling cost and improving the model accuracy since real life conditions are analysed. Data from two years of passenger operation of a suburban light rail in Karlsruhe, Germany, is used. A standard physical model for a HVAC system and a train compartment is developed. This model is parametrised using data driven modelling and model training. For data driven modelling, the conducted data is analysed and suitable models are derived. For example, the cooling system is modelled using a look-up table based approach developed with the data. For model training, the data is first separated into test and training data. The training data is then separated into different batches (heating up, winter-night, winter-day and summer), to parameterise different physical quantities of the model. Using a systematic grid search, parameters that fit the training data in an optimal manner are found. ... mehrFinally, the overall model is validated using the test data. Over a large temperature range from -5 °C to 35 °C the model shows good consistency with the test data. The mean absolute percentage error over all test data is 13 %. Within the batches, the summer data was modelled best with an error of about 8 %. The described method allows a fast and reliable method to develop an accurate physical HVAC model.