Data Quality in Hybrid Neuro-Fuzzy based Soft-Sensor Models: An Experimental Study

Soft sensor models are used to infer the critical process variables that are otherwise difficult, if not impossible, to measure in broad range of engineering fields. Adaptive Neuro-Fuzzy Inference System (ANFIS) has been employed to develop successful ANFIS based sensor models. In addition to the structure of the model, the quality of the training as well as of the testing data also plays a crucial role in determining the performance of the soft sensor. This paper investigates the impact of data quality on the performance of an ANFIS based soft sensor model that is designed to estimate the average air temperature in distributed heating systems. The average air temperature is estimated based upon the available information, including solar radiation (Qsol), energy used by boiler (Qin) and external temperature (T0). For this problem, with the measurement errors caused by reading and equipment of all three variables, it is not unusual to have some uneven patterns in dataset which will decrease the model accuracy. The article investigates the impact of data quality on the performance of the soft sensor model. The results of two experiments are reported. The results show that the performance of ANFIS based sensor models is sensitive to the quality of data. The paper also discusses how to reduce the sensitivity by an improved mathematical algorithm.