Soft Sensors: Software-Based Sensors
One of the major trends driving and driven by technological advancement is increasing use of the sensors and instruments in the world around us. Examples of sensors enabling the proliferated use of instruments around the globe are in great number presented in this chapter. A consequence of this effect is that increasing amount of digital data is available for further processing and exploitation. In fact, data are developing into a precious commodity, which is often traded at very high price. These data are the essence of soft sensors discussed in this work, and as such there is a strong pressure on the quality of the data. With some exceptions (e.g., ), algorithms for soft sensors development require a substantial amount of high-quality historical data in order to be able to develop useful soft sensors, and yet at the same time the quality of real-life industrial data is often very low. By analyzing the data obtained from their sources, one very often finds data impurities like outliers, missing value, and measurement noise . The causes of these issues are numerous: it can be the physical limits of the hardware sensors, for example, in case of noisy data, and hardware sensor failures or maintenance, and in cases of missing data. For these reasons, the historical data often have to be manually treated to remove the impurities mentioned previously.