Skip to main content

Nature-Inspired Adaptive Architecture for Soft Sensor Modelling.

Kadlec, P. and Gabrys, B., 2007. Nature-Inspired Adaptive Architecture for Soft Sensor Modelling. In: NiSIS'2007 Symposium: 3rd European Symposium on Nature-inspired Smart Information Systems, 26- 27 November 2007, St Julian's, Malta.

Full text available as:

Kadlec_Gabrys_NISIS2007.pdf - Accepted Version


Official URL:


This paper gives a general overview of the challenges present in the research field of Soft Sensor building and proposes a novel architecture for building of Soft Sensors, which copes with the identified challenges. The architecture is inspired and making use of nature-related techniques for computational intelligence. Another aspect, which is addressed by the proposed architecture, are the identified characteristics of the process industry data. The data recorded in the process industry consist usually of certain amount of missing values or sample exceeding meaningful values of the measurements, called data outliers. Other process industry data properties causing problems for the modelling are the collinearity of the data, drifting data and the different sampling rates of the particular hardware sensors. It is these characteristics which are the source of the need for an adaptive behaviour of Soft Sensors. The architecture reflects this need and provides mechanisms for the adaptation and evolution of the Soft Sensor at different levels. The adaptation capabilities are provided by maintaining a variety of rather simple models. These particular models, called paths in terms of the architecture, can for example focus on different partition of the input data space, or provide different adaptation speeds to changes in the data. The actual modelling techniques involved into the architecture are data-driven computational learning approaches like artificial neural networks, principal component regression, etc.

Item Type:Conference or Workshop Item (Paper)
Uncontrolled Keywords:Soft Sensors, process industry, adaptive systems, evolving systems, ANN, genetic algorithms, natureinspired computational learning method
Group:Faculty of Science & Technology
ID Code:8517
Deposited On:21 Dec 2008 16:47
Last Modified:14 Mar 2022 13:19


Downloads per month over past year

More statistics for this item...
Repository Staff Only -