Generation and verification of learned stochastic automata using k-NN and statistical model checking

• Published in: Applied intelligence (Dordrecht, Netherlands) Vol. 52; no. 8; pp. 8874 - 8894
• Main Authors: Baouya, Abdelhakim, Chehida, Salim, Ouchani, Samir, Bensalem, Saddek, Bozga, Marius
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.06.2022; Springer Nature B.V; Springer Verlag
• Subjects: Algorithms; Approximation; Artificial Intelligence; Computer Science; Forecasting; Infrastructure; Internet of Things; Machine Learning; Machines; Manufacturing; Markov analysis; Mechanical Engineering; Neural networks; Processes; Sensors; Statistical methods; Statistical models; Temporal logic; Verification; Water supply systems; Stochastic automata; Sensors; Learning automata; Temporal logic; Statistical model checking
• ISSN: 0924-669X; 1573-7497
• DOI: 10.1007/s10489-021-02884-4

Deriving an accurate behavior model from historical data of a black box for verification and feature forecasting is seen by industry as a challenging issue especially for a large featured dataset. This paper focuses on an alternative approach where stochastic automata can be learned from time-series observations captured from a set of deployed sensors. The main advantage offered by such techniques is that they enable analysis and forecasting from a formal model instead of traditional learning methods. We perform statistical model checking to analyze the learned automata by expressing temporal properties. For this purpose, we consider a critical water infrastructure that provides a scenario based on a set of input and output values of heterogeneous sensors to regulate the dam spill gates. The method derives a consistent approximate model with traces collected over thirty years. The experiments show that the model provides not only an approximation of the desired output of a feature value but, also, forecasts the ebb and flow of the sensed data.