Quantifying Uncertainty in ASM Models with Markov Processes
Describing discrete real world processes in the presence of uncertainty often allows for more realistic behavior predictions when combining qualitative modeling with quantitative aspects derived from observed data. Increasingly, historic data from past observations of real world process behavior is becoming available in amble quantities. We propose here a systematic approach to formally describe knowledge and understanding of discrete process behavior as abstract state machine models combined with probabilistic descriptions of behavioral patterns expressed in terms of Markov processes. In realistic scenarios, states of the underlying stochastic process are often assumed to be not directly observable, rather one can observe output signals associated with state transition events in a probabilistic sense. This view leads to hidden Markov models, allowing to quantify the likelihood of output sequences and related state transitions. We illustrate the approach for processes naturally characterized by time series analysis and forecasting in the context of situation analysis and anomaly detection.
Will be continued