Identifying behavior models for process plants

Using behavior models for anomaly detection in hybrid systems

“…However, this algorithm does not account for discrete signals. In [5] we presented the HyBUTLA algorithm, that is to the best of our knowledge the first hybrid automata learning algorithm that can model both discrete and continuous dynamics.…”

“…return anomaly: unknown event (4) else (5) while isN otEmpty(T ) do (6) if t ∈ δ(e = (scurr, a, snew)) then (7) scurr := snew (10) if y = 0 and y p = 0 then (11) return anomaly: zero value of the signal (12) else if dif f ≥ ε then (13) return anomaly: signal drop (14) else if dif f ≤ −ε then (15) return anomaly: signal jump (16) else (17) return OK (18) end if (19) else if t < min(δ(e)) then (20) return anomaly: event too early (21) else if t > max(δ(e)) then (22) return anomaly: event too late (23) end if (24) end while (25) end if is detected. This is illustrated in figure 7, where the anomaly is represented by a non-existing path through the automaton, i.e.…”

“…AUTOMATED LEARNING OF BEHAVIOR MODELS Generally speaking, every hybrid system can be seen as a set of interconnected components that work in parallel and often asynchronously, where each individual component comprises strictly sequential behavior. Such components are defined by a parallelism model of the system's structure ( [5]). There are no algorithms for identifying this model automatically.…”

“…Behavior models are identified using an offline HyBUTLA algorithm for learning hybrid automata (presented in [5]). Similar to ALERGIA and MDI algorithms (see section II-B), HyBUTLA first builds a prefix tree acceptor (PTA) from the given observations, and then merges the compatible states using specific compatibility criteria.…”

“…A promising alternative to manual modeling are the datadriven approaches that come from the field of learning theory ( [4]). In this paper behavior models are identified using the HyBUTLA algorithm (presented in [5]). It is, to the best of our knowledge, the first algorithm capable of automatically learning models of hybrid systems in the modeling formalism of hybrid automata ( [6]).…”