Fuzzy systems properly integrated with Qualitative Reasoning approaches yield a hybrid identification method, called FS-QM, that outperforms traditional data-driven approaches in terms of robustness, interpretability and efficiency in both rich and poor data contexts. This results from the embedment of the entire system dynamics predicted by the simulation of its qualitative model, represented by fuzzy-rules, into the fuzzy system. However, the intrinsic limitation of qualitative simulation to scale up to complex and large systems significantly reduces its efficient applicability to real-world problems. The novelty of this paper deals with a divide-and-conquer approach that aims at making qualitative simulation tractable and the derived behavioural description comprehensible and exhaustive, and consequently usable to perform system identification. The partition of the complete model into smaller ones prevents the generation of a complete temporal ordering of all unrelated events, that is one of the major causes of intractable branching in qualitative simulation. The set of generated behaviours is drastically but beneficially reduced as it still captures the entire range of possible dynamical distinctions. Thus, the properties of the correspondent fuzzy-rule base, that guarantee robustness and interpretability of the identified model, are preserved. The strategy we propose is discussed through a case study from the biological domain.
Log Analysis-Based Resource and Execution Time Improvement in HPC: A Case Study