Evaluation Heuristics Sample Clauses

Evaluation Heuristics. The other aspect of the ASIL optimisation algorithm is the evaluation heuristics. These are used to evaluate each individual in a population to determine their “fitness” scores – high-scoring individuals are kept, poor-scoring ones are not. Typically the optimisation problem itself is defined in terms of objectives – e.g. minimise cost, maximise safety – and thus we need to be able to evaluate according to those objectives. In a single objective approach, multiple evaluation scores are combined into one value, which is then to be either maximised or minimised. In multiple objective approaches, the scores remain separate and can be maximised or minimised individually. In the case of the ASIL optimisation, there are two objectives to consider. The first is cost: how expensive a particular ASIL allocation might be to actually implement. This need not be particularly accurate or realistic – it only matters that it helps the algorithm to differentiate between different solutions. In our case, we chose a simple heuristic that scores each ASIL in an allocation like so:  QM = 0 points  ASIL A = 1 point  ASIL B = 10 points  ASIL C = 100 points  ASIL D = 1000 points Other schemes are also possible, including linear scales (e.g. A = 1, B = 2, C = 3, D = 4) and powers of two (2, 4, 8, 16 etc.). The ASIL values in an SRA are then summed together to find a total cost for that particular SRA. Thus an SRA of AABD will get a cost of 1012, whereas AACC would have a cost of 202. The other heuristic to consider is the number of invalid (failed) allocations. Each cut set in the FTA results is subject to one or more DSR – derived safety requirement – that indicates the minimum ASIL that has to apply to that cut set in order to meet the overall system safety requirements. Unlike the exhaustive ASIL decomposition algorithm, which guarantees that all SRAs will be valid, the optimisation makes no such guarantees (and in fact doing so could hamper the algorithm’s performance and success). Thus we award 1 penalty point for every cut set that does not meet its DSRs; the goal is obviously to obtain an SRA with 0 penalty points.