Modified real-value negative selection algorithm and its application on fault diagnosis

Analyze the drawbacks of common real-value negative selection algorithm applied on fault diagnosis, and the modified real-value negative selection algorithm is presented based on the corresponding innovations. Firstly, the fault detector set is partitioned into remember-detector set covering known-fault space and random-detector set covering unknown-fault space. Secondly, taking all known states including normal state as self set in training period, get the random-detector set through negative selection and distribution optimization. Lastly, in order to avoid `Fail to Alarm´ event caused by the Hole, the two-time-matching method is presented in detecting period which takes the normal state as self set. A resistance circuit fault diagnosis experiment shows that compared with the common real-value negative selection algorithm, the modified real-value negative algorithm can effectively avoid `Fail to Alarm´ event, and has higher diagnostic accuracy.