Selecting Fuzzy Rules Using Clonal Algorithms And Its Application Study

For a classification problem, we must extract the features of the classified objects first, then, complete this task according to the features. Whatever measure and technique were used to extract features, the feature values gained all have no clear boundary, and they are fuzzy, that bring us a lot of difficulty for the problem resolving. However, fuzzy logic that is based on nature language and imitates the fuzzy thought of human is just a power tool to character and resolve thus problems, and its core is fuzzy if-then rules constructed by the study pattern features. Conventionally, fuzzy rules are extracted from experts and that is a very fussy work. For a classification task, the extracted rules not only are optimized in classification power, but also the relation among the rules, so, features extracted artificially is difficult. At present, measure about rule extracted automatically is few. This paper the elite rules were extracted from a mass of fuzzy rules produced by multiple fuzzy partition by clone selection algorithm proposed according to clone selection theory of artificial immune system, thus a pool of effective rules are fund, that have lesser rule number and very high classification correctness, and very short identification time. The elite rules were used in real-time vehicle classification for examining its performance, and has more exact classification capability by the computer simulation, it is satisfied. The clone selection algorithm includes two most important process, hyper-mutation and receptor editing. Hyper-mutation mechanism allow the immune system to explore local areas by making small steps towards an antibody with higher affinity, leading to a local optima, receptor editing offers the ability to escape from local optima on an affinity landscape and reach the global optimum. This two mechanic can keep the diversity of antibody population, and lead to a fast maturation of antibody-antigen affinity, as a result the algorithm convergence to a optimal solution more quickly and more stable, that is the cause that the algorithm was selected in this paper. In addition, another pivotal problem, feature extraction, was studied in vehicle classification when the measure proposed in this paper was applied to this issue. In this paper, we extracted four wavelet features about the vehicle inductive signal including wavelet energy, average value and diviation. These features overcome the influence of vehicle speed in vehicle identification dynamically, and ensure the accuracy of the vehicle classification.