Genetic Algorithms For Merging The Testing Data And Simulated Data

In reliability evaluation, the testing information would be lost by using simple merge method. With development of computer simulation technology, the simulated data is more and more reliable, so it is important to merge the testing data and simulated data in application. Since genetic algorithm does not depend on gradient information, it provides a common framework for solving complex system problems. Consequently, it does not depend on the specific areas of questions and has a strong robustness for the problem types, so it is widely used in many sciences fields.Among the life of the distributions, Weibull distribution is applied in many fields, this paper has mainly research on genetic algorithms for merging simulated data and test data of Weibull distribution.First of all, the paper gave the basic concept of the reliability of Weibull distribution, and the basic genetic algorithm structure was introduced and the main computer program for genetic algorithm was given.Also, it discussed the information merging principle and merging functions, and constructed genetic algorithm fitness functions based on method of moment, athwart-moment-method, Kolmogorov test, linear merging method, and extreme value distribution for a Weibull distribution unit, respectively. With the improved genetic algorithm, the optimal values of the unknown parameters in integration function are obtained. Furthermore, the methods were compared by examples, the most reasonable and effective merging method is given.Finally, since the distribution types of Weibull distribution cascade system is complex and merging function is not easy to be solved, this paper provided three methods, including series systems approximate fitting, unit parameter's maximum likelihood estimation method, and unit parameter's P quantile filling method. Then we discussed three methods by examples.