In the existing rule-based performance optimization approaches at software architecture (SA) level

it has not been fully concerned that the count and the order of each rule usage are uncertain in the optimization process.As a result

the search space for performance improvement is limited and the better solutions are hard to find.Aiming to this problem and taking the system response time minimum as the optimization objective

firstly

a model called RPOM is defined to abstract rule-based software performance optimization at SA level as the mathematical problem for solving the optimal rule sequence.Secondly

a framework named RSEF is designed to support the execution of a rule sequence.Furthermore

an evolutionary algorithm named EA4PO is proposed to find the optimal performance improvement solution by introducing statistical learning

constraint checking and repairing.Finally

a web application is taken as a case in the experiments for comparing with the existing methods.The experimental results indicate that the shorter system response time can be obtained and the statistical learning can obviously improve the convergence rate and the solution quality in our approach.