In view of the problem that the transmission power and efficiency decrease in the inductive power transfer（IPT） system when mutual inductance varies，a composite resonant network is proposed with a refined genetic algorithm for parameter optimization. Through analyzing the topology of a traditional resonant network，a composite resonant network is proposed to improve the transmission power and efficiency characteristics against wide mutual inductance variations. With the maximum transfer power and efficiency as the objective function，a nonlinear programming mathematical model is established. A mixed method of the nonlinear simplex method and the logistic model is considered to generate the fine initial population，and two-time normalization is proposed to deal with the objective function and constraints. The feasibility rules are used to choose better individuals，and a mixed mutation is designed to improve the global searching capability. Final results and experiment show that the refined method can find the global optimum parameters quickly，which well meets the design requirements with good transmission power and efficiency characteristic when mutual inductance changes in a large scale.