During the operation of a multi-module Input Parallel and Output Parallel (IPOP) system, the system parameters of each module are hardly completely consistent due to the influence of parasitic parameters, and the traditional flow-sharing control method cannot make the system run in the best efficiency. At the same time, due to the existence of soft switch, the input current of high-gain circuit has the problem of sudden change, which is difficult to control. To solve the above problems, this paper proposes an efficiency optimization scheme based on genetic algorithm. Firstly, BP neural network is used to fit the efficiency curve of the single module circuit. Then the optimal current distribution scheme under corresponding working conditions is obtained by off-line training of genetic algorithm. Then low-pass filtering is used to collect the average value of input current, which can effectively solve the current mutation problem. Finally, according to the existing distribution scheme, the IPOP system is distributed by using the way of online table lookup, so that the system has the highest efficiency in a wide load range. Finally, a 640W prototype was built to verify the scheme. The proposed efficiency optimization scheme based on genetic algorithm can improve the efficiency of the system in the full load range, and the maximum efficiency can be increased by 5.95%.