An optimization algorithm for heat exchanger networks and a new kind of gene matrix are proposed in the paper. Objective function is established as fitness of genetic algorithm and the weighted values are used to represent the structure of heat exchanger networks. Due to that with the genetic algorithm the optimal solution can be searched randomly at the whole solution space, multi-layer-perceptron trained by genetic algorithm is used to derive the best structure and parameters in the evolutionary computation. According to the results, the genetic-perceptron model is proved to be helpful to solve multi-peak nonconvex problem and it is efficient to optimize heatexchanger networks by genetic-perceptron model.