The protein secondary structure (PSS) prediction system presented in this paper is a subsystem of potato bioinformation research platform. The proposed method is a novel and practical PSS prediction method, which is based on nucleic acid sequence (NAS), uses an combined neural network (CNN) and takes an improved genetic algorithm (GA) to optimize the connection weights of CNN. The experimental results indicate that, not only the proposed method is feasible, but compared with the traditional PSS prediction methods, its prediction accuracy is higher, its use is more convenient, its search speed is faster and it has confidentiality in a certain degree.