In this paper, a portfolio selection problem is discussed. Firstly, based on the credibility theory, the rate of return of a security is regarded as a triangular fuzzy number and the variance is replaced by the rate's lower half variance of-absolute semi-deviation to express the portfolio risk, Thus, a mean-absolute semi-deviation --sine entropy portfolio selection model is proposed, in which cardinal constraints and threshold constraints are considered in a fuzzy random environment. Secondly, an Improved Non-dominated Sorting Genetic Algorithm (INSGA-II) is designed to solve the proposed model, in which a chaotic optimization strategy is added into when the population is initialized. And a crowding distance formula is introduced, in which the variance is considered. This can make the solution more efficient. Finally, the historical data of some stocks in the Shanghai 50 Index constituent stocks are selected for numerical analysis to prove the effectiveness of the proposed model and the algorithm.