Abstract:TThe combination choice of scientific and technological innovation projects is an important decision-making problem for the survival and sustainable development of many new research and development institutions(new R&D institutions). From the perspective of Markowitz's modern investment theory, this paper constructs Markowitz's mean-variance theory model, and uses non-dominated sequencing genetic algorithm to simulate and test the applicability and appropriateness of decentralization strategy in the venture capital of new R&D institutions, and its influence on institutional risk prevention and investment performance. The results show that, compared with the centralized investment, the diversified investment portfolio strategy can effectively reduce the non-systematic risks brought by the single project of the new R&D institutions. Moreover, with the expansion and increase of the investment portfolio, the new R&D institutions can integrate the internal and external innovation resources more effectively, form the scope of technology ecosystem at the level of the invested enterprises, and give full play to the synergistic effect between technologies. Further numerical simulation analysis shows that there is a theoretical moderate portfolio size for the venture capital of new R&D institutions. Excessive and blind diversification of the portfolio will only reduce the marginal utility, resulting in the increase of the number of institutional investments but the decline of investment quality and the increase of risk exposure, which cannot make the venture capital reach the Pareto optimal allocation state. Based on this, it is proposed that new R&D institutions should adhere to the diversified investment strategy of investment quantity and investment quality, and constantly improve their own value-added service ability.