This paper focus on the control parameter of estimation of distribution algorithms based on sequential importance sampling particle filters and Cholesky decomposition. The calculation formula of initial value of was given. This initial value can makes the algorithm search all areas within the range in the early stage of evolution, and does not reduce the convergence. Then, several - generation curves were given, and the property and the impact on algorithm performance of these curves are discussed. Finally, the selection principles of - generation curve for different optimization problems were given, the curve that controls the population diversity change with the evolution is also applicable to other evolutionary optimization algorithms. The experimental results show that these principles are correct and efficient.