Since nonsmooth optimization problems are difficult to solve by deterministic algorithms based on subgradient information, the heuristic algorithm was considered.The optimization mechanism and characteristics of the basic great deluge algorithm(GDA) were analyzed and the solving steps were given as well.Then improved GDAs for unconstrained and box constrained problems were proposed respectively, where the parameter up was omitted.For the unconstrained case, the random walk algorithm with respect to neighborhood search was proposed.For the box constrained case, the method of choosing a feasible initial point and a chaos optimization algorithm with respect to neighborhood search were proposed.The improved GDAs were tested by taking several typical nonsmooth optimization problems as examples and were compared with other algorithms.The test results show that the improved GDAs are efficient and superior to other algorithms mentioned in the paper.So it can be used as a practical method for solving nonsmooth optimization problems.