The invention discloses a distributed type system network resource operation management reliability raising method, which comprises the following steps: mapping the received tasks according to the use conditions of the resources and the learning experience of a deep learning neural network; based on the cost computation of the deep learning neural network, adjusting the number of the task weights in the mapping; and executing above operations to all tasks and updating the output weight of the deep learning neural network. With the method, under the condition of a hard-to-predict complex scenario, it is possible to not add complexity to the computation, but to adapt for best solutions to the problematic space.