Generative adversarial networks (GANs), which have powerful fitting ability and thus generate diverse samples, are efficient deep neural networks for generation tasks. GANs are rarely used to solve combinatorial optimization problems though they have great potential in the field of operational research. This paper presents a hybrid evolutionary algorithm (DCG-EA/I) driven by deep convolutional generative adversarial networks (DCGAN). First, the evolutionary individuals are encoded to fit the training of GANs. Then, an escaping strategy driven by DCGAN is proposed to expand evolutionary individuals space and enhance evolutionary population diversity. Moreover, we use a iterative local search 2-opt method to improve the quality of the solutions. Finally, a negative-as-positive mechanism is constructed so as to stabilize the training process of DCGAN and reduce the harm caused by mode collapse. The algorithm is tested on TSP standard library instances and real-world instances. Experimental results show that the proposed algorithm can mitigate the problem of local convergence and achieve competitive performances over other GAN-based algorithms.