In a field battlefield environment, there are marine units composed of multiple local mobile clouds, and each cloud has several devices. If there are few resources distributed within a local cloud, it will result in an inability to handle the complex tasks. If each cloud handles its own tasks independently, it will cause a decrease in task completion rate. To solve this problem, tasks can be offloaded to other local mobile clouds to achieve collaborative services between multiple mobile clouds. In addition, the mobility of local clouds will cause distance changes, and once the distance exceeds the communication distance, disconnection will occur. So, the impact of mobility needs to be considered. In response to this situation, this paper first constructs the network environment, delay model and mobility model of multiple local mobile clouds. Then the problem is modeled as a task scheduling optimization problem that minimizes the joint objective, and a task scheduling algorithm (AC-GA) based on the combination of actor-critic and genetic algorithm is proposed. We use the advantages of the AC algorithm to solve high-dimensional and complex problems. The trained strategy is then used as the initial data of the genetic algorithm to update and iterate to find the global optimal solution for task scheduling. The obtained task scheduling strategy can effectively reduce the impact of device mobility and delay constraints on tasks and improve task completion rate.