Industrial Internet based on distributed computing and cloud computing platform forms a “cloud-edge-end” cooperative system. Facing the problem of computing task offloading for machine-type communication devices in industrial Internet scenarios, this paper transforms the task offloading problem into a Markov decision process problem, proposes an online task offloading algorithm based on deep Q neural network (DQN), and designs an optimal scheduling method based on iterative optimization for industrial Internet resources. Simulation experiments are conducted by comprehensively considering the network environment and server state during the task offloading process, and compared with other resource optimization scheduling strategies. The results show that the DQN algorithm converges in about 9000 steps and has good convergence performance. The offloading strategy based on the DQN algorithm can effectively reduce the delay, energy consumption and total overhead of the computational task offloading system in the economy.