Abstract: Faced with the increasing requirements of data transmission and computing in 5G and industrial Internet, mobile edge computing (MEC) has gradually become a novel methodology, which can effectively deal with the shortage of computing capacity of industrial Internet devices and alleviate network congestion. However, when many devices send computing requests at the same time, the computing load of the edge computing server is always exceeded. In addition, industrial Internet devices are usually equipped with limited battery, unable to execute the task with excessive energy consumption, and the huge number of devices also determine the network connection, data computing and other system overhead. Therefore, a computation offloading decision method of machine-type communication devices based on Q-learning in the industrial Internet scenario in this paper was proposed. The network environment and server state in the process of computation offloading was considered, and the delay, energy consumption and economic overhead caused by the process of offloading were jointly optimized. The simulation results demonstrate that the proposed scheme can effectively reduce total overhead of the delay, energy consumption and economy in the computing task offloading system.