The recent advent of the Industry 4.0 era has led to the need to transform the industrial Internet of Things towards green, low-carbon and sustainable development. This is due to the fact that traditional industries consume too much energy. It is urgent to make use of digital technology for energy saving and emission reduction. However, there are still some unresolved issues in the transformation process: 1) the inability to use equipment resources thoroughly and efficiently, 2) the waste caused by overly simple resource management. In this paper, based on the above issues, we develop the ambient backscatter system to optimize the overall resource scheduling scheme and combine intelligent algorithms to solve the problem of offloading tasks. The solution optimizes offloading decisions to minimize system energy consumption and latency. Meanwhile, the proposed optimization problem is designed as a Markov decision process by combining the proposed federated learning assigned with asynchronous advantage actor-critic algorithm to obtain the optimal policy. The final evaluation results significantly show that the system performance indicator based on our proposed solution is better than others.