As　the　industrial　Internet　of　Things　(IIoT)　is　being　built　and　promoted,　digital　and　intelligent　production　methods　are　advancing　rapidly.　However,　with　the　increasing　number　of　deployed　devices　and　complicated　resource　optimization　schemes,　several　inevitable　problems,　such　as　excessive　energy　overhead,　are　brought.　Driven　by　these　issues,　we　propose　and　design　a　green　system　architecture　and　optimization　scheme　that　consists　of　perceptual　control,　heterogeneous　system　model,　and　decision　optimization.　Based　on　this　aspect,　we　focus　on　optimizing　energy　efficiency　in　the　IIoT　by　utilizing　ambient　backscatter　communication　(AmBC)　technology　in　the　perception　control　layer,　and　reducing　training　costs　through　collective　deep　reinforcement　learning　(CDRL)　among　edge　nodes.　Simulation　results　show　that　our　proposed　scheme　has　significant　advantages　in　the　area　of　energy　efficiency.