Abstract: Actively leveraging the energy support role of self built biomass reactors, photovoltaic equipment, and photothermal equipment by farmers, guiding the construction of low-carbon micro comprehensive energy systems that are oriented towards individual users and have the ability to operate independently in isolated islands, is beneficial for reducing the reliance on centralized energy supply paths for users. A multi-layer convolutional neural network (PI-CNN) embedded with master-slave distributed physical knowledge is proposed to address the problem of carbon emission path ambiguity caused by the diversification of rural green energy production equipment and the complexity of energy consumption behavior. Through deep interaction between the master and slave knowledge bases, the mapping relationship between user energy consumption behavior and carbon emissions is explored, achieving accurate tracking of carbon emission paths. Drawing on the learning results of the PI-CNN network, optimize the weights of carbon emission paths, radial distribution network power flow, and user heat and electricity time of use price allocation, design an energy management strategy evaluation model that takes into account comprehensive carbon emissions, global power flow uniformity of the power grid, and total user energy consumption revenue, and quantify the quality of energy consumption schemes. Considering the difficulty of solving centralized non-uniform multi-objective optimization problems, a comprehensive energy management strategy that considers economy, low-carbon, and robustness is proposed by combining the adaptive gradient algorithm (AdaGrad) to solve mathematical problems. Taking a village in southern China as an example, conduct simulation analysis to verify the feasibility and superiority of the proposed PI-CNN energy management strategy.