Abstract:

The proactive utilization of self-built biomass reactors by rural residents， photovoltaic equipment， and solar thermal equipment in energy supply can facilitate the development of low-carbon micro-integrated energy systems with islanded autonomous operation capabilities for individual users， thereby reducing their reliance on centralized power supply. To address the ambiguity in carbon emission pathways resulting from the diversified green energy production equipment and complex energy consumption behaviors in rural areas， a multi-layer physics-informed convolutional neural Network （PI-CNN） embedded with primary and secondary knowledge was proposed. Through deep interaction between primary and secondary knowledge bases， the model uncovered the mapping relationships between user energy consumption behaviors and carbon emissions， enabling precise tracking of carbon emission paths. Based on the learning outcomes of the PI-CNN network， appropriate weights were assigned to carbon emission pathways， radial distribution network power flows， and time-of-use heat and electricity pricing. An energy management strategy evaluation model was then designed by considering operational integrated carbon emissions， global power flow uniformity of the grid， and total benefits of user energy consumption， to quantify the quality of energy solutions. To address the challenges in solving centralized heterogeneous multi-objective optimization problems， an integrated energy management strategy was developed to balance economic efficiency， low-carbon performance， and robustness by combining adaptive gradient algorithms for mathematical problem solving. Simulation analysis conducted in a village in southern China verified the feasibility of PI-CNN and demonstrated the advantages of the proposed energy management strategy.

Key words: rural house, low-carbon operation, convolutional neural network, integrated energy system, adaptive gradient algorithm, carbon emission pathway