Search for a command to run...
The shift to smart cities with visions of becoming carbon neutral requires smart and smart adaptive energy management processes that could increase the savings and minimize the emissions and at the same time maintain occupant comfort. The present paper introduces EcoRL-TransNet, a new framework based on reinforcement learning which combines spatio-temporal modeling, explainable decision-making and multi-objective optimization to control a smart grid. The architecture also comprises dual graph encoder to model building-wise spatial interactions and temporal usage patterns, a transformer-based context aggregator to model spans over a wide range of time, and Carbon Attribution Attention (CA fr It optimises energy efficiency, carbon footprint and comfort with a personalised multi-objective actor-critic reinforcement learning algorithm. EcoRL-TransNet was tested against the CityLearn virtuality or simulation environment which was representative of the real energy consumption in multi building city environments. Experimental findings show that the suggested model performs better than previous models regarding all of its key metrics. In particular, it acquired a 21 percent decrease in average energy usage, a 25.3 percent decrease in maximum demand, and also a 487 kg reduction of annual CO 2 emissions. It also had low variance and robust policies with a comfort score of 0.91 in various training runs. The system gives practical and understandable outputs that would be applicable in a real time smart grid. The findings confirm that EcoRL-TransNet meets the most important requirements of green, effective and transparent energy systems of contemporary cities. It has innovative architecture that implies that it is a likely candidate of the next generation carbon-aware smart energy.