Energy Consumption and Carbon Emission Prediction of District Heating System in Residential Communities Based on SSA-LSTM Model

Against global dual-carbon targets, urban residential central heating dominates building energy use and carbon emissions. Conventional LSTM forecasting requires manual hyperparameter adjustment and easily falls into local optima; micro-community carbon prediction also lacks accurate energy models and policy-based multi-scenario analysis for targeted low-carbon renovation. This study adopts the 2018–2023 hourly heating data of a community in H Province. It builds a preprocessing workflow with boxplot-Isolation Forest anomaly detection and MissForest filling, then constructs an SSA-LSTM hybrid model optimized by Sparrow Search Algorithm to predict heat and power loads precisely. Combined with carbon accounting and three policy scenarios, it evaluates carbon peak timing and emission reduction potential of heating renovations. Results show that SSA-LSTM attains 2.48% MAPE for heat and 3.20% for power, surpassing LSTM and BP. Only moderate and ideal renovation scenarios realize carbon peaks in the 2023–2024 heating period, with cumulative cuts of 138.19 t and 254.2 t by 2031–2032; household heat meters deliver 28% of total reductions. The framework offers quantitative support for community heating operation, renovation evaluation and carbon quota management.