Residual correction with genetic algorithm regressive ensemble (RC-GARE) model for effective air temperature predictions of urban cities
摘要
Climate change mitigation resulting from rising temperatures represents one of humanity’s most pressing challenges, driven by rapid industrial expansion, urban development, substantial greenhouse gas emissions, and deforestation associated with urbanization. These factors collectively pose an unprecedented threat to environmental and climatic stability. Accurate air temperature forecasting has become essential due to the highly variable nature of temperature patterns across different geographical regions. Reliable temperature prediction capabilities can enhance the credibility of future environmental planning initiatives and support the maintenance of urban environmental sustainability and climate health. This research introduces residual correction (RC) based deep learning models, incorporating LSTM, BiLSTM, GRU, CNN, and RNN architectures. The methodology integrates a Genetic Algorithm (GA) for optimal ensemble formation, specifically through aggregation ensemble (AE) with equal weighting and regression-based ensemble (RE) approaches, resulting in the RC-GAAE and RC-GARE models, respectively. The RC-based deep learning models demonstrate substantial improvements over traditional approaches across multiple performance metrics. The average enhancement ranges include RMSE improvements of 76.82%–327.67%, MAE improvements of 64.25%–331.23%, MAPE improvements of 49.71%–319.72%, MSE improvements of 200.98%–1725.69%, and