An efficient joint evolutionary algorithm-based neural network model for air quality prediction

Air Quality Index (AQI) prediction is vital for pollution control. Deep learning improves accuracy but suffers from three flaws: weak temporal modeling, no feature differentiation, and costly manual tuning. This paper proposes a joint optimization network model (JONM) to solve them. JONM uses a spliced multi-step predictor to avoid error accumulation. It processes the primary AQI feature and secondary features through separate network depths, preventing signal dilution. Eight complementary evolutionary algorithms jointly optimize hyperparameters, and particle swarm optimization adaptively fuses their outputs. On a ten‑year Beijing dataset, JONM achieves MSE of 203.5 and RMSE of 14.3. These values are 22% and 12% lower than the best baseline, respectively. Experiments on two extra cities and under pollution peaks and noise confirm its strong generalization and robustness.