Interpretable Causal Structure Learning for Multivariate Air Quality Forecasting: An Integrated Evolutionary-Cognitive Approach

Accurate Air Quality Index (AQI) forecasting is essential for public health protection and sustainable urban planning. Existing data-driven forecasting models often provide limited interpretability of pollutant interactions, while conventional Fuzzy Cognitive Map (FCM) learning algorithms may struggle to optimize causal weights effectively. To address these limitations, this study proposes IGWO-FCM, an evolutionary-cognitive framework that integrates an Improved Grey Wolf Optimizer (IGWO) with FCMs for interpretable causal structure learning and AQI forecasting under uncertainty. The framework employs FCMs to represent pollutant interactions and feedback mechanisms, and introduces a hierarchical transformation operator and adaptive learning strategy to enhance causal weight optimization. On a real-world urban AQI dataset comprising 744 hourly records and six pollutants, as well as two public benchmarks, IGWO-FCM achieves the lowest MAE and RMSE among the compared FCM learning algorithms for five of the six pollutants, reducing these errors by up to 29.6% and 40.2%, respectively, relative to the strongest baseline. These results demonstrate the predictive performance of IGWO-FCM and its potential to provide interpretable insights into pollutant interactions for AQI forecasting.