Abstract
Forecasting of the Air Quality Index (AQI) in high-altitude cities is significantly challenged by nonlinear temporal dynamics, particularly due to complex interactions between atmospheric physics and pollutant transport. A hybrid forecasting model integrating a simulated Complex-Valued Optical Convolution Accelerator (CVOCA) with a Bidirectional Long Short-Term Memory (BiLSTM) network is presented to improve prediction accuracy and computational efficiency. The simulated CVOCA module enables high-speed, complex-valued convolution to extract amplitude and phase features from high-dimensional AQI sequences, while the BiLSTM component captures bidirectional temporal dependencies. Performance was evaluated using daily AQI data from Lhasa (2014-2024, 4018 samples). The hybrid model achieved RMSE = 1.60, MAE = 0.93, MAPE = 1.60%, R = 0.997, and R(2) = 0.996, outperforming standalone LSTM and BiLSTM models. The proposed model captures complex AQI dynamics in plateau environments and provides a scalable solution for real-time forecasting and environmental decision support.