Adaptive Q-Learning-Based Demand Forecasting for Intermittent Spare Parts

Intermittent spare parts are characterized by high value,demand disruption and criticality. Effective demand forecasting helps optimize inventory management, ensuring timely maintenance of equipment to maintain the reliability and safety required for normal operation. Therefore, a self-adaptive forecasting method for intermittent spare parts based on adaptive Q-learning is proposed. Firstly, this method preprocesses and clusters spare parts demand data using similarity metrics and K-means. Secondly, it dynamically adjusts the matching relationship between the clustered spare parts categories and forecasting models using the Q learning algorithm, adapting the forecasting model for each category of data to achieve better forecasting accuracy. Finally, the method is validated using a dataset of intermittent spare parts. The results show that adaptive Q-learning can significantly reduce spare parts forecasting errors, with E_nrmse, E_mae, and E_mase metrics all lower than six other comparison models including Croston, ARIMA, and Deep Renewal Exact, with mean values of 8.921 0, 1.485 6, and 0.395 0 respectively for each category. Therefore, the proposed adaptive Q-learning method provides accurate and reliable forecasting accuracy and applies to intermittent spare parts demand forecasting.