Prediction of Aviation Spare Parts Consumption Based on Kepler-NeuralProphet Modeling

Aviation spare parts are essential resources for the aviation industry, playing a crucial role in ensuring the safety and reliability of aircraft flights. Nevertheless, because of their high cost and specific requirements, it is necessary to develop precise procurement plans. This paper addresses the challenges associated with poor prediction accuracy and limited explanatory capability of current methods when dealing with small-sample aviation spare parts data. It introduces a small-sample aviation spare parts consumption prediction method based on the Kepler-N euralProphet model. The NeuralProphet model has a strong combinatorial interpretation capability, which allows it to maintain interpretability while achieving accurate predictions. It utilizes trend, seasonal, holiday, and autoregressive terms to extract the features of time series data related to aviation spare parts consumption. In order to enhance the feature extraction capability of NeuralProphet model, the hyperparameters of NeuralProphet model were optimized using improved Kepler algorithm, which is based on the optimization of the Logistic-Tent-Cosine chaotic operator. In the empirical tests, the Kepler-NeuralProphet model yielded MAE, MAPE, and RMSE results of 3.31, 3.39%, and 5.59, respectively, outperforming the NeuralProphet, Prophet, and NARX models.