Improved CEEMDAN-based aero-engine gas-path parameter forecasting using SCINet

Accurate gas-path parameter forecasting is very important for normal operations of aero-engines. In this study, the sample convolution and interaction network (SCINet), which is a variant of the temporal convolutional network, is applied to the forecasting of gas-path parameters for the first time. A new model, namely CEEMDAN-SCINet, is developed by improving SCINet based on complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN). CEEMDAN-SCINet solves the problems of future “unknown” data involved in prediction and high computational complexity of traditional decomposition-based forecasting models. SCINet and CEEMDAN-SCINet are compared with three typical deep learning forecasting models on aero-engine gas-path parameter data provided by a laboratory in China. The experimental results indicate that SCINet significantly improves the accuracy and stability of single-step and multi-step forecasting for aero-engine gas-path parameters, and that the forecasting of CEEMDAN-SCINet outperforms that of SCINet. Therefore, SCINet and CEEMDAN-SCINet offer broad application prospects for aero-engine parameter forecasting.