A remaining useful life prediction method based on DCS-optimized BiTCN-BiGRU

The remaining useful life (RUL) prediction of mechanical equipment is crucial for prognostics and health management. Currently, the RUL prediction based on deep learning still faces challenges such as difficulty in feature extraction and parameters tuning. To address these issues, a novel RUL prediction model is proposed, which integrates bidirectional temporal convolutional networks (BiTCN) with bidirectional gated recurrent units (BiGRU) and introduces a differentiated creative search (DCS) algorithm. Specifically, BiTCN is employed to extract local features from time series data, while BiGRU is utilized to capture long-term dependencies. Additionally, the DCS algorithm is applied to identify the optimal hyperparameters for the BiTCN-BiGRU model. Experimental results demonstrate that the proposed DCS-BiTCN-BiGRU model can achieve superior performance in RUL prediction tasks on the C-MPASS aero-engine dataset. Compared to the BiTCN-BiGRU model without DCS optimization, the DCS-BiTCN-BiGRU model reduces RMSE by 8.73% and MAPE by 14.34%. Meanwhile, compared to the models like BiGRU-TSAM, GGCN, and PI-DCNN-LSTM, it achieves the lowest RMSE reduction of 22.29% and MAPE reduction of 71.64%, thus provides more accurate and reliable RUL predictions for aero-engines.