Abstract:

To ensure the safe and stable operation of critical equipment in oil and gas stations, reduce safety accidents and improve maintenance efficiency while reducing operating costs, this paper proposed a novel residual life prediction model for gas engine generator sets in oil and gas stations based on deep learning models and genetic optimization algorithms. Firstly, the operational data of gas engine generator sets collected from the oil and gas stations were preprocessed, and the Principal Component Analysis (PCA) algorithm was applied to extract the main components of the multi-dimensional sensor features as input features. A deep learning residual life prediction model was constructed using 1-Dimensional Transposed Convolution (TransConv1D) and Long Short-Term Memory (LSTM) for training. To solve the problem of feature stacking redundancy during model training, the QGA was employed to optimize two hyperparameters of the model. Through case verification and comparative analysis with traditional methods, the proposed model achieved at least a 2.46% reduction in MAE and a 7.85% reduction in RMSE in the prediction results. The research demonstrates that the proposed model outperforms the traditional models with smaller prediction errors and can predict the residual life of gas engine generator sets more accurately.

Key words: gas engine generator set, residual life prediction, deep learning, quantum genetic algorithm (QGA), oil and gas station