Compressor Oil Temperature Prediction Based on Optimization Algorithms and Deep Learning

Abstract

 The prediction of lubrication oil temperature plays a crucial role in the performance optimization and fault diagnosis of twin-screw refrigeration compressors. However, due to the complexity of operating conditions and the nonlinear characteristics of the data, traditional prediction methods still face challenges in terms of accuracy and generalization ability. This paper proposes a lubrication oil temperature prediction method based on optimization algorithms and deep learning, integrating an improved Beluga Whale Optimization (BWO) algorithm with a Temporal Convolutional Network (TCN) and an Attention Mechanism. The improved BWO algorithm enhances global search capability and convergence speed by introducing a survival strategy and an optimal position update strategy while optimizing hyperparameter selection to improve the predictive performance of the deep learning model. In experiments, we conducted a comparative analysis of various optimization algorithms and the improved BWO. The experimental results demonstrate that the BWO-optimized TCN-Attention model (BWO-ATCNS) outperforms traditional methods in prediction accuracy, with a lower Mean Squared Error (MSE), reduced Root Mean Squared Error (RMSE), and a higher coefficient of determination (R²). This study provides an efficient and reliable solution for lubrication oil temperature prediction in twin-screw refrigeration compressors and offers new insights for industrial predictive modeling.