Enhanced Predictive Maintenance for Al Sabiya Steam Power Plant using BiLSTM Attention Networks and Optimization Algorithms: A Comparative Study

Abstract

Predictive maintenance plays a role in optimizing operations as showcased at the Al Sabiya Steam Power Plant Especially in boiler feed pumps . In this research, there are three optimization focused models—PSO BiLSTM Attention, HHO BiLSTM Attention, and CSO BiLSTM Attention—utilizing BiLSTM networks with attention mechanisms were evaluated. The accuracy of these models was examined, with the PSO BiLSTM Attention model standing out for achieving an accuracy score of 1.0 in both two class and four class classification tasks. Particularly noteworthy is that the PSO BiLSTM Attention model accomplished this feat with a training time of around 50.64 seconds for the two-class scenario and 40.48 seconds for the four-class scenario surpassing its counterparts. Following behind is the HHO BiLSTM Attention model with training times of 49.77 seconds for the two-class scenario and 51.59 seconds for the four-class scenario while the CSO BiLSTM Attention model required training time at 117.63 seconds for the two-class scenario and 100.09 seconds for the four-class scenario. This study emphasizes how PSO optimization can effectively enhance both reliability and operational efficiency in BiLSTM networks equipped with attention mechanisms, for maintenance purposes.