Research on Ship Equipment Health State Assessment Method Based on BP Neural Network-Random Forest (BP-RF) and Combined Weighting

In view of the diversity and varying complexity of ship equipment, and the difficulty of existing state assessment methods in effectively handling the differences in the influence of different characteristic parameters on the equipment’s health state, leading to poor evaluation results, this paper proposes a ship equipment health state assessment method based on BP-RF and combined weighting. This method utilizes the BP-RF model to mine the implicit relationship between ship equipment feature parameters and state patterns, converting monitoring data into state pattern probability information. A trapezoidal membership function is used to determine the membership degree of each state pattern probability to different health state levels. The combined weighting method, which reflects a symmetric concept, balances expert experience and data information by integrating the subjective and objective weights of each state pattern probability, thus determining the equipment’s health state level. Through a case study of a specific type of ship’s gas turbine, the BP-RF model achieves a diagnostic accuracy of 98.3%, with F1 scores improved by 5.1%, 5.0%, 8.6%, 9.9%, 6.7%, 3.4%, and 5.3% compared to the BP neural network, RF, Support Vector Machine (SVM), Convolutional Neural Network (CNN), BP-SVM, SVM-RF, and CNN-SVM models, respectively. Additionally, the evaluation results of this method exhibit clear boundaries for each state membership degree, effectively addressing the problem of unbalanced contributions from characteristic parameters and comprehensively reflecting the relative importance and correlation of each parameter. Overall, this method provides a more accurate and comprehensive assessment of ship equipment health compared to other methods, offering reliable support for ship equipment maintenance and assurance.