A hybrid model for evaluating the accuracy of failure forecasts in ship power plants

The growing complexity of ship power plants (SPP) and the heterogeneity of diagnostic data have made it increasingly difficult to ensure reliable failure forecasting and maintain operational safety. The purpose of this study was to analyse the correspondence between predicted and actual failures in SPP diagnostics using a comprehensive approach that combines knowledge-based methods, probabilistic modelling, and simulation experiments, and to test the proposed methodology by applying the developed model. The research integrated Case-Based Reasoning (CBR), probabilistic modelling, and simulation-based degradation analysis within a unified framework. Using diagnostic data from 150 operational cases of typical SPPs (main engine, generator, pump, cooling, and power systems), the model’s forecasting accuracy was quantitatively compared with baseline approaches, including classical CBR, adaptive CBR, ARIMA, and Decision Tree models. The integrated approach achieved the best performance (Root Mean Square Error = 0.32, R² = 0.93), reducing average error by 30-50% relative to classical methods. Each component of the hybrid framework contributed distinctly: the CBR layer ensured contextual consistency, probabilistic modelling reduced uncertainty by ≈ 20%, and simulation-based analysis increased stability under incomplete or noisy data by ≈ 15%. At the subsystem level, the greatest improvements were observed for the main engine (root mean square error reduction by 46%) and the pump system (by 39%), confirming the model’s adaptability to variable operating conditions. Robustness tests with 10-20% missing or perturbed input data demonstrated sustained accuracy (R² > 0.90), validating the model’s resilience and practical applicability. The proposed methodology has practical value for improving the efficiency of maintenance and reliability of ship power systems, as it provides accurate, interpretable, and noise-resistant equipment failure prediction