Objectives  In response to the new requirements for the accurate forecasting of ship planned maintenance costs, a forecasting method via case-based reasoning is proposed.

  Methods  First, the feature vectors composed of the main feature attributes of various types of ships and their maintenance costs are represented by cases. The K-nearest neighbor (KNN) algorithm based on weighted Euclidean distance is then used for case retrieval, and the attribute importance of rough set theory is introduced. Second, the similarity between the retrieved case and target case is used as the adjustment coefficient, and each case is revised in combination with the idea of combined forecasting. Finally, the latest case obtained from the forecasting is added to the case library for the continuous accumulation of data.

  Results  The comparative analysis results of this method, the linear regression forecasting method and the radial basis function (RBF) neural network method against real ship maintenance data show that the average forecasting relative errors are 8.7%, 10.4% and 10.2%, verifying this method's forecasting accuracy and validity.

  Conclusion  The results of this study can provide references for the formulation and optimization of ship maintenance cost plans.