This article aims to address the challenges of data scarcity, individual differences and multidimensional nonlinear relationships in complex equipment cost prediction. A fusion prediction model based on the residual category merging operator (RCMO) and the non-equidistant GM(1,1) model is proposed to improve prediction accuracy and generalization ability.
By constructing “virtual time factors,” multiple dimensions of cost-influencing factors are integrated into a single sequence. Combined with the non-equidistant GM(1,1) model to handle non-uniformly spaced data, the residual classification mechanism is introduced and the RCMO is designed to generate virtual similar samples, optimizing the robustness of the model.
In the rocket cost case, the non-equidistant RCMO-GM(1,1) model has an average relative error of 0.02650, significantly better than the multiple linear regression model (0.39237) and I-GM(0, N) model (0.11969) and can effectively capture nonlinear trends and complex patterns, demonstrating stronger adaptability and reliability.
Although this model demonstrates a remarkable advantage in prediction accuracy, it has not yet been fully optimized in handling dynamic data updates. In the future, it is possible to explore the introduction of dynamic data update mechanisms to further enhance the robustness and applicability of the model.
This model solves the problems of data scarcity and individual differences through multi-factor fusion and residual correction mechanisms, providing a high-precision and stable tool for complex equipment cost prediction, and has broad engineering application potential.
