Complex Equipment Health Prediction Based on Improved Multi -Candidate Markov Model

This paper addresses the limitations of traditional Markov models, which rely solely on the previous state for predictions, cannot utilize longer historical information, and lack a multi-candidate mechanism and dynamic inference capabilities. We propose an improved multi-candidate Markov model that introduces tolerance parameters and a custom historical window, allowing it to capture longer sequence patterns and produce multiple prediction results, enhancing adaptability to uncertain scenarios. The model assigns weights to possible states based on their historical frequencies, with higher weights indicating more frequent occurrences, thus granting them higher probabilities during predictions and enabling dynamic inference. Additionally, frequency statistics and Laplace smoothing strengthen the robustness of the state transition matrix, preventing zero probability issues and enhancing prediction stability. Finally, we validate the model using a LIMU system as a case study, combined with evidential reasoning method, demonstrating excellent feasibility and accuracy.