Abstract: In view of the present difficulty in engine state evaluation under the condition of prior knowledge absence, an evaluation method of engine state based on improved Kmeans clustering algorithm is proposed in this paper. This method uses Kmeans algorithm to avoid the influence of subjective factors during evaluation. Correlation indicators are put forward, the algorithm is modified, and the corresponding weights are given according to the properties of feature parameters. A selection method of initial clustering centers with minimum variance heuristic algorithm is used to avoid the interference of acnodes and noise points during initial clustering center selection in the condition of small sample size, and Bootstrap small sample statistical method is adopted to weaken the effects of the randomness of test samples on evaluation model. Finally, the feasibility and validity of the method are verified by real example evaluation, indicating that compared with the traditional method, this method is more objective and stable.

Key words: engine, state evaluation, K_means algorithm, small sample statistics, engine, state evaluation, Kmeans algorithm, small subsample statistics