关键词: 贝叶斯分类器, 公共自行车系统, 故障诊断, 贡献率

Abstract: In view of the facts that PBS was not equipped with on-board detection sensors, bicycle failure might not be found directly in time and make rider to suffer the safety risks. To solve such an issue, a bicycle fault diagnosis method was proposed based on naive Bayesian classifier. By analyzing the PBS historical data including rental records, maintenance records and user evaluation etc., 15 state classification features were extracted as feature vectors for bicycle fault detection. Based on the posterior probability of Naive Bayesian classifier, the contribution rate of each state classification feature to the class was obtained; the recall rate was used as an evaluation index to predict the diagnosis of failed bikes. Based on the running data of 2016 from Hangzhou PBS, the model was verified by experiments. Through the optimization of the input eigenvalue of the prediction model, the recall rate of the test samples reaches 85.79% with satisfactory accuracy.

Key words: Bayesian classifier, public bicycle system(PBS), fault diagnosis, contribution rate