Application of Machine Learning Technique to The Weapon System Critical Spare Parts Forecasting System

• Main Authors: Lee, Chin-Yung, 李敬庸
• Other Authors: S. Deng
• Format: Others
• Language: zh-TW
• Published: 2010
• Online Access: http://ndltd.ncl.edu.tw/handle/02462531418494363524

碩士 === 國防大學理工學院 === 兵器系統工程碩士班 === 98 === This study is to focus on the inventory forecasting model of the electronic erratic and critical spare parts in the weapon system. We first simulate the training data by Poisson distribution, than apply the support vector machines （SVM） methods, back-propagation neural network （BPN）, trend-adjusted exponential smoothing method and exponential smoothing method to forecast the failure demand in the testing data periods, and obtained its failure Poisson probability respectively. At the end we use the support vector machines （SVM） method to classify the testing data by the training model. We hope get the relationship between the demand of spare parts, the average of the demand and the probability of Poisson distribution. We take the H type air defense missile system data as sample. On the condition of the shortage of sample information, we can get greater than 80% accuracy. So we confirm the module can keep the demand in the decimal point number off to avoid the defect of human decision, and get the accurate demand of spare parts without the reliability, operating and maintenance time, etc.