考虑不同状态划分方法的马尔科夫链空调车备件需求预测  

Research on Forecast of Air-conditioned Car Spare Parts by Markov Chain Considering Different State Division Methods

作　　者：王炳忠 毛宇 胡新生 胡海洋 Wang Bingzhong;Mao Yu;Hu Xinsheng;Hu Haiyang(Naval Aviation University Qingdao Campus,Qingdao,266041)

机构地区：[1]海军航空大学青岛校区,青岛266041

出　　处：《制冷与空调（四川）》2023年第2期175-179,201,共6页Refrigeration and Air Conditioning

摘　　要：针对空调车备件需求随机性和波动性的特点,利用灰色马尔科夫组合模型对空调车备件消耗量进行了预测。以状态数量和区间长度为基准,提出四种马尔科夫链状态划分方式,得到预测曲线,并拟合出与实际曲线的相关系数进行比较。结果表明:第7,11,12,14月的灰色预测误差均在5%以下,可考虑在状态划分时加入不变状态;从状态数量来看,五状态对应的相关系数整体上高于四状态,原因是五状态在四状态基础上,除了能预测备件需求上升和下降的强弱状态,还可以对上述灰色预测误差较小的月份不作修正,从而提高预测精度;从区间长度来看,五状态下的等值划分和等量划分,相关系数分别为0.98和0.85,由于等量划分时的各区间长度较为失衡,最终使得概率矩阵在第13个月状态预测出现偏差。Aiming at the randomness and volatility of air-conditioned car spare parts,the gray Marcow combined model is used to predict the consumption of air-conditioning car spare parts.Based on the quantity and interval length,four Malcov chain state division methods are proposed to obtain a predictive curve and compare the correlation coefficients of the actual curve.The results show that:the gray prediction errors in the 7th,11th,12th and 14th are below 5%,and you can consider adding an uncomvituted state when the state division is divided.From the perspective of the number of states,the correlation coefficient corresponding to the five states is higher than the four states as a whole.The reason is that on the basis of the four states,in addition to predicting the strong or weak state of the demand for spare parts,it can also not modify the month with less gray prediction errors,thereby improving the prediction accuracy.From the perspective of interval length,the equivalent and equivalent division in the five states,the correlation coefficients are 0.98 and 0.85,respectively.Due to the unbalanced length of each interval of equivalent division,the probability matrix predicts the deviation of the state prediction in the 13th month.

关 键 词：空调车 备件 需求预测 灰色马尔科夫模型 状态划分 

分 类 号：N941.5[自然科学总论—系统科学]