为减小市场需求不确定性的影响，汽车供应链中下游企业会将实际需求放大后提供给上游企业，逐层传递后会形成供应链的牛鞭效应，导致上游企业制造成本较高、生产稳定性较差。因此，为减小供应链中需求放大造成的不良影响，汽车供应链各企业开始关注客户需求预测研究。本文以Z汽车部件公司客户需求预测误差较大、缺乏科学预测工具为出发点，使用不同方法建立需求预测模型，旨在为Z公司确定合适的需求预测模型，解决其在客户需求预测方面的问题，为企业进行需求管理和相关决策提供理论基础与模型支持。

本文以寻求Z公司最优预测模型为目标，主要基于历史需求数据建立模型进行预测研究。首先对需求的影响因素进行分析，从6个方面选择15个影响因素衡量指标，通过企业调研和相关统计网站获取相关数据；再将数据拆分为训练集和测试集，基于训练集数据建立数据挖掘的三个预测模型--支持向量机模型、BP 神经网络模型和随机森林模型；预测结果表明基于15个因素输入的三个模型模型的预测误差仍然较大，分析模型受“噪声”数据影响导致预测精度下降，接着通过灰色关联分析法确定15个因素中的6个主要影响因素，以6个主要影响因素作为输入建立基于大样本灰色关联分析的随机森林预测模型、支持向量机预测模型、BP神经网络模型。同时，为检验数据挖掘预测模型的有效性，本文基于需求数据特征使用传统时间序列预测方法建立ARIMA模型和灰色GM(1,N)模型，用于模型预测结果的对比和分析。最后通过统一误差评价指标分析不同预测方法的8种模型的预测效果，结果显示基于大样本灰色关联分析的支持向量机预测模型效果最好，性能最优。

考虑到需求波动会对预测模型精度产生影响，本文还分析了不同需求波动下各预测模型的适用性情况。使用案例推理（CBR）法分析出突发事件对汽车销量或产量的影响范围为[-80%,20%],在此影响范围内，假设Z公司2018年遇到4种突发事件造成不同程度的需求波动，通过比较各预测模型在不同需求波动下的预测误差来确定需求波动下的最优预测模型。实验结果表明，基于大样本灰色关联分析的支持向量机预测模型在不同需求波动下均为最优预测模型，能够较好应对突发事件造成需求波动的影响。本文研究结果不仅解决Z公司需求预测误差较大、缺乏科学预测工具等实际问题，还为相关制造型企业进行需求预测研究提供一定的理论参考。

In order to reduce the impact of market demand uncertainty, midstream and downstream enterprises in the automotive supply chain will amplify actual demand and provide it to upstream enterprises. The layer-by-layer transmission will incur bullwhip effect in the supply chain, resulting in high manufacturing costs and poor production stability. Therefore, so as to reduce the negative influence caused by the expansion of demand in the supply chain, companies in the automotive supply chain have begun to pay attention to customer demand forecast. This article takes Z auto parts company's error of customer demand forecast and lack of scientific forecasting tools as a starting point, using different methods to establish demand forecast models determine the appropriate demand forecast model for Z company,and solves its problems in customer demand forecast.Finally, this article provides theoretical basis and model support for enterprises to carry out demand management and related decisions.

This article builds a model based on historical demand data for forecasting research to seek the optimal forecasting model of Z company.Firstly, it analyzes the influencing factors of demand,and selects 15 influencing factor measurement indicators from 6 aspects.At the same time, this article obtains relevant data through enterprise research and related statistical Websites;then splits the data into training and test sets, and builds three prediction models of data mining based on the training set—support vector machine model, BP neural network model and random forest mode. The prediction results show that the prediction errors of the three models based on 15 factor inputs are still large, and affected by“noise”data.And the analysis model’s prediction accuracy decreases.Then this article determines 6 main influencing factors of 15 factors through the gray correlation analysis method, taking the 6 main influencing factors as input to establish a random forest prediction model based on large sample gray correlation analysis, supporting vector machine prediction model and BP neural network model. At the same time, in order to test the effectiveness of the data mining prediction model, this article uses the traditional time series forecasting method to establish the ARIMA model and the gray GM (1, N) model based on the characteristics of demand data for model comparison and analysis. Finally, the error evaluation index is used to analyze the prediction effects of the eight models of different prediction methods. The results show that the support vector machine prediction model based on large sample gray correlation analysis has the best effect and the best performance.

Considering that the fluctuations of demand will affect the accuracy of the forecasting model, this paper also analyzes the applicability of each forecasting model under different fluctuations in demand. Use case reasoning (CBR) method to analyze the impact range of emergency events on car sales or output, and the impact range is [-80%, 20%]. Within this impact range, supposed that Z company encountered four kinds of emergency events in 2018 that caused different degree of demand fluctuations, the article compares the prediction errors of various forecasting models under different demand fluctuations to determine the optimal forecasting model under demand fluctuations. The experimental results show that the support vector machine prediction model based on large sample gray correlation analysis is the optimal prediction model under different demand fluctuations, which can better cope with the impact of demand fluctuations caused by unexpected events. The research results of this paper not only solve practical problems such as Z company’s large demand forecast error and lack of scientific forecasting tools, but also provide a certain theoretical reference for related manufacturing enterprises to conduct demand forecast research. 

[1] 蔡晨阳. 济宁地区民用汽车市场需求预测研究[D]. 北京:北京交通大学, 2017.

[2] 张曼琳. 共享单车短时需求影响因素分析[J]. 中国商论,2020(05):103-104+155.

[3] 冯晓明. 城市森林生态服务居民需求偏好及影响因素分析[J]. 统计与决策,2019,35(14):95-99.

[4] Zhu X, Li L L, Zhou K L, et al. A meta-analysis on the price elasticity and income elasticity of residential electricity demand[J]. Journal of Cleaner Production, 2018, 201(08):169-177.

[5] 刘博瑶. 基于结构方程模型的长期护理保险需求影响因素研究[D]. 厦门:厦门大学, 2018.

[6] 张金良，关轶群. 基于IO-SDA模型的电力行业碳排放影响因素分析[J]. 华北电力大学学报(社会科学版), 2019(06):39-46.

[7] 郭淑婷. 社区居家养老服务需求意愿及影响因素研究[D]. 北京:首都经济贸易大学, 2018.

[8] 虢芳. 江西省农村物流需求的影响因素与预测研究[D]. 南昌:华东交通大学, 2018.

[9] 史彦虎, 郝彩娟, 朱先奇. 基于灰关联模型的汽车市场需求影响因素分析[J]. 山西农业大学学报(社会科学版), 2012, 11(02):155-159.

[10] Wang Z H, Wang C, Hao Y. Influencing factors of private purchasing intentions of new energy vehicles in China[J]. Journal of Renewable & Sustainable Energy, 2013, 5(6):1-14.

[11] 程波. 我国人口结构转变对城镇住房需求的影响研究[D]. 南京:南京大学, 2018.

[12] 马丽平, 张建辉. 基于误差修正模型的汽车需求影响因素分析[J]. 科技管理研究, 2014, 34(07):106-109+122.

[13] 谭章禄, 陈晓. 基于ISM模型的我国能源需求影响因素研究[J]. 工业技术经济, 2015, 34(06):74-81.

[14] Farzad A, Giovanni C, Susan H, et al. What influences travelers to use Uber? Exploring the factors affecting the adoption of on-demand ride services in California[J]. Travel Behaviour & Society, 2018, 13(06):88-104.

[15] Constantin S, Mihaela G, Corina I. Factors Influencing Touristic Demand and its Modelling Possibilities[J]. Procedia - Social and Behavioral Sciences, 2012, 62(09):1184-1189.

[16] Menezes M B, Silveira G J, Guimaraes R, et al. Estimating demand variability and capacity costs due to social network influence: The hidden cost of connection[J]. International Journal of Production Economics, 2018, 197(01):317-329.

[17] Fan H, Macgill I, Sproul A B, et al. Statistical analysis of drivers of residential peak electricity demand[J]. Energy and Buildings, 2017, 141(141):205-217.

[18] Jayarathna L, Rajapaksa D, Managi S, et al. A GIS based spatial decision support system for analysing residential water demand: A case study in Australia[J]. Sustainable Cities and Society, 2017, 32(03):67-77.

[19] 翁智雄, 马忠玉, 葛察忠, 等. 不同经济发展路径下的能源需求与碳排放预测——基于河北省的分析[J]. 中国环境学, 2019, 39(08):3508-3517.

[20] Xie J, Li H, Ma Z, et al. Analysis of Key Factors in Heat Demand Prediction with Neural Networks[J]. Energy Procedia, 2017, 105(03):2965-2970.

[21] 谢和平, 吴立新, 郑德志. 2025年中国能源消费及煤炭需求预测[J]. 煤炭学报, 2019, 44(07):1949-1960.

[22] 代业明, 高红伟, 高岩, 等. 具有电力需求预测更新的智能电网实时定价机制[J]. 电力系统自动化, 2018, 42(12):58-63.

[23] 李林波, 何思远, 梁潇. 一致性特征下综合体停车需求预测方法[J]. 同济大学学报(自然科学版), 2018, 46(03):340-345.

[24] 朱桃杏, 任建新, 张雪燕. 基于效用函数的京津冀高铁旅游出行需求研究[J]. 铁道工程学报, 2018, 35(03):102-108.

[25] 韩慧健, 韩佳兵, 张锐. 基于模糊认知图的物流需求预测模型研究[J]. 系统工程理论与实践, 2019, 39(06):1487-1495.

[26] 胡迪, 高庆水, 张楚, 等. 基于系统动力学的风力机备件需求预测研究[J]. 太阳能学报, 2019, 40(03):666-672.

[27] 胡铮, 袁浩, 朱新宁, 等. 面向5G需求的人群流量预测模型研究[J]. 通信学报, 2019, 40(02):1-10.

[28] 唐伦, 赵培培, 赵国繁, 等. 基于深度信念网络资源需求预测的虚拟网络功能动态迁移算法[J].电子与信息学报, 2019, 41(06):1397-1404.

[29] 段宗涛, 张凯, 杨云, 等. 基于深度CNN-LSTM-ResNet组合模型的出租车需求预测[J]. 交通运输系统工程与信息, 2018, 18(04):215-223.

[30] 李浩, 王铁宁. 基于组合预测的装甲装备器材需求确定[J]. 系统工程与电子技术, 2018, 40(10):2276-2281.

[31] 李文强, 段振云, 赵文辉. 基于偏最小二乘模型的无人机航材需求预测方法[J]. 系统工程理论与实践, 2018, 38(05):1354-1360.

[32] 刘庆, 刘秀丽, 汪寿阳. 基于合理膳食结构的2020-2050年我国食物用粮需求测算[J]. 系统工程理论与实践, 2018, 38(03):615-622.

[33] Rice W L, Park S Y, Pan B, et al. Forecasting campground demand in US national parks[J]. Annals of Tourism Research, 2019, 75(01):424-438.

[34] Johannesen N J, Kolhe M, Goodwin M, et al. Relative evaluation of regression tools for urban area electrical energy demand forecasting[J]. Journal of Cleaner Production, 2019, 218(01):555-564.

[35] Angelopoulos D, Siskos Y, Psarras J, et al. Disaggregating time series on multiple criteria for robust forecasting: The case of long-term electricity demand in Greece[J]. European Journal of Operational Research, 2019, 275(01):252-265.

[36] Goodwin P, Gonul M S, Onkal D, et al. When providing optimistic and pessimistic scenarios can be detrimental to judgmental demand forecasts and production decisions[J]. European Journal of Operational Research, 2019, 273(03): 992-1004.

[37] Saloux E, Candanedo J A. Forecasting District Heating Demand using Machine Learning Algorithms[J]. Energy Procedia, 2018, 149(08):59-68.

[38] Tanizaki T, Hoshino T, Shimmura T, et al. Demand forecasting in restaurants using machine learning and statistical analysis[J]. Procedia CIRP, 2019, 79(02):679-683.

[39] Merkuryeva G, Valberga A, Smirnov A, et al. Demand forecasting in pharmaceutical supply chains: A case study[J]. Procedia Computer Science, 2019, 149(01):3-10.

[40] Lobo B J, Brown D E, Grazaitis P J, et al. Long-term forecasting of fuel demand at theater entry points[J]. International Journal of Forecasting, 2019, 35(2):502-520.

[41] Hribar R, Potocnik P, Silc J, et al. A comparison of models for forecasting the residential natural gas demand of an urban area[J]. Energy, 2019, 167(10):511-522.

[42] Polunina N A . Forecasting sales volume of the company products on the basis of regression model[J]. 2016, 10:147-149.

[43] Pandey P, Kumar S, Shrivastav S, et al. Forecasting using Fuzzy Time Series for Diffusion of Innovation: Case of Tata Nano Car in India[J]. National Academy Science Letters-india, 2013, 36(3):299-309.

[44] Hulsmann M, Borscheid D, Friedrich C M, et al. General Sales Forecast Models for Automobile Markets and their Analysis[J]. Transactions on Machine Learning and Data Mining, 2012, 5(2): 65-86.

[45] Dwivedi A, Niranjan M, Sahu K, et al. A Business Intelligence Technique for Forecasting the Automobile Sales using Adaptive Intelligent Systems (ANFIS and ANN)[J]. International Journal of Computer Applications, 2013, 74(9):7-13.

[46] 陈艳华. 基于机器学习的时间序列预测方法研究与应用[D]. 兰州:兰州大学, 2017.

[47] 况东钰. 基于时间序列分析的网约车需求短时预测研究[D]. 北京:北京交通大学, 2019.

[48] 郑琰, 黄兴, 肖玉杰. 基于时间序列的商品需求预测模型研究[J]. 重庆理工大学学报(自然科学), 2019, 33(09):217-222.

[49] 王赟. 基于灰色关联分析的多因子选股模型研究[D]. 北京:北京交通大学, 2017.

[50] 刘香云. 基于灰色关联度的道路交通事故组合预测方法研究[D]. 北京:北京交通大学, 2015.

[51] 谷苏单. 灰色预测模型GM(1,N)在邯郸市成安县地下水矿化度预测中的应用[D]. 邯郸:河北工程大学, 2018.

[52] 李运. 机器学习算法在数据挖掘中的应用[D]. 北京:北京邮电大学, 2015.

[53] 张静娜. 数据挖掘在石油基础数据分析中的应用研究[D]. 西安:西安石油大学, 2013.

[54] 周志华. 机器学习[M]. 北京:清华大学出版社, 2016:225-226.

[55] 于扬红雪. 出租车需求量预测模型的研究[D]. 大连:大连海事大学, 2017.

[56] Breima L. Random Forests[J]. Machine Learning, 2001, 45(1): 5-32.

[57] 王曼. 基于数据挖掘的手机销量分析及预测[D]. 北京:北京工业大学, 2019.

[58] 高豪杰. 基于BP神经网络的阀门制造企业库存需求预测应用研究[D]. 北京:北京交通大学, 2015.

[59] Vapnik V N . The Nature of Statistical Learning Theory[M]. Springer, 2000.

[60] 赵猛. 基于数据挖掘技术的大气环境预测研究[D]. 北京:北京交通大学, 2017.