1、1,第三部分 计划供应链中的需求与供给,第七章 供应链的需求预测第八章 供应链的综合计划第九章 供应链的供给和需求计划：管理预计的可变性,第四章 供应链的需求预测Chapter 4 Demand forecasting in a supply chain,Supply Chain Management,7-2,3,Outline,7.1 预测在供应链中的作用7.2 预测的特征7.3 预测的组成部分及预测方法7.4 预测的时间序列法7.5 预测误差的度量,4,7.1 预测（Forecasting）在供应链中的作用,对未来需求的预测构成了供应链中所有战略性和规划性决策的基础。如推动流程是根据对顾客
2、需求预测来运行的。如Dell所有供应链规划活动都是以预测顾客最终购买行为及发生的时间为基础。如：生产：日程安排、库存管理、总体计划营销：销售资源配置、促销、新产品开发财务：生产线（设备）的投资和预算规划人事：雇员计划、雇佣、解雇拥有稳定需求的成熟产品最容易预测，如牛奶、纸巾等日常用品对于销售季节很短的时尚商品和高技术产品，需求预测较难,5,7.2 预测的特征,预测经常会出错，要包括预期结果和对误差的测量（因为预测经常会出错）；长期预测通常没有短期预测精确，即长期预测误差的标准差相对于均值要大一些； 产品差异化延迟（ postponement ）综合预测通常要比独立预测准确得多；在公司中越往供应
3、链的上游靠近（或者距离顾客越远），接收到的信息失真就越多。,6,7.3 预测的组成部分及预测方法,识别影响未来需求的因素，确定这些因素与未来需求之间的关系；平衡主观和客观两方面因素；此外，还需了解 过去的需求、产品提前期、广告计划或其他的营销努力、经济状况、计划的价格折扣、竞争对手已经采取的行动、天气、紧急事件等。,7,预测方法,定性法（Qualitative）：基本上是主观的，依赖于人们的判断和意见做出预测。在缺少历史数据或专家关于市场的见解对于预测十分重要时；时间序列法（Time Series）：利用历史数据预测未来需求；基于假设历史时期的需求是对未来需求的一种很好的暗示。适用于外界环境稳
4、定、基本需求模式年度变动不大；因果关系法（Causal）：假定预测的需求与有关外界因素（如经济环境、利率等）高度相关，利用对外界因素的预测来预测未来的需求；仿真法（Simulation）：通过模拟消费者选择进行需求预测，利用这种方法，公司可以将时间序列法和因果关系法结合起来，回答问题。如价格提升将会带来什么样的影响 并将各种预测结果结合起来作为最终的预测结果比单独运用某种方法更为有效。,8,预测的组成部分及预测方法,被考察需求(O)=系统成分Systematic component (S)+随机成分Random component (R)系统成分衡量需求的期望值由需求水平Level扣除季节因素
5、影响后的目前需求；需求趋势Trend下一时期需求的增长或衰减率；季节性需求Seasonality可预测需求的季节性变动。随机成分指预测中偏离系统需求的那部分，不能用目前的需求变动所解释预测的目的：过滤出随机成分（噪音），估计系统成分的需求。,9,需求预测的基本步骤,1、理解预测的目标2、把供应链的需求计划和预测整合起来3、了解和识别顾客群4、识别影响需求预测的主要因素5、确定合适的预测技术6、设定预测绩效和误差测度,10,7.4 预测的时间序列法,每一个预测的目的都是支持以预测为基础的决策，都是预测系统需求部分和估计随机需求部分。系统需求部分的数据在一般形式下包含需求水平、需求趋势和季节系数，
6、它也能表现为如下列方程所示的多种形式。乘法型：系统成分需求水平需求趋势季节系数加法型：系统成分需求水平需求趋势季节系数混合型：系统成分（需求水平需求趋势）季节系数,11,时间序列法分为：静态法和适应法。静态法（Static）：只对需求中系统成分的各个要素（需求水平、需求趋势、季节系数）预测一次，不根据观察到的新需求更新系统成分。适应法（Adaptive）：根据观察到的新需求更新系统成分的各个要素的预测。包括移动平均、指数平滑和进行需求趋势及季节性需求修正后的指数平滑。,12,静态方法,假设混合模型:Systematic component = (level + trend)(seasonal
7、factor)L基期的需求水平估计（对基期剔除季节性影响后的需求预测）T需求趋势的估计Stt期的季节性系数估计Dt t期实际观测到的需求值Ft预测的t期需求预测公式：Ft+l=L+(t+l)TSt+l步骤：剔除季节性需求的影响，用线性回归预测需求水平和需求趋势；估计季节系数。,13,Time Series Forecasting (Table 7.1)例：NaturalG,Forecast demand for thenext four quarters.,14,预测需求水平和需求趋势,剔除季节影响后的需求Deseasonalized demand = demand that would ha
8、ve been observed in the absence of seasonal fluctuations时期数Periodicity (p)：在周期内包含的所有时期之后，季节性周期将重复进行 for demand at NaturalG(Table 7.1, Figure 7.1) p = 4,15,Deseasonalizing Demand,16,17,剔除季节性影响后需求以一个固定比率变化，即剔除季节性影响后的需求与时间t之间存在一个线性关系Dt = L + tTwhere Dt = deseasonalized demand in period tL = level (dese
9、asonalized demand at period 0)T = trend (rate of growth of deseasonalized demand)In the example, L = 18,439 and T = 524,18,Time Series of Demand(Figure 7.3),19,估计季节性系数,Use the previous equation to calculate deseasonalized demand for each periodSt = Dt / Dt = seasonal factor for period tIn the exampl
10、e, D2 = 18439 + (524)(2) = 19487 D2 = 13000S2 = 13000/19487 = 0.67,20,Estimating Seasonal Factors (Fig. 7.4),21,预测季节性系数,The overall seasonal factor for a “season” is then obtained by averaging all of the factors for a “season”如果数据中存在一个r的季节性循环，对所有pt+i, 1ip为形式的时期，定义In the example, there are 3 seasonal
11、 cycles in the data and p=4, soS1 = (0.42+0.47+0.52)/3 = 0.47S2 = (0.67+0.83+0.55)/3 = 0.68S3 = (1.15+1.04+1.32)/3 = 1.17S4 = (1.66+1.68+1.66)/3 = 1.67,22,预测,Using the original equation, we can forecast the next four periods of demand:F13 = (L+13T)S1 = 18439+(13)(524)(0.47) = 11868F14 = (L+14T)S2 =
12、18439+(14)(524)(0.68) = 17527F15 = (L+15T)S3 = 18439+(15)(524)(1.17) = 30770F16 = (L+16T)S4 = 18439+(16)(524)(1.67) = 44794,23,适应性预测法,Ft+l = (Lt + lTt )St+l = forecast for period t+l in period t Lt = Estimate of level at the end of period t Tt = Estimate of trend at the end of period t St = Estimate
13、 of seasonal factor for period t Ft = Forecast of demand for period t (made period t-1 or earlier)Dt = Actual demand observed in period t Et = Forecast error in period t At = Absolute deviation for period t = |Et|MAD = Mean Absolute Deviation = average value of At,24,适应法预测步骤,初始化: Compute initial est
14、imates of level (L0), trend (T0), and seasonal factors (S1,Sp). This is done as in static forecasting.预测: Forecast demand for period t+1 using the general equation Ft+l = (Lt + lTt )St+l .估计误差: Compute error Et+1 = Ft+1- Dt+1 修正预测值: Modify the estimates of level (Lt+1), trend (Tt+1), and seasonal fa
15、ctor (St+p+1), given the error Et+1 in the forecastRepeat steps 2, 3, and 4 for each subsequent period,25,移动平均法（Moving Average）,当需求没有可观测的趋势或季节性变动 需求的系统成分需求水平将最近N期的需求平均值作为t期的需求水平预测：当观测到t+1期需求后，移动平均法给过去N期数据同样的权重，同时忽略所有比新的移动平均数陈旧的数据。,26,Moving Average Example,From NaturalG example (Table 7.1)At the end
16、 of period 4, what is the forecast demand for periods 5 through 8 using a 4-period moving average?L4 = (D4+D3+D2+D1)/4 = (34000+23000+13000+8000)/4 = 19500F5 = 19500 = F6 = F7 = F8Observe demand in period 5 to be D5 = 10000Forecast error in period 5, E5 = F5 - D5 = 19500 - 10000 = 9500Revise estimat
17、e of level in period 5:L5 = (D5+D4+D3+D2)/4 = (10000+34000+23000+13000)/4 = 20000F6 = L5 = 20000 = F7 = F8,27,简单指数平滑法（Simple Exponential Smoothing）,当需求没有可观测的趋势或季节性变动时 系统需求需求水平预测观测到t+1期需求Dt+1后，为需求水平的平滑系数，0 1。 值越大，预测值与最近的观测值越相关；反之亦然。,28,Simple Exponential Smoothing Example,From NaturalG example, forec
18、ast demand for period 1 using exponential smoothingL0 = average of all 12 periods of data= Sum(i=1 to 12)Di/12 = 22083F1 = L0 = 22083Observed demand for period 1 = D1 = 8000Forecast error for period 1, E1, is as follows:E1 = F1 - D1 = 22083 - 8000 = 14083Assuming = 0.1, revised estimate of level for
19、 period 1:L1 = D1 + (1-)L0 = (0.1)(8000) + (0.9)(22083) = 20675F2 = L1 = 20675Note that the estimate of level for period 1 is lower than in period 0,29,需求趋势修正后的指数平滑法Trend-Corrected Exponential Smoothing (Holts Model),系统需求有需求水平和需求趋势没有季节性变动 系统成分需求水平需求趋势Obtain initial estimate of level and trend by run
20、ning a linear regression of the following form:Dt = at + b T0 = a L0 = bIn period t, the forecast for future periods is expressed as follows:Ft+1 = Lt + Tt Ft+n = Lt + nTt 观测到t+1期需求后，修正Lt+1 = aDt+1 + (1-a)(Lt + Tt)Tt+1 = b(Lt+1 - Lt) + (1-b)Tt 为需求水平的平滑系数，0 1 为需求趋势的平滑系数，0 1。,30,Trend-Corrected Expone
21、ntial Smoothing Example,Example: Tahoe Salt demand data. Forecast demand for period 1 using Holts model (trend corrected exponential smoothing)Using linear regression,L0 = 12015 (linear intercept) T0 = 1549 (linear slope)Forecast for period 1: F1 = L0 + T0 = 12015 + 1549 = 13564Observed demand for p
22、eriod 1 = D1 = 8000E1 = F1 - D1 = 13564 - 8000 = 5564Assume = 0.1, = 0.2L1 = D1 + (1-)(L0+T0) = (0.1)(8000) + (0.9)(13564) = 13008T1 = (L1 - L0) + (1- )T0 = (0.2)(13008 - 12015) + (0.8)(1549) = 1438F2 = L1 + T1 = 13008 + 1438 = 14446F5 = L1 + 4T1 = 13008 + (4)(1438) = 18760,31,需求趋势和季节性需求修正后的指数平滑法Tre
23、nd- and Seasonality-Corrected Exponential Smoothing （Winter Model）,系统需求有需求水平、需求趋势和季节性变动 系统需求（需求水平需求趋势）季节性需求Assume periodicity pObtain initial estimates of level (L0), trend (T0), seasonal factors (S1,Sp) using procedure for static forecastingIn period t, the forecast for future periods is given by:F
24、t+1 = (Lt+Tt)(St+1) and Ft+n = (Lt + nTt)St+n,32,Trend- and Seasonality-Corrected Exponential Smoothing (continued),After observing demand for period t+1, revise estimates for level, trend, and seasonal factors as follows:Lt+1 = (Dt+1/St+1) + (1-)(Lt+Tt)Tt+1 = (Lt+1 - Lt) + (1-)TtSt+p+1 = (Dt+1/Lt+1) + (1-)St+1 a 为需求水平的平滑系数，01b 为需求趋势的平滑系数，0 1g为季节性需求的平滑系数，06，高估了需求；TS-6，低估了需求，应该更换一种新的预测方法,36,思考题,1、预测在像戴尔公司这样的按单生产的制造商的供应链中起到什么作用？2、你认为在巧克力和面粉的需求中系统成分和随机成分各是什么？,37,Thank you!,
