资源描述
木炉工业案例分析报告第五小组木炉工业案例分析报告 完成时间:2008年5月20日Data, Models and Decisions Case 3:The Wood Stove Industry Wood stoves have become increasingly popular for home heating since the advent of higher energy prices. In many urban areas where wood stoves have become popular, they have contributed to local air pollution, thus drawing considerable attention from governmental agencies charged with monitoring air quality. Proposals have been offered to legislative bodies that wood stoves, like automobiles, require pollution control devices or standards. Some members of the wood stove industry have become concerned that imposed standards might require difficult and expensive modifications that would make their products less competitive in the market for home-heating equipment. The claim is made that a substantial reduction in the air pollution produced by a wood stove can be made by a slight modification of most stoves and by informing wood stove users of better methods for operating their stoves. As a result of concerns over the impact of such regulation, one wood stove manufacturer has undertaken research to determine what product or operational modifications would minimize the air pollution produced by its brand of wood stoves. The research involved the measurement of particulate matter, as observed by a photoelectric cell, shown by the gases venting out of a wood stove chimney. This method, used in conjunction with a measure of air flown out of the stack, provided a relative measure of particulate matter in terms of the percentage of light blocked by (or passing through) the venting gases. Since the manufacturers product could be used with two sizes of flue pipe, the experiment involved varying the air intake setting (1/4, 1/2, 3/4, and full open) and the flue size, and measuring the temperature and the relative level of particulate matter exiting the stack. The experiment was conducted by first building a fire in the stove and then closing the stove and adjusting the air intake valve to the appropriate level. After half an hour (to allow the fire to stabilize at the set air flow level), measurements of temperature and particulate matter were recorded, see the accompanying table. Analyze these data by the development of an appropriate multivariable regression model relating relative particulate matter concentration to the air intake setting, flue size, and flue temperature. Interpret your results. Observation Number Relative Particulate Matter Concentration Air Intake Setting Flue Size Flue Temperature 1 44 1/4 S 106 2 28 1/2 S 248 3 26 3/4 S 385 4 31 Open S 534 5 42 1/4 L 124 6 26 1/2 L 211 7 29 3/4 L 374 8 34 Open L 487 9 42 1/4 S 131 10 28 1/2 S 230 11 27 3/4 S 353 12 36 Open S 515 13 40 1/4 L 144 14 26 1/2 L 255 15 27 3/4 L 286 16 33 Open L 517 17 42 1/4 S 117 18 27 1/2 S 235 19 25 3/4 S 378 20 34 Open S 510 21 39 1/4 L 139 22 30 1/2 L 248 23 29 3/4 L 302 24 34 Open L 521 案例3:木炉工业由于高能源价格的出现,家庭木炉取暖变得日益流行。在许多木炉取暖流行的城市,木炉取暖导致了当地的空气污染,引起了负责监控空气质量的政府部门的高度重视。他们向立法部门提出建议:木炉需要像汽车一样要求有空气污染控制装置或者标准。有些木炉工业的成员担心,强制的标准会更难达到,修正的成本也会很高,这会导致他们的产品在家庭取暖设备的市场上竞争力减弱。他们声称,可以通过对大多数木炉做轻微的调整或者告知木炉使用者更好的操作方法,来确实的减少木炉对空气的污染。由于担心这项规则的影响,一个木炉生产商如何最大限度地减少他们的品牌对空气所产生的污染,这取决于他们着手研究什么样的产品或是修正哪些操作细节。这项研究主要是通过木炉烟囱中排放的气体来测量微颗粒物,例如观察光电细胞。这种方法和测量烟囱的排气量方法相结合,便提供了一种通过光源在通风口被阻塞的比例来测量相关微颗粒物的方法。既然生产商的产品可用于两种规格大小的暖气管道,那么这个实验就包括调整进气管的设置(1/4,1/2,3/4或完全开放)、调整暖气管道的尺寸,并且测量温度和相关微小物质的含量。第一次点燃木炉,然后关闭炉灶,将进气管阀门调整到适当的水平,半个小时后(在设置的空气流动水平下允许炉火稳定),测试并记录相关的微量物质和温度,见所附表格。建立相关一个适当的多元回归模型来分析这些数据,模型中相关微量物质含量与空气进出口的设置,暖气管大小和烟囱温度有关。请解释你结果。观察序号相关微量物质含量进风口设置暖气管大小暖气管温度1441/4S1062281/2S2483263/4S385431OpenS5345421/4L1246261/2L2117293/4L374834OpenL4879421/4S13110281/2S23011273/4S3531236OpenS51513401/4L14414261/2L25515273/4L2861633OpenL51717421/4S11718271/2S23519253/4S3782034OpenS51021391/4L13922301/2L24823293/4L3022434OpenL521案 例 分 析本案例中欲研究木炉燃烧取暖时烟囱排出气体中的微量物质含量与木炉进气口设置、暖气管大小和暖气片温度三个之间的关系。通过得出的结论来判断是否可以在可控制的成本内,且不影响产品的市场竞争力条件下,通过对木炉做轻微调整及告知消费者使用更好的操作方法减少排放尾气中颗粒物质的含量,从而达到环境保护部门的空气排放要求。一、变量类型的确定这里我们知道排除尾气中微量物质含量为因变量,其他三个为自变量。设定微量物质含量为Y,木炉进风口设置为X1、暖气管大小为X2,暖气片温度为X3。对于进风口的设置X1有4个水平,分别是1/4, 1/2, 3/4, Open. 将它们分别换算成一定的比率数值,分别为:0.25,0.50,0.75,1.00。三个自变量中,暖气管有2种型号,分别为S和L,作为定性变量,我们取1个哑变量X2来分析,X2= 0时,表示S型号,X2=1时,表示L型号。二、对Y和X1,X2,X3直接多元线性回归,观察显著性和拟合度 回归方程为相关微量物质含量 = 38.9 - 48.3 进风口设置 + 0.77 暖气管大小 + 0.0763 暖气管温度自变量 系数 系数标准误 T P常量 38.891 2.955 13.16 0.000进风口设置 -48.29 21.25 -2.27 0.034暖气管大小 0.769 2.273 0.34 0.739暖气管温度 0.07630 0.04109 1.86 0.078S = 5.45411 R-Sq = 31.5% R-Sq(调整) = 21.2%方差分析来源 自由度 SS MS F P回归 3 273.01 91.00 3.06 0.052残差误差 20 594.95 29.75合计 23 867.96Durbin-Watson 统计量 = 2.10494 相关微量物质含量 残差图 分析以上模型有关指标:R-Sq = 31.5% R-Sq(调整) = 21.2%,即模型中方程仅能对尾气中的颗粒物质Y的变异解释21.2%。显然不能满意。标准化残差图显示,数据有随着Y值增大而增加的趋势。即使将所有变量都加入回归,仍然不能良好的解释Y值的变异趋势,我们考虑其他模型。三、由于对Y和X1,X2,X3直接多元线性回归拟合度不高,故考虑将进风口的比例与温度的交互作用纳入模型。回归方程为相关微量物质含量(Y) = 65.0 - 41.8 进风口设置(X1) + 0.173 暖气管大小(X2)- 0.150 暖气管温度(X3) + 0.171 X1*X3自变量 系数 系数标准误 T P常量 65.033 2.447 26.58 0.000进风口设置(X1) -41.838 7.562 -5.53 0.000暖气管大小(X2) 0.1734 0.8084 0.21 0.832暖气管温度(X3) -0.15048 0.02409 -6.25 0.000X1*X3 0.17123 0.01448 11.82 0.000S = 1.93561 R-Sq = 91.8% R-Sq(调整) = 90.1%PRESS = 107.608 R-Sq(预测) = 87.60%方差分析来源 自由度 SS MS F P回归 4 796.77 199.19 53.17 0.000残差误差 19 71.19 3.75合计 23 867.96来源 自由度 Seq SS进风口设置(X1) 1 170.41暖气管大小(X2) 1 0.04暖气管温度(X3) 1 102.56X1*X3 1 523.76异常观测值 进风口设 微量物质含 拟合值 标准化观测值 置(X1) 量(Y) 拟合值 标准误 残差 残差 6 0.50 26.000 30.602 0.705 -4.602 -2.55RR 表示此观测值含有大的标准化残差Durbin-Watson 统计量 = 2.49205相关微量物质含量(Y) 残差图 删后t残差1Hi(杠杆率)1Cook 距离110.5086145710.3018129540.0232732852-0.0156508690.1523539689.2944E-0630.4689738420.1873860030.0105775924-2.09852910.2317313320.22530097150.3416924220.2053080010.0063267546-3.0650611030.1324806130.19900777972.1158631570.1752357190.16081093380.2961037490.2061552480.00478351890.8728145870.1891648880.03599655510-0.6520882120.1290543780.012994672110.6216279290.1466031280.013719542121.2373341710.1978806610.073484812130.4345445250.2241684650.01139867914-1.016478870.2092751450.05459590715-0.3539473960.4482600260.02133889416-0.63152080.2168384150.022806247170.0135060340.2335542831.17346E-0518-1.0385615760.1293637050.03192104119-0.1755854940.1652985680.001286717200.1231344340.1964913590.00078209121-0.4444559240.2103493620.010988401221.0223208440.172501520.043470995231.150600970.3119481520.11803204424-0.1042909370.2267841090.00067306比较值3(4+1)/24=0.63从上述数据可以看出, Y = 65.0 - 41.8 X1 + 0.173X2- 0.150 X3 + 0.171 X1*X3,这个方程能较好的拟合Y值的变异,判定系数达到91.8%,调整后也为90.1%。另外,根据杠杆率的经验法则判断,Hi大于3(p+1)/n来识别出有影响的观测值,自变量的个数4个,观测值个数为24个,所以Hi=3(4+1)/24=0.63尽管数据显示有异常观测值,但杠杆率Hi均小于0.63,所以基本判断无有影响的检测值。异常观测值:进风口设 物质含 拟合值 标准化观测值 置(X1) 量(Y) 拟合值 标准误 残差 残差6 0.50 26.000 30.602 0.705 -4.602 -2.55R 分析该回归方程存在的问题: 对于变量X2暖气管类型,p=0.8320.05,无显著性。 对各变量做相关性分析,发现进风口设置(X1) 和暖气管温度(X3)相关系数高达0.982,存在严重多重共线性。相关: 相关微量物质含量(Y), 进风口设置(X1), 暖气管大小(X2), 暖气管温度(X3) 相关微量物质含量 进风口设置(X1) 暖气管大小(X2)进风口设置(X1) -0.443 0.030暖气管大小(X2) -0.007 0.000 0.974 1.000暖气管温度(X3) -0.370 0.982 -0.039 0.075 0.000 0.858单元格内容: Pearson 相关系数P 值四、增加哑变量进行逐步回归在实验过程中,炉子的暖气管类型是定性变量,有2个水平;每个炉子都有4个进风口设置(1/4, 2/4, 3/4, 4/4),因此在实验中单单炉子的选择和操作对尾气的影响可已有8个组合,即(S, 1/4; S, 2/2; S, 3/4; S, Open; L, 1/4; L, 1/2; L, 3/4; L, Open)。因此,我们将每个组合作为一个变量,由于有8个定性变量,我们设置7个哑变量来表示这些不同的组合。定义如下:Xa=1, S, 1/2; Xa=0, 其他;Xb=1, S, 3/4; Xb=0, 其他;Xc=1, S,Open; Xc=0, 其他;Xd=1, L, 1/4; Xd=0, 其他;Xe=1, S, 1/2; Xd=0, 其他;Xf=1, L, 3/4; Xa=0, 其他;Xg=1, L, Open; Xg=0, 其他。因此,我们得到的Xa Xg 的值分别为:进风口设置暖气管大小XaXbXcXdXeXfXg1/4S0.000.000.000.000.000.000.001/2S1.000.000.000.000.000.000.003/4S0.001.000.000.000.000.000.00OpenS0.000.001.000.000.000.000.001/4L0.000.000.001.000.000.000.001/2L0.000.000.000.001.000.000.003/4L0.000.000.000.000.001.000.00OpenL0.000.000.000.000.000.001.00可以将相关颗粒物质看做是这些不同组合的变量及暖气片温度(X3)的关系,可得下表:观察序号相关微量物质含量(Y)暖气管温度(X3)XaXbXcXdXeXfXg1441060000000228248100000032638501000004315340010000542124000100062621100001007293740000010834487000000194213100000001028230100000011273530100000123651500100001340144000100014262550000100152728600000101633517000000117421170000000182723510000001925378010000020345100010000213913900010002230248000010023293020000010243452100000011. 对以上数据作逐步回归,使用Minitab软件得出如下数据逐步回归: 相关微量物质含量(Y)与 暖气管温度(X3), Xa, Xb, Xc, Xd, Xe, Xf, Xg 入选用 Alpha: 0.05 删除用 Alpha: 0.05响应为 8 个自变量上的 相关微量物质含量(Y),N = 24步骤 1 2 3 4 5常量 37.17 38.99 41.51 42.16 44.01暖气管温度(X3) -0.0154 -0.0184 -0.0225 -0.0201 -0.0200T 值 -1.87 -2.37 -3.28 -3.74 -8.60P 值 0.075 0.028 0.004 0.001 0.000Xe -7.3 -8.8 -10.1 -11.9T 值 -2.15 -2.96 -4.30 -11.49P 值 0.043 0.008 0.000 0.000Xa -8.5 -9.7 -11.6T 值 -2.85 -4.16 -11.17P 值 0.010 0.001 0.000Xb -8.7 -10.5T 值 -3.77 -10.29P 值 0.001 0.000Xf -9.2T 值 -9.08P 值 0.000S 5.83 5.41 4.67 3.63 1.58R-Sq 13.71 29.27 49.72 71.21 94.84R-Sq(调整) 9.79 22.53 42.18 65.15 93.41Mallows Cp 287.0 233.7 162.9 88.5 6.4通过上述步骤得出逐步回归的方程为:, 各系数都有显著性意义。 总判定系数R-sq=94.84%,调整后为93.41%。观察序号删后t残差1Hi(杠杆率)1Cook 距离111.542171630.1836818710.08284723220.409959040.3335663970.01469962430.196953720.333702210.0034205944-1.69632550.1829033750.09721331150.32442530.1675408080.0037151726-1.50812270.3349245210.17827439471.388754210.3395419010.1571450718-0.16240330.1439032620.00078115490.418868890.16164570.00590889100.135643360.3334616280.001622633110.470496320.3341212880.019349739121.689938990.1659761760.08586994113-0.76183160.1512651440.01765133814-0.76203410.3339641340.04968625115-1.65222040.3359564530.21000046516-0.43618010.1676837740.006689189170.229854360.173649820.00195317218-0.54822220.3333488550.02606003419-0.6726780.3334119110.038904668200.146413120.1617835740.00072922821-1.59606190.1551703580.071807705222.542304980.3335516030.41360301230.221127780.3340938830.004316851240.297164550.1711513560.003201263比较值3*(8+1)/24=1.1251虽然有数据显示有异常观测值: 相关微量 暖气管温 物质含 拟合值 标准化观测值 度(X3) 量(Y) 拟合值 标准误 残差 残差 22 248 30.000 27.133 0.911 2.867 2.23R但由于该组数据杠杆率3*(8+1)/24=1.125, 第22个观测值杠杆率0.333551603,且所有观测值的杠杆率均小于1.125,所以本组数据无有影响的观测值。该方程较好的解释了在实验给定的温度范围内木炉尾气中的颗粒物质与木炉进气口设置、暖气管型号、暖气片温度之间的关系。,(S, 1/4)=44.01-0.02X3(S,1/2)=44.01-11.6-0.02X3(S,3/4)=44.01-10.5-0.02X3(S,Open)=(S, 1/4)=44.01-0.02X3(L,1/4)=(S, 1/4)=44.01-0.02X3(L,1/2)=44.01-11.9-0.02X3(L,3/42)=44.01-9.2Xf-0.02X3(L,Open)=(L, 1/4)=44.01-0.02X32. 对以上数据做最佳子集回归最佳子集回归: 相关微量物质含量(Y)与 暖气管温度(X3), Xa, Xb, Xc, Xd, Xe, Xf, Xg 响应为 相关微量物质含量(Y) 暖气管温度 R-Sq(调 Mallows X3 X X X X X X X方差 R-Sq 整) Cp S ) a b c d e f g 1 24.5 21.1 248.7 5.4578 X 1 16.5 12.7 277.2 5.7404 X 2 36.0 29.9 209.8 5.1445 X X 2 31.2 24.7 226.7 5.3316 X X 3 71.1 66.7 87.0 3.5442 X X X 3 50.0 42.5 162.0 4.6598 X X X 4 80.1 75.9 56.9 3.0179 X X X X 4 78.9 74.5 61.1 3.1051 X X X X 5 94.8 93.4 6.4 1.5772 X X X X X 5 92.1 89.9 16.1 1.9511 X X X X X 6 95.4 93.8 6.2 1.5261 X X X X X X 6 94.9 93.0 8.3 1.6206 X X X X X X 7 95.8 93.9 7.0 1.5138 X X X X X X X 7 95.5 93.5 8.1 1.5689 X X X X X X X 8 95.8 93.5 9.0 1.5618 X X X X X X X X由以上第3步知,X3与进风口设置存在多重共线性,所以最好不能同时作为变量出现在回归模型中。取以上划横线的变量做后向消元回归如下:逐步回归: 相关微量物质含量(Y) 与 Xa, Xb, Xc, Xd, Xe, Xf, Xg 后退法。 删除用 Alpha: 0.05响应为 7 个自变量上的 相关微量物质含量(Y),N = 24步骤 1 2常量 42.67 41.50Xa -15.0 -13.8T 值 -12.14 -12.05P 值 0.000 0.000Xb -16.7 -15.5T 值 -13.48 -13.50P 值 0.000 0.000Xc -9.0 -7.8T 值 -7.28 -6.82P 值 0.000 0.000Xd -2.3T 值 -1.89P 值 0.077Xe -15.3 -14.2T 值 -12.41 -12.34P 值 0.000 0.000Xf -14.3 -13.2T 值 -11.60 -11.47P 值 0.000 0.000Xg -9.0 -7.8T 值 -7.28 -6.82P 值 0.000 0.000S 1.51 1.62R-Sq 95.78 94.83R-Sq(调整) 93.93 93.01Mallows Cp 8.0 9.6得到回归方程:分析该回归方程:该方程判定系数95.83%,调整后为93.01%,各个变量的系数都有显著性意义。即尾气中的颗粒物质含量与炉子不同的暖气管型号、进气口的设置关系密切,这两者的组合可以解释空气中尾气的颗粒物质变异的93%以上,该炉子实验厂家可以通过调整炉子使用的暖气管型号及使用者调整进气口设置两种方法来控制尾气污染物质的排放,从而达到政府环保部门的要求。
展开阅读全文