Vertiefung zu Logistik/SCM [1.5ex] Discrete Location with Stochastic
Transcrição
Vertiefung zu Logistik/SCM [1.5ex] Discrete Location with Stochastic
16.04.2013 Vertiefung zu Logistik/SCM 1 / 37 Vertiefung zu Logistik/SCM Discrete Location with Stochastic Demand Univ.-Prof. Dr. Knut Haase Institut für Verkehrswirtschaft 16. April 2013 16.04.2013 Vertiefung zu Logistik/SCM Inhalt 1. Introduction Assumptions Application in mind 2. Modeling the behavior of the potential customers Choice Probability Derived from the Multinomial Logit Model Independence of irrelevant alternatives (IIA-property) Estimation of Discrete Choice Model 3. Integration of the MNL in location planning Notation 4. Nonlinear location model Linear reformulation by Benati/Hansen (2002) Linear reformulation by Haase (2009) Linear reformulation by Zhang et al. (2012) 5. Numerical investigations 6. Case study and numerical results 2 / 37 16.04.2013 Vertiefung zu Logistik/SCM 3 / 37 Assumptions I I I I Given set of potential locations Static competition Customer choose location with maximum utility Choice behavior can be modeled by multinomial logit model 16.04.2013 Vertiefung zu Logistik/SCM Application in mind I I I I I Network of branch offices of parcel service provider City of Dresden Considered company: Deutsche Post Competitor: Hermes Logistics Group 4 / 37 16.04.2013 Vertiefung zu Logistik/SCM 5 / 37 Vertiefung zu Logistik/SCM 6 / 37 Application in mind II 16.04.2013 Inhalt 1. Introduction Assumptions Application in mind 2. Modeling the behavior of the potential customers Choice Probability Derived from the Multinomial Logit Model Independence of irrelevant alternatives (IIA-property) Estimation of Discrete Choice Model 3. Integration of the MNL in location planning Notation 4. Nonlinear location model Linear reformulation by Benati/Hansen (2002) Linear reformulation by Haase (2009) Linear reformulation by Zhang et al. (2012) 5. Numerical investigations 6. Case study and numerical results 16.04.2013 Vertiefung zu Logistik/SCM 7 / 37 Choice Probability Derived from the Multinomial Logit Model I Notation ij independently and identical extreme value distributed (i = individual, j = alternative) zijm value of attribute m jm utility per unit of attribute m; variable to be estimated 16.04.2013 Vertiefung zu Logistik/SCM 8 / 37 Density-function and cumulative distribution function of Gumbel distribution (extreme value distribution) F (ij ) = e 1 e ij 0:8 0:6 0:4 0:2 f (ij ) = e ij 0 2 0 2 ij 4 6 e e ij 16.04.2013 Vertiefung zu Logistik/SCM 9 / 37 Utility of location j for individual i uij = vij + ij vij = X jm zijm m Choice probability (Train, 2009) e vij pij = P v ik ke 16.04.2013 Vertiefung zu Logistik/SCM 10 / 37 Independence of irrelevant alternatives (IIA-property) I Ratio of choice probabilities of two alternatives j ; j 0 : pij pij 0 = = = vij Pe vik ke = vij 0 Pe vik ke e vij e vij 0 constant I Adding or removing an alternative does not change the ratio of choice probabilities (constant substitution patterns). 16.04.2013 Vertiefung zu Logistik/SCM 11 / 37 Estimation of Discrete Choice Model Software packages for determination of ˆ : I I I I 16.04.2013 BIOGEME, LIMDEP, NLOGIT, SPSS and R. Vertiefung zu Logistik/SCM Estimation of an MNL with BIOGEME BIOGEME is an acronym for BIerlaire Optimization toolbox for Generalized Extreme value Model Estimation (see [Bie03] und [Bie08]). The software public domain and can be downloaded under http://biogeme.epfl.ch for the standard operating systems. Execution of BIOGEME from command window (cmd.exe). 12 / 37 16.04.2013 Vertiefung zu Logistik/SCM 13 / 37 Files to be prepared Model-file Includes the definition of the utility functions. File name needs the extension .mod, e.g. mymodel.mod. Data-file Contains the observations as table. Filename, e.g.: mysample.dat Excution of BIOGEME biogeme mymodel mysample.dat Result file mymodel.rep. 16.04.2013 Vertiefung zu Logistik/SCM Example Students which have no car available go to their university on foot (j = 1), by bike (j = 2) or by bus (j = 3). The decision of student i depends on the travel time dij [in minutes] and on the sex si (1= female, 0 = male). The deterministic component of the utiliy function is as follows: vi1 vi2 vi3 01 + 11 di1 + 21 si = 02 + 12 di2 + 22 si = 03 + 13 di3 + 23 si = For avoiding problems of identification we set (in this case) 01 = 11 = 21 = 0; i.e. vi1 = 0 for all students n 14 / 37 16.04.2013 Vertiefung zu Logistik/SCM 15 / 37 Our model file mymodel.mod is constructed as follows: [Choice] Choice [Beta] // Name b01 b02 b03 b11 b12 b13 b21 b22 b23 Value 0 0 0 0 0 0 0 0 0 LowerBound -10000 -10000 -10000 -10000 -10000 -10000 -10000 -10000 -10000 UpperBound 10000 10000 10000 10000 10000 10000 10000 10000 10000 status (0=variable, 1=fixed) 1 0 0 1 0 0 1 0 0 [Utilities] // Id Name Avail linear-in-parameter expression 16.04.2013 1 2 3 Vertiefung zu Logistik/SCM Foot av1 b01 * one + b11 * d1 + b21 * s Bike av2 b02 * one + b12 * d2 + b22 * s Bus av3 b03 * one + b13 * d3 + b23 * s [Expressions] one = 1 av1 = 1 av2 = 1 av3 = 1 [Model] $MNL 16 / 37 16.04.2013 Vertiefung zu Logistik/SCM 17 / 37 Excerpt of our data file mysample.dat: ID Choice 1 1 2 1 3 1 4 2 5 1 : : 95 3 96 1 97 1 98 2 99 1 100 1 16.04.2013 d1 6 8 12 21 6 : 51 18 6 33 19 15 d2 8 9 9 12 7 : 21 11 9 15 11 12 d3 13 13 15 17 12 : 24 14 12 16 13 16 s 0 1 1 0 0 : 0 1 0 1 0 1 Vertiefung zu Logistik/SCM Excerpt of the output file of BIOGEME mymodel.rep: Model: Number of estimated parameters: Number of observations: Number of individuals: Null log-likelihood: Cte log-likelihood: Init log-likelihood: Final log-likelihood: Likelihood ratio test: Rho-square: Adjusted rho-square: Final gradient norm: Diagnostic: Iterations: Run time: Multinomial Logit 6 100 100 -109.861 -88.991 -109.861 -68.002 83.718 0.381 0.326 +4.082e-07 Convergence reached... 11 00:00 18 / 37 16.04.2013 Vertiefung zu Logistik/SCM Utility parameters ****************** Name Value Std err t-test p-val Rob. std err Rob. t-test Rob. p-val ---- ----- ------- ------ ----- ------------ ----------- ---------b01 0.00 --fixed-b02 -5.57 1.15 -4.84 0.00 1.01 -5.50 0.00 b03 -10.4 2.17 -4.78 0.00 2.02 -5.13 0.00 b11 0.00 --fixed-b12 0.324 0.0746 4.35 0.00 0.0665 4.87 0.00 b13 0.554 0.123 4.52 0.00 0.115 4.81 0.00 b21 0.00 --fixed-b22 -0.128 0.617 -0.21 0.84 * 0.580 -0.22 0.83 b23 -0.317 0.678 -0.47 0.64 * 0.666 -0.48 0.63 16.04.2013 * * Vertiefung zu Logistik/SCM Deterministic components of utility function: vi1 = vi2 = vi3 = Model statistics L(ˆ ) -68.002 2 0.381 0 5:57 + 0:324 di2 10:4 + 0:554 di2 19 / 37 0:128 si 0:317 si 20 / 37 16.04.2013 Vertiefung zu Logistik/SCM 21 / 37 Inhalt 1. Introduction Assumptions Application in mind 2. Modeling the behavior of the potential customers Choice Probability Derived from the Multinomial Logit Model Independence of irrelevant alternatives (IIA-property) Estimation of Discrete Choice Model 3. Integration of the MNL in location planning Notation 4. Nonlinear location model Linear reformulation by Benati/Hansen (2002) Linear reformulation by Haase (2009) Linear reformulation by Zhang et al. (2012) 5. Numerical investigations 6. Case study and numerical results 16.04.2013 Vertiefung zu Logistik/SCM 22 / 37 Modified notation I Notation typically in location literature i demand point (homogenous customers) M locations; indices j ; k J locations of considered company (J M) Utility of homogenous customer in i to choose location j uij = vij + ij (1) 16.04.2013 Vertiefung zu Logistik/SCM 23 / 37 Modified notation II Choice probability pij = P =P 16.04.2013 e vij vik k 2M e e vij P vij vik + e k 2M nJ j 2J e Vertiefung zu Logistik/SCM Inhalt 1. Introduction Assumptions Application in mind 2. Modeling the behavior of the potential customers Choice Probability Derived from the Multinomial Logit Model Independence of irrelevant alternatives (IIA-property) Estimation of Discrete Choice Model 3. Integration of the MNL in location planning Notation 4. Nonlinear location model Linear reformulation by Benati/Hansen (2002) Linear reformulation by Haase (2009) Linear reformulation by Zhang et al. (2012) 5. Numerical investigations 6. Case study and numerical results (2) (3) 24 / 37 16.04.2013 Vertiefung zu Logistik/SCM 25 / 37 Nonlinear location model Variables 2 J is established by considered company (0, yj =1, if location j sonst) xij expected market share of location j in demand point i (corresponds to pij ) 16.04.2013 Vertiefung zu Logistik/SCM 26 / 37 Model formulation Assumption In each demand point the in i : 1 unit max F = X 2 P i I subject to X 2 P 2 n j M J 2 j Je e vij + vij y Pj vij j 2J e yj yj = r j J yj 2 f0; 1g (4) (5) 8j 2J (6) 16.04.2013 Vertiefung zu Logistik/SCM 27 / 37 Linearization by Benati/Hansen (2002) Additional defined parameter 'ij = max F = P e vij (7) vik k 2M nJ e XX 2 2 'ij (yj xij ) (8) i I j J subject to (5), (6), and yj xij + X 2 'ik (yk xik ) 1 8 i 2 I; j 2 J (9) xij 0 8 i 2 I; j 2 J (10) k J 16.04.2013 Vertiefung zu Logistik/SCM 28 / 37 Linear reformulation by Haase (2009) Non-negative variable (fraction of competitors): x̃i max F = XX 2 2 xij (11) i I j J subject to (5), (6), (10), and x̃i + X 2 xij 1 8i 2I (12) j J xij 'ij yj 1 + 'ij xij 'ij x̃i x̃i 0 0 0 8 i 2 I; j 2 J 8 i 2 I; j 2 J 8i 2I (13) (14) (15) The model of [Haa09] has been analogously presented by [AVMM13]. 16.04.2013 Vertiefung zu Logistik/SCM 29 / 37 Linear reformulation by Zhang (2012) Non-negative variable: zijk (11) subject to (5), (6), (10), and xij 'ij yj + X 2 'ik zijk = 0 k J zijk xij zijk yk xij + yk zijk zijk 16.04.2013 0 0 1 0 8 i 2 I; j 2 J 8 i 2 I; 8 i 2 I; 8 i 2 I; 8 i 2 I; 2J j; k 2 J j; k 2 J j; k 2 J j; k Vertiefung zu Logistik/SCM Inhalt 1. Introduction Assumptions Application in mind 2. Modeling the behavior of the potential customers Choice Probability Derived from the Multinomial Logit Model Independence of irrelevant alternatives (IIA-property) Estimation of Discrete Choice Model 3. Integration of the MNL in location planning Notation 4. Nonlinear location model Linear reformulation by Benati/Hansen (2002) Linear reformulation by Haase (2009) Linear reformulation by Zhang et al. (2012) 5. Numerical investigations 6. Case study and numerical results (16) (17) (18) (19) (20) 30 / 37 16.04.2013 Vertiefung zu Logistik/SCM 31 / 37 Numerical investigations I GAMS (23.7) and use CPLEX 12 on a 64-bit Windows Server 2008 with 1 Intel Xeon 2.4 GHz processor and 24 GB RAM I Cartesian coordinates blocks (demand nodes) and locations by a uniform distribution in the interval [0; 30] I vri = 0:2 dri jR j jI j 200 25 50 400 25 50 S CPU [BH02] GAP DEV CPU [Haa09] GAP DEV CPU [ZBV12] GAP DEV 2 4 2 4 2 4 2 4 10.52 267.27 153.40 3591.20 35.41 976.47 745.41 3600.00 0 0 0 11.99 0 0 0 29.12 0 0 0 0.05 0 0 0 0.81 8.04 64.22 67.10 443.97 38.10 213.06 241.86 1548.51 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1348.36 3600.00 3600.00 3600.00 3323.68 3600.00 3600.00 3600.00 0 28.91 19.01 9.79 - 0 5.44 4.39 15.72 1.65 14.62 19.16 18.48 16.04.2013 Vertiefung zu Logistik/SCM Inhalt 1. Introduction Assumptions Application in mind 2. Modeling the behavior of the potential customers Choice Probability Derived from the Multinomial Logit Model Independence of irrelevant alternatives (IIA-property) Estimation of Discrete Choice Model 3. Integration of the MNL in location planning Notation 4. Nonlinear location model Linear reformulation by Benati/Hansen (2002) Linear reformulation by Haase (2009) Linear reformulation by Zhang et al. (2012) 5. Numerical investigations 6. Case study and numerical results 32 / 37 16.04.2013 Vertiefung zu Logistik/SCM 33 / 37 Case study and numerical results I Deutsche Post in City of Dresden I I I I 6.406 blocks (i) Post: 49 shops (I ) Hermes (competitor): 69 shops j M j= 49 + 69 = 118 16.04.2013 Vertiefung zu Logistik/SCM 34 / 37 Case study and numerical results II Utility function Empirical investigation (revealed preferences) (Hoppe, 2009) Variable Log. distance to shop [m] Distance to shop < 1800 m? (ja=1, nein=0) Is shop of Deutsche Post? (ja=1, nein=0) Shop located in centre (ja=1, nein=0) business hours per week [h] number of observations ¯2 ˆ -1,86 1,02 1,48 1,60 0,028 637 0,689 t-Wert -7,30 2,95 4,09 3,70 3,47 16.04.2013 Vertiefung zu Logistik/SCM 35 / 37 Optimal solution of location model Haase (2009) F : 0.6699 16.04.2013 CPU: 957.11 seconds Vertiefung zu Logistik/SCM Literaturhinweise I Aros-Vera, Felipe, Vladimir Marianov und John E. Mitchell: p-Hub approach for the optimal park-and-ride facility location problem. European Journal of Operational Research, 226(2):277 – 285, 2013. Benati, Stefano und Pierre Hansen: The maximum capture problem with random utilities: Problem formulation and algorithms. European Journal of Operational Research, 143:518–530, 2002. Bierlaire, M.: BIOGEME: a free package for the estimation of discrete choice models. In: 3rd Swiss Transport Research Conference, 2003. Bierlaire, M.: An introduction to BIOGEME (Version 1.8), 2008. Haase, Knut: Discrete Location Planning. Technischer Bericht WP-09-07, Institute for Transport and Logistics Studies, University of Sydney, 2009. 36 / 37 16.04.2013 Vertiefung zu Logistik/SCM Literaturhinweise II Zhang, Yue, Oded Berman und Vedat Verter: The impact of client choice on preventive healthcare facility network design. OR Spectrum, 34:349–370, 2012. 37 / 37