WOOD 492 MODELLING FOR DECISION SUPPORT Lecture 10 Introduction to Sensitivity Analysis Why Sensitivity Analysis? • LPs are used for finding an optimal plan – Solution of key decision variables that generates the best value for the objective function, based on given parameters • LP solutions are based on the “certainty” assumption – What if we are not 100% sure about the parameter values? – How can we determine the impact of parameter changes on the optimal solution? – Which parameters are the most important one to estimate correctly based on the sensitivity of the objective function? Sept 24, 2012 Wood 492 - Saba Vahid 2 Sensitivity Analysis • Is extremely useful when making decisions based on the results of an optimization model • Helps us answer questions like: – What is the value of additional capacity/resources? – How much would the prices have to change before ….. – How sensitive is the objective value to the estimates of parameter values? – What is the range of parameter values for which the optimal solution stays the same? Sept 24, 2012 Wood 492 - Saba Vahid 3 Example for Sensitivity Anlaysis • • • • Producing 5 products using 3 moulders, 2 sander, and labour Factory works 2 shifts, 8 hours per shift, 6 days / week Each product takes 20 hrs of labour 8 workers working 48 hours / week Price Moulder Hours Sander Hours Sept 24, 2012 Prod 1 550 12 10 Prod 2 600 20 8 Prod 3 350 Prod 4 400 25 Prod 5 200 15 16 Wood 492 - Saba Vahid 4 LP formulation Price Moulder Hours Sander Hours Prod 1 Prod 2 Prod 3 Prod 1 550 12 10 Prod 2 600 20 8 Prod 4 Prod 3 350 Prod 4 400 25 Prod 5 200 15 16 Prod 5 Z Answer Upper Lower Obj Moulder Sander Labor 12.00 7.20 0.00 0.00 0.00 $550 12 10 20 $600 20 8 20 $350 $400 25 $200 15 20 20 16 20 $10,920 <= <= <= RHS 288 192 384 1. What is the value of an extra hour of moulding, sanding or labour? 2. How much more expensive should product 3 be before we start producing it? Sept 24, 2012 Wood 492 - Saba Vahid 5 Shadow price (Answering Q1) by Trial & Error Answer Upper Lower Obj Moulder Sander Labor Prod 1 Prod 2 Prod 3 Prod 4 Prod 5 12.00 7.20 0.00 0.00 0.00 $550 12 10 20 $600 20 8 20 $350 $400 25 $200 15 20 20 Z new = $10,926.25 Z old = $10,920 16 20 $10,920 <= <= <= RHS 288 192 384 Change to 289 Difference (shadow price)= $6.25 = value of an additional hour of moulder time Shadow price: marginal value of a resource/constraint. Can be calculated by adding 1 to the RHS of a constraint and calculating the difference in the objective function. Reduced Cost (Answering Q2) by Trial & Error Answer Upper Lower Obj Moulder Sander Labor Prod 1 Prod 2 Prod 3 Prod 4 Prod 5 12.00 7.20 0.00 0.00 0.00 $550 12 10 20 $600 20 8 20 $350 $400 25 $200 15 20 20 16 20 $10,920 <= <= <= RHS 288 192 384 Increase gradually Price of product 3 has to increase by $125 before it would be produced. Reduced cost = -125 Reduced Cost: If a variable = 0 in the optimal solution, then its reduced cost is the amount its objective function coefficient (price in this example) needs to change before it will come into the solution (>0). New value of Obj. Coefficient = Old value of Obj. Coeff – Reduced Cost Sept 24, 2012 Wood 492 - Saba Vahid 7 Sensitivity Analysis with Simplex • When LP problems are large with many variables and constraints – Re-solving LPs may require a large computational effort – Simplex algorithm eliminates the need to resolve the LP for every change in parameters – While we won’t get into the details of sensitivity analysis with Simplex method, we can view the results in Excel Solver’s sensitivity report Sept 24, 2012 Wood 492 - Saba Vahid 8 Example of the sensitivity report in Excel • Furniture Manufacturer (refer to Example_3 files on the course website) How sensitive is the solution to this estimated price? Variable Tables Desks Dressers Answer 22.00 33.33 20.00 $3,160 Obj Ftn $50.00 $45.00 $28 used sign RHS Saw-hrs 9 3 3 358 <= 500 Assemble-hrs 5 6 2 350 <= 350 Finish-hrs 5 3 2 250 <= 250 1 20 <= 20 Sales Demand How would the solution change if we had more assembly hours? Sensitivity report Sept 24, 2012 Wood 492 - Saba Vahid 9 Lab 3 Preview CB 1 4000 m3 Sawdust (m3) Log deck CB 2 5000 m3 Logs sorted by diameter (m3) Head saw Planks and boards (MBF) Trimmer Lumber products (MBF) Market Chips (m3) CB 3 2500 m3 Lab 3 Preview Sept 24, 2012 Wood 492 - Saba Vahid 10 Lab 3 Preview • Objective: maximize profits ($) • Decision variables: – how much to cut from each cut block (m3) – How much logs to process with each sawing pattern(m3) – Which lumber products to produce (mbf) • Constraints – – – – – Available timber (volume in the cut blocks) Machine hours available, downtime Material balance related to conversion factors No demand constraints Sawing patterns are not mutually exclusive (we can use a combination of SP1 and SP2) Sept 24, 2012 Wood 492 - Saba Vahid 11 Some unit conversions • 1 MBF = volume of a block of 1” by 12” by 12” – 1BF= 144 cubic inch – 1 BF=0.0254*0.3048*0.3048 cubic meters ≈ 0.002360 m3 – 1000 BF=1 MBF= 2.36 m3 • Calculating chip conversion factors (m3/m3*, unitless): *Note: In this lab, the chip conversion factors are given in relation to the total log volume, not the lumber volume Total log vol = lumber vol+ sawdust + chips (all units should be in m3) = lumber vol(MBF)*2.360 (m3/MBF)+ 0.1 * total log vol+ chips chips= total log vol - 0.1 total log vol- lumber vol * 2.360 = 0.9 total log volume – (lumber recovery factor* total log vol)*2.360 =total log volume * (0.9 – lumber recovery factor * 2.360) Sept 24, 2012 Wood 492 - Saba Vahid 12 Lab 3 preview • Speed of Head saw = 250 ft/minute = 15,000 ft/hr – Accounting for 10% downtime= 15000 *0.9= 13,500 ft/hr – Each log is 8 ft tall and there’s a 4 ft gap between logs so for every 12 ft of conveyor, there is one log passing – 13,500/12= 1125 logs/hour (passing through head saw) – Combination of this log/hr rate and log volumes is used to calculate processing time (hr/m3) for the head saw, as shown in the Excel file • Speed of Trimmer= 95 lugs/minute = 5,700 lugs/hr – Accounting for 90% coverage= 5,700 *0.9=5,130 boards/hr – Accounting for 10% downtime= 5,130 *0.9= 4,617 boards/hr – Combination of this board/hr rate and lumber volumes is used to calculate processing time (hr/MBF) for the trimmer, as shown in the Excel file Sept 24, 2012 Wood 492 - Saba Vahid 13 Next Class • Quiz on Friday, Sept 28 • More sensitivity analysis examples Sept 24, 2012 Wood 492 - Saba Vahid 14