1 / 20

MGTSC 352

MGTSC 352. Lecture 15: Aggregate Planning Altametal Case Summary of Optimization Modeling. AltaMetal Ltd. (Case 8, pg. 111, and pgs. 87 – 92). Another aggregate planning problem 1,000 products aggregated to 9 groups. AltaMetal Ltd. (Case 8, pg. 111, and pgs. 87 – 92).

giulio
Download Presentation

MGTSC 352

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. MGTSC 352 Lecture 15: Aggregate Planning Altametal Case Summary of Optimization Modeling

  2. AltaMetal Ltd. (Case 8, pg. 111, and pgs. 87 – 92) • Another aggregate planning problem • 1,000 products aggregated to 9 groups

  3. AltaMetal Ltd.(Case 8, pg. 111, and pgs. 87 – 92) Is it possible to satisfy demand? If so, how? (production plan by product group) Excel …

  4. Active Learning • Pairs, 1 min. • Formulate AltaMetal’s problem in English • What to optimize, by changing what, subject to what constraints …

  5. To many change-overs … • The JIT (“just-in-time”) plan we found may require too many changeovers • What if we require a minimum lot size of 30 tons? • Daily capacity = 90 tons  At most 3 lots per day • Changing cells: • Old: # of tons of product X to produce in month Y • New: # of __ of product X to produce in month Y Excel …

  6. Tired of Waiting for Solver? • Hit Escape key

  7. LINEARINTEGERNONLINEAR“Programming” MODELS A Summary

  8. CLASSIFICATION Decision Variables FunctionsFractionalInteger LinearLPILP NonlinearNLPINLP

  9. LP • SIMPLEX method (linear algebra) • Corner point optimality • Move from corner-to-corner, improve obj. • Very efficient • Can solve problems with thousands of variables and constraints

  10. ILP • Branch & Bound (divide-and-conquer) • Solve the LP, ignoring integer constraints • Select a fractional variable, x6 = 15.7 • Create two new problems: x6≤ 15, x6 16 • Solve the new problems • Continue until all branches exhausted • # of branches is exponential in # of var.

  11. NLP • Gradient method (uses derivatives) • Repeat until convergence • Find an improving direction • Move in the improving direction • Converges to local optimum • Multiple starts recommended

  12. INLP • Ignore integer constraints, solve the NLP • Use Branch & Bound • Solve a series of NLPs • Computationally demanding • No guarantee of optimality • YUCK!

  13. Formulating Optimization Models (pg. 93) • Formulate the problem in English • Or French, or Chinese, or Icelandic, …  • Start with data in spreadsheet • Define decision variables – turquoise cells • Express performance measure (profit, or cost, or something else) as function of the decision variables • Express constraints on decision variables • Scarce resources • Physical balances • Policy constraints

  14. Solving Optimization Problems • Try simple values of the decision variables to check for obvious errors • Guess at a reasonable solution and see if model is ‘credible’ (sniff test) • Look for missing or violated constraints • Is profit (cost) in ballpark?

  15. Optimizing with Solver • Use Simplex LP method (‘assume linear model’) whenever possible • Set Options properly • automatic scaling, assume non-negative • Watch for diagnostic messages – do not ignore! (infeasible, unbounded) • Interpret solution in real-world terms and again check for credibility

  16. Things to Remember • The Simplex LP method always correctly solves linear programs • Solver is a slightly imperfect implementation of the Simplex method (but you should generally assume that it is correct) • The biggest source of errors is in the model building process (i.e., the human)

More Related