IBM - Decision Optimization
- 1. © 2017 IBM Corporation Decision Optimization Sebastian Fink Client Technical Professional Decision Optimization sebastian.fink@de.ibm.com
- 2. 3 © 2017 IBM Corporation ▪ Let’s look at an example of sequence dependent setups ▪ Time shown is the downtime necessary to switch from one job to the next. ▪ Determine a sequence of jobs that minimizes the combined set-up time. ▪ Assuming Job 1 is running, what is the best sequence? Optimization – an example Job 1 Job 2 Job 3 Job 4 Job 5 1 2 3 4 5 1 -- 8 4 5 3 2 -- -- 12 5 5 3 -- 9 -- 7 3 4 -- 1 10 -- 15 5 -- 7 13 3 -- FromJob# To Job #
- 3. 4 © 2017 IBM Corporation Optimization – an example: Schedules and Evaluation 1st 2nd 3rd 4th 5th Setup 1-2 Setup 2-3 Setup 3-4 Setup 4-5 Total Setup % optimal Job # 1 2 3 4 5 8 12 7 15 42 382% 1 4 5 3 2 5 15 13 9 42 382% 1 2 4 5 3 8 5 15 13 41 373% 1 4 5 2 3 5 15 7 12 39 355% 1 2 5 3 4 8 5 13 7 33 300% 1 3 2 4 5 4 9 5 15 33 300% 1 3 4 5 2 4 7 15 7 33 300% 1 5 3 2 4 3 13 9 5 30 273% 1 4 3 2 5 5 10 9 5 29 264% 1 5 2 3 4 3 7 12 7 29 264% 1 2 3 5 4 8 12 3 3 26 236% 1 2 4 3 5 8 5 10 3 26 236% 1 2 5 4 3 8 5 3 10 26 236% 1 4 3 5 2 5 10 3 7 25 227% 1 5 2 4 3 3 7 5 10 25 227% 1 5 4 3 2 3 3 10 9 25 227% 1 4 2 5 3 5 1 5 13 24 218% 1 5 3 4 2 3 13 7 1 24 218% 1 3 2 5 4 4 9 5 3 21 191% 1 4 2 3 5 5 1 12 3 21 191% 1 3 5 2 4 4 3 7 5 19 173% 1 5 4 2 3 3 3 1 12 19 173% 1 3 4 2 5 4 7 1 5 17 155% 1 3 5 4 2 4 3 3 1 11 100% First come, first serve
- 4. 5 © 2017 IBM Corporation Optimization – an example: Schedules and Evaluation 1st 2nd 3rd 4th 5th Setup 1-2 Setup 2-3 Setup 3-4 Setup 4-5 Total Setup % optimal Job # 1 2 3 4 5 8 12 7 15 42 382% 1 4 5 3 2 5 15 13 9 42 382% 1 2 4 5 3 8 5 15 13 41 373% 1 4 5 2 3 5 15 7 12 39 355% 1 2 5 3 4 8 5 13 7 33 300% 1 3 2 4 5 4 9 5 15 33 300% 1 3 4 5 2 4 7 15 7 33 300% 1 5 3 2 4 3 13 9 5 30 273% 1 4 3 2 5 5 10 9 5 29 264% 1 5 2 3 4 3 7 12 7 29 264% 1 2 3 5 4 8 12 3 3 26 236% 1 2 4 3 5 8 5 10 3 26 236% 1 2 5 4 3 8 5 3 10 26 236% 1 4 3 5 2 5 10 3 7 25 227% 1 5 2 4 3 3 7 5 10 25 227% 1 5 4 3 2 3 3 10 9 25 227% 1 4 2 5 3 5 1 5 13 24 218% 1 5 3 4 2 3 13 7 1 24 218% 1 3 2 5 4 4 9 5 3 21 191% 1 4 2 3 5 5 1 12 3 21 191% 1 3 5 2 4 4 3 7 5 19 173% 1 5 4 2 3 3 3 1 12 19 173% 1 3 4 2 5 4 7 1 5 17 155% 1 3 5 4 2 4 3 3 1 11 100% Rule based heuristics First come, first serve
- 5. 6 © 2017 IBM Corporation Optimization – an example: Schedules and Evaluation 1st 2nd 3rd 4th 5th Setup 1-2 Setup 2-3 Setup 3-4 Setup 4-5 Total Setup % optimal Job # 1 2 3 4 5 8 12 7 15 42 382% 1 4 5 3 2 5 15 13 9 42 382% 1 2 4 5 3 8 5 15 13 41 373% 1 4 5 2 3 5 15 7 12 39 355% 1 2 5 3 4 8 5 13 7 33 300% 1 3 2 4 5 4 9 5 15 33 300% 1 3 4 5 2 4 7 15 7 33 300% 1 5 3 2 4 3 13 9 5 30 273% 1 4 3 2 5 5 10 9 5 29 264% 1 5 2 3 4 3 7 12 7 29 264% 1 2 3 5 4 8 12 3 3 26 236% 1 2 4 3 5 8 5 10 3 26 236% 1 2 5 4 3 8 5 3 10 26 236% 1 4 3 5 2 5 10 3 7 25 227% 1 5 2 4 3 3 7 5 10 25 227% 1 5 4 3 2 3 3 10 9 25 227% 1 4 2 5 3 5 1 5 13 24 218% 1 5 3 4 2 3 13 7 1 24 218% 1 3 2 5 4 4 9 5 3 21 191% 1 4 2 3 5 5 1 12 3 21 191% 1 3 5 2 4 4 3 7 5 19 173% 1 5 4 2 3 3 3 1 12 19 173% 1 3 4 2 5 4 7 1 5 17 155% 1 3 5 4 2 4 3 3 1 11 100% Optimal Rule based heuristics First come, first serve
- 6. 7 © 2017 IBM Corporation IBM Bluemix + WDCE ▪ How did your solution do? ▪ There are 24 possible combinations to schedule 5 jobs : ‒ 4 Options to chose from for the second job ‒ 3 for the third ‒ 2 to chose the fourth job ‒ The last job remaining gets to go fifth ▪ number of combinations grows very fast Decision Optimization – an example # Jobs # combinations 3 2 4 6 5 24 6 120 7 720 8 5.040 9 40.320 10 362.880 11 3.628.800 12 39.916.800 13 479.001.600
- 7. 8 © 2017 IBM Corporation Optimization – The Science of Better Decisions What to build, where and when? How to best allocate aircrafts and crews? Risk vs. potential reward? Inventory cost vs. customer satisfaction? Cost vs. carbon emission? Optimization helps businesses: • create the best possible plans • explore alternatives and understand trade-off • respond to changes in business operations
- 8. 9 © 2017 IBM Corporation Predictive Analytics Why did it happen? What will happen? Proactive Analytics What should we do? Descriptive Analytics What has happened? Information Insight Plan DecideAnalyze Business Value Act Competitive Advantage + + Standard/ad Hoc Reports Query, Drill down Optimization, Uncertainty mgt Predictive modelling, Forecast, Simulation, Alerts Proactive Analytics for Decision Planning and Action What is Optimization?
- 9. 10 © 2017 IBM Corporation Prescriptive Analytics – How Does It Work? Inputs typically come from ERP systems, silo’d DBs or entered manually 1 In many cases ”Demand to be Met” comes from a Predictive Analytics Solution like SPSS 2 The model is typically created by an OR Professional and is specific to the business problem and industry 3 The engine is the run-time component of the IBM Decision Optimization software 4 Output is typically written back to an ERP or a DB and can be consumed through a BI solution 5
- 10. 11 © 2017 IBM Corporation IBM Bluemix + WDCE Optimization Engines – CPLEX – Overview ▪ Roots in Analytics Geometry and Matrix Algebra ▪ Used to solve strategic and tactical resource allocation questions where optimality is important. ▪ Typical applications ‒ Large-scale. ‒ High performance. ‒ Mission critical. ‒ Versatile: embeddable & stand-alone. ‒ Sophisticated analysis. ▪ Problem Types: ‒ Linear (LP), Mixed-Integer (MIP) and Quadratic (QP, QCP, MIQP, MIQCP). ▪ What is inside? ‒ Simplex optimizers (Primal, Dual, Network), Barrier Optimizer ‒ Branch-and-Cut algorithm ‒ Heuristics (Genetic algorithms, solution polishing, neighborhood search etc.)
- 11. 12 © 2017 IBM Corporation IBM Bluemix + WDCE Optimization Engines – CPLEX – Key Features ▪ Parallelization of Algorithms ‒ Takes advantage of multiple cores and speeds up search (shared memory architectures) ‒ “remote object” API to allow parallelization on distributed memory architectures ▪ Infeasibility Analysis ‒ Detects minimal set of constraints that causes infeasibility ‒ Provides recommendations for relaxations that will fix infeasibility ▪ Solution Pool ‒ Generation of multiple solutions for every problem ‒ Alternative optimal solutions or solution within a % of optimality ▪ Parameters and Callbacks ‒ Extensive set of parameters that experts can fine tune ‒ Callback code written by experts can be executed at runtime together with native code
- 12. 13 © 2017 IBM Corporation IBM Bluemix + WDCE Optimization Engines – CPLEX – Key Concepts ▪ Optimality Gap ‒ How far away is my current solution from the best possible solution? TIME SOLUTIONCOST Current Solution Theoretical Bound Large gap indicating a poor solution. As the search progresses the gap is reduced. When the solution and the theoretical optimal become the same we have proof of optimality.
- 13. 16 © 2017 IBM Corporation IBM Bluemix + WDCE Challenge Solution Benefits/ROI Saving approximately €3,36m and gaining efficiency in medical staff scheduling Composed of four major hospitals the client was challenged with accurately assigning its medical professionals to rotating shifts while adhering to their preferences, regulations and equitable distribution of vacation time. The hospital uses IBM Optimization technology to schedule their staff and optimize the availability of more than 7,000 medical staff in a rotating shift environment. The solution automatically applies thousands of rules and constraints such as industry regulations, staff preferences, and specialty areas to rapidly generate daily shifts and help ensure equitable scheduling while accommodating for unplanned situations in near-real time with as many reiterations of schedules as needed. Reduced manual procedures for each supervisor by 35%, saving approximately €10,500 per manager annually Decreases manual scheduling tasks by 4 %, saving €960 per nurse or totally €3,36m Reduces operational costs by eliminating pay errors, accurately tracking hours worked and accounting for the remuneration that vary with each employee This leading public health center and hospital in Madrid was founded in 1964. With 7,000 employees it serves more than 50,000 inpatients and 220,000 outpatients annually. “With the new solution, we’re more confident in our ability to handle frequent and unexpected changes that affect daily shifts and to meet the quality and level of care our patients expect from our organization.” Customer Profile
- 14. 17 © 2017 IBM Corporation IBM Bluemix + WDCE Challenge Solution Benefits/ROI Crew Recovery at Continental Airlines Re-assign crews quickly following service disruptions to cover open flights and return them to their original schedules in a cost-effective manner. CrewSolver decision-support system for Continental Airlines to generate globally optimal, or near optimal, crew-recovery solutions. Customer Profile Continental Airlines is the fifth largest United States airline and operates more than 2,000 daily departures to 123 domestic and 93 foreign destinations. •First airline to recover following the December 2000 and March 2001 Nor’easter snowstorms, the June 2001 Houston flood, and the September 11th terrorist attacks. •$40MM in savings for each major service disruption •Rated first in on-time performance for the 12 months ending in August 2002.
- 15. 19 © 2017 IBM Corporation IBM Bluemix + WDCE
- 16. 20 © 2017 IBM Corporation IBM Bluemix + WDCE 20 Challenge Solution Benefits/ROI ▪ An IBM Decision Optimisation based price optimisation solution to provide ability to pinpoint optimal pricing for hotel rooms quickly and accurately that increases the revenue it derives and decrease vacancies ▪ The solution gathers data from multiple sources, including the reservation system and market response, demand forecast and competitive rates databases ▪ Using powerful algorithms, the solution determines optimal room rates based on factors such as occupancy, price elasticity and competitors’ prices Customer Profile ▪ Increases the revenue per available room (RevPAR) by 2.7% ▪ Increases incremental revenue by USD 300 million per year ▪ Reduces vacancies and missed revenue opportunities, sharpening the competitive edge ▪ Striving to improve profitability and limit vacancies by setting optimal prices for its rooms ▪ Used spreadsheets and reports to analyse factors such as competing prices, market research and seasonal demand trends when setting daily rates ▪ With more than 76,000 potential pricing decisions to consider daily, the efficiency of the spreadsheet approach is not consistent, leading to missed revenue ▪ Needed a solution to analyse pricing factors in near-real time and offer its inventory of rooms to customers at optimal rates ▪ A leading hotel company with global presence ▪ Operates more than 4,600 hotels in nearly 100 countries ▪ Manages an inventory of 687,000 rooms, spread over 9 major brands
- 17. 21 © 2017 IBM Corporation IBM Bluemix + WDCE 21 Challenge Solution Benefits/ROI Unit Commitment at Red Electrica de Espana • Optimization solutions from IBM provided operational advantages to REE’s managers and engineers enabling them to simplify all maintenance tasks and changes made to the model, thereby significantly reducing planning time and associated costs. Customer Profile • Reduced production costs by between €50,000 and €100,000 per day. • Reduced its carbon emissions by approximately 100,000 tons of CO2 annually. • Simplifies all maintenance tasks and any changes made to the model, which are very frequent. • Brought greater trust in the solution and a significant reduction in planning time required by users. • Red Eléctrica de España, in charge of managing the Spanish national power grid needed to replace the approximate heuristic methods it had been using for the last 20 years. • Red Eléctrica is the sole transmission agent and operator of the Spanish electricity system. • Its mission is to ensure the global functioning of the system guaranteeing at each moment the continuity and security of supply. Generation
- 18. 23 © 2017 IBM Corporation IBM Bluemix + WDCE
- 19. 24 © 2017 IBM Corporation IBM Bluemix + WDCE
- 20. 25 © 2017 IBM Corporation IBM Bluemix + WDCE More examples https://github.com/IBMDecisionOptimization/docplex-examples/tree/master/examples/cp/jupyter Sudoku House Building Box Placement Unit Commitment
- 21. 26 © 2017 IBM Corporation IBM Bluemix + WDCE ▪ Discover and learn Optimization through a 4h online class training: ‒ mathematical-optimization-for-business-problems/ (course code: CP0101EN on cognitiveclass.ai) ▪ CPLEX Tutorial Notebooks on our github: ‒ Linear_Programming.html ‒ Beyond_Linear_Programming.html ▪ Scheduling Tutorial Notebooks: ‒ keep an eye on DSX blog in the coming days! New to Optimization?