Optimization
Once you know what is likely, the question is what to do. We write the decision down as a mathematical model and solve it.
What we work on
- Allocation
- Which stock, machines, people or budget go where, when there is less of them than everyone would like.
- Planning and scheduling
- What is produced on which line in which order. Shift plans, maintenance windows, project schedules.
- Networks and logistics
- Which site serves which customer, where stock is held, how goods are routed.
- Purchasing and pricing
- How much to buy and when, at which price to sell, which contracts to take.
These are examples. If your problem isn't here, ask us.
What a model consists of
Every optimization model has the same three parts. Writing them down is often the most useful step of a project, because it forces everyone to agree on what the goal is and which rules really apply.
- Decisions
- What you are free to choose: quantities, assignments, sequences, dates.
- Objective
- What you want to achieve, as a number: lowest cost, highest margin, fewest late orders, or a stated weighting of several goals.
- Constraints
- What limits you: capacities, shift rules, contracts, delivery dates, regulations.
A small example
Several plants supply several customers. How much should each plant ship to each customer so that transport costs are lowest?
| Decision | Quantity shipped from plant i to customer j | xij ≥ 0 |
|---|---|---|
| Objective | Minimise total transport cost | min ΣiΣjcij xij |
| Constraint | No plant ships more than it can produce | Σjxij ≤ si |
| Constraint | Every customer receives what they ordered | Σixij = dj |
Real models have more rules and often yes-or-no decisions, such as whether to open a warehouse. The structure stays the same.
Why you can rely on the plan
- Hard rules are never broken
- A plan that violates a capacity limit or a legal rest period is not a plan. Such rules are constraints in the model and hold in every solution.
- If no plan exists, you learn why
- When the rules cannot all be met at once, the model reports which of them conflict, so that you can decide which one to relax.
- Results in your units
- Hours, pallets, francs and late orders. You compare the plan with your own, line by line.
- You can ask what if
- Change a capacity, a price or a deadline and see what it does to the plan and what it costs.
- We state how good the solution is
- For many problems the solver gives a bound: the plan is proven to be within a stated percentage of the best possible. Where no such proof is available, we say so.
Planning with uncertain numbers
A plan built on average demand fails whenever demand is not average. Our models use the whole forecast range, so the plan holds up in the likely cases and you know what the unlikely ones would cost.
Methods
Linear and mixed-integer programming, constraint programming, stochastic and robust optimization, and heuristics for problems too large to solve exactly.
Tell us about your problem
The first conversation is free. If we don't think we can help, we will say so.