Differentiation of "optimization"
Posted: Mon Jan 08, 2024 10:34 am
Hi there,
here's just a quick differentiation of how to understand "optimization"... Often this term needs to be aligned between parties dealing with our APIs...
Bernd
here's just a quick differentiation of how to understand "optimization"... Often this term needs to be aligned between parties dealing with our APIs...
- Best route: this simply means to determine the best geometry between 2 or more waypoints without changing the sequence of the waypoints. The target function may be based on abstract "costs" (xRoute1, check the term MALUS) or monetary costs (derived by direct costs per distance, per time, per fuel and indirect costs such as toll based on vehicle dimensions)
- Optimal routes: green= shortest, blue= fastest (car based)
- Reconstruction of a historical route: While most "target function" based optimizations refer to an output that is a "recommendation for a future activity" let me add this generic "how has this been done in the past?" approach. MapMatching means to reconstruct a route that already has been driven in the past. So here the input is usually a dense sequence of GPS coordinates where a vehicle has been. In this case the purpose of the output is not the geometry itself but the derived KPIs such as "toll on a route" or "emission that was emitted".
- Matching of a historical route
- Best sequence: this means that to determine the optimal sequence of a set of waypoints from the perspective of a single vehicle. The target function which defines the "optimum" may be based on "minimum total distance" or "minimum toal driving time" or further aspects (such as minimum costs). On top of the target function there's also the view to "constraints" (vehicle capacities, driver working hours, waypoint opening times...) which may exclude some "sequences" as not valid.
- Optimal sequence of 10 stops based on single vehicle (airline, no time windows or capas)
- Best (operational) tour optimization: in this case there's not just one vehicle in the scope of the function but a fleet. So the challenge for the algorithm is not only to determine sequences but also to decide about "which ressource is supposed to take care of which order/waypoints". This is the highest complexity of optimization. This also deals with target function (highest revenue, lowest expenses) and constraints. Compared to "best sequence" there's also the new constraint level based on "skills" (aka equipment).
- Optimal tours (assignment) of 3 vehicles dealing with 20 stops
- Clustering: Strategic area optimization: In this case the target of the optimization is to assign locations (screenshot: balls) to area centers (screenshot: pyramids) in a way that the assigned workload per area considers some kind of a balancing of KPIs (upper chart: same bar size)
- Area optimization: a "balls" size reflects the locations activity. A perfectly balanced area output (standard case) is given if all areas have an equal sum of assigned activity. (Must not be "equal number of assigned locations by the way)
- Clustering: Multi weeks planning: Now this case tries to assign periodical visits to given days in a so-called plalling cycle. Target is (mofre or less) to have an equal workload on each day of the cycle. Can be used on short cycles (weekly, biweekly) and also on a long term perspective (24 weeks or even 48 weeks).
- Biweekly planning of a single technicians cycle.
Bernd