We really care about optimization. You may have seen us discussing the fundamentals. So it's no surprise that we couldn't wait to share the performance enhancements in this release. Hybrid optimization? Available on Nextmv Cloud? Yes please.
Make every decision more efficient with our hybrid solver that combines the power of decision diagrams and heuristics on Nextmv Cloud. The pairing of these techniques blends the best of both approaches so customers can find better routes for their fleet of vehicles faster.
A quick ALNS primer
ALNS is a heuristic solving technique that takes a feasible solution (like one from our decision diagram) and then adapts the use of operators to find better solutions in the neighborhood. ALNS uses a dynamic weighting system to select the most efficient operators, making it easy to add domain knowledge for a more informed search.
Learn more about how ALNS works and enjoy this abstract visual representation of ALNS in action.
How does hybrid optimization impact performance?
The introduction of ALNS as part of a hybrid solver to Nextmv Cloud means immediate performance benefits for all customers. Let’s have a look by diving into an example that consists of a fleet of 15 vehicles with 59 stops and includes delivery precedence and time windows.
Find more improving solutions faster
Ok, so what does “more improving solutions faster” mean?
Let’s take a look at what the result charts show us. We’ll do a direct comparison of the output with and without hybrid optimization. On the x-axis we see time in milliseconds and on the y-axis we see total value measured in seconds. Each point on the chart represents a feasible solution found.
More solutions returned faster: The total number of solutions found is typically much higher with hybrid optimization (than with a single solving paradigm). In our example, the hybrid solver found 25 more solutions in the given runtime of 3 seconds.
Improving solutions returned faster: It typically takes less time to find the next best solution with hybrid optimization (than with a single solving paradigm). In our example, the hybrid solver found a better solution in the first 175 ms.
What makes a solution better? In the context of this routing problem, we want to minimize time spent on the road (as well as other penalties that discourage behaviors like lateness) which are all represented in the solution value. Solution values are numerical representations of the states a solver explores as part of an optimization problem. Therefore, in our example, the solver interprets lower solution values as better.
Find more diverse solutions
The hybrid solver also allows for the discovery of more diverse solutions by employing random and parallel searches. This allows the solver to reach different areas of the search space more efficiently. The outcome: more heterogeneous results. For example, this diversity might surface in the vehicles utilized in each solution, as seen in the charts below.
The first chart, representing the composition of value by vehicle without hybrid optimization, shows solutions that utilize the same 10 vehicles. In the second chart, we see a more diverse set of vehicles being utilized in each of the returned solutions with hybrid optimization — as the hybrid approach inherently allows for the exploration of more diverse solutions in order to find the next best one.
Get started with Nextmv Cloud
Have questions? We’re all ears.