Package org.matsim.core.replanning

Packages
package	annealing
package	choosers
package	conflicts
package	inheritance
package	modules
package	selectors
package	strategies

Classes
interface	GenericPlanStrategy
class	GenericPlanStrategyImpl
interface	GenericStrategyManager
class	GenericStrategyManagerImpl
interface	PlanStrategy
class	PlanStrategyImpl
interface	ReplanningContext
class	ReplanningContextImpl
class	ReplanningUtils
class	StrategyManager
class	StrategyManagerModule

Detailed Description

The StrategyManager controls the replanning of all agents in a population.

Various strategies can be created and added to the org.matsim.core.replanning.StrategyManager. The StrategyManager selects for each agent randomly one of the strategies. When adding strategies to the strategy manager, a weight must be provided which influences the probability the strategy gets chosen for a person. The probability for a strategy to be chosen is defined as the strategy's weight divided by the sum of all strategies' weights.

A org.matsim.core.replanning.PlanStrategyImpl always consists of exactly one PlanSelector and zero or more StrategyModules. A PlanSelector selects one of the agents' existing plans, while a StrategyModule modifies a plan. A StrategyModule can change the route information, time information, change the sequence of activities, change locations of activities and so on.

After a plan is selected, the strategy makes a copy of the plan before it is modified by a strategy module. If a strategy contains no strategy module, the plan will not be duplicated. If a strategy contains multiple strategy modules, the plan will be modified by one strategy module after the other, in the order the strategy modules were added to the strategy.

Example

The strategy manager contains the following 5 strategies:

1. the time information of the currently selected plan is modified
2. the route information of a randomly selected plan is modified
3. first, the time information of a randomly selected plan is modified, and afterwards new routes are assigned according to the new times
4. the plan with the best score is selected for the agent, but not modified
5. a random plan is selected from the agent, but plans with higher score have a higher probability to be selected

This could be visualized as follows, including the probabilities each strategy is chosen:

StrategyManager
    |
    +--- 5% -- KeepSelected
    |          |_ TimeAllocationMutator
    |
    +--- 7% -- RandomPlanSelector
    |          |_ ReRoute
    |
    +--- 8% -- RandomPlanSelector
    |          |_ TimeAllocationMutator
    |          |_ ReRoute
    |
    +-- 75% -- BestScoreSelector
    |
    +--- 5% -- ExpBetaSelector

And here is the corresponding code to setup such a StrategyManager:

    StrategyManager manager = new StrategyManager();

    // strategy1
    PlanStrategy strategy1 = new PlanStrategy(new KeepSelected());
    strategy1.addStrategyModule(new TimeAllocationMutator());

    // strategy2
    PlanStrategy strategy2 = new PlanStrategy(new RandomPlanSelector());
    strategy2.addStrategyModule(new ReRoute());

    // strategy3
    PlanStrategy strategy3 = new PlanStrategy(new RandomPlanSelector());
    strategy3.addStrategyModule(new TimeAllocationMutator());
    strategy3.addStrategyModule(new ReRoute());

    // strategy4
    PlanStrategy strategy4 = new PlanStrategy(new BestScoreSelector());

    // strategy5
    PlanStrategy strategy5 = new PlanStrategy(new ExpBetaSelector());

    // add the strategies to the manager
    manager.addStrategy(strategy1, 0.05);
    manager.addStrategy(strategy2, 0.07);
    manager.addStrategy(strategy3, 0.08);
    manager.addStrategy(strategy4, 0.75);
    manager.addStrategy(strategy5, 0.05);

    // run it
    // Plans plans = new Plans();
    manager.run(plans);