From ILMS-Wiki
Jump to: navigation, search
Package: jams.components.optimizer
Autor: Christian Fischer

This component performs a model calibration. It finds optimal parameter values by using the SCEM (Shuffle Complex Evolution) algorithm for optimization. (Vrugt et al., 2003) The SCEM component is similar to the SCE optimization component. The difference is, that SCE performs a single criterion optimization and SCEM a multiple criterion optimization. In general the solution of a multiple criterion optimization problem is not a single set of parameters, but infinitely many sets of parameters. This is called Pareto Front, which SCEM tries to approximate. For applications, the calibration - model is inserted into the context of the SCEM component. After that the parameters, their feasible range, and the optimization criterions are specified. It is possible to maximize and minimize the criterions during optimization. The process is influenzed by the number of complexes and the number of individuals in a population. A higher number of complexes and individuals increases the chance of finding optimal points, but increases computing time also. In most cases the number of complexes lies between 2 and 10 and the population-size between 5 and 50.


<contextcomponent class="jams.components.optimizer.SCEM.SCEM" name="SCEM">
           <var name="dirName" context="J2000g" attribute="workspaceDir"/>
           <var name="parameterIDs" value="Snow.ddf;Snow.t_thres;InitSoilWater.FCAdaptation;SoilWater.latVertDist;GroundWater.k"/>
           <var name="boundaries" value="[1.0<10.0];[-10.0<10.0];[0.0<10.0];[0.0<10.0];[1.0<10.0];"/>
           <var name="effMethodName" value="e2,e1,vol"/>
           <var name="effValue" context="J2000g" attribute="e2,e1,mse"/>
           <var name="MaximizeEff" value="1,1,3"/>
           <var name="NumberOfComplexes" value="5"/>
           <var name="Population" value="11"/>
           <var name="sampleCount" value="10000"/>

           [..] insert model here [..]
dirName name of working directory
parameterIDs name of parameters for optimization
boundaries feasible range of every parameter
effMethodName names of goodness of fit criterions
effValue modelvariables of goodness of fit criterions
MaximizeEff optimization mode, 1 = maximization, 4 = minimization
NumberOfComplexes number of complexes
Population number of individuals in population
sampleCount maximal number of function evaluations


├─<SCEM Context>
│      └─<Catchment Context/>
│              ├─<InitCatchmentContext/>
│              ├─<InitHRUContext/>
│              ├─<TemporalContext</>
│              │     ├─<HRUContext/>
│              │     └─<ReachContext/>
│              └─<EntityOutputContext>


Personal tools