Description: This package implements a state-of-the-art version of the Particle Swarm Optimisation (PSO) algorithm, with a special focus on the calibration of environmental models. hydroPSO is model-independent, allowing the user to easily interface any model code with the calibration engine (PSO). hydroPSO includes a series of controlling options and PSO variants to fine-tune the performance of the calibration engine. An advanced sensitivity analysis function together with user-friendly plotting summaries facilitate the interpretation and assessment of the calibration results. Bugs reports/comments/questions are very welcomed.
Description: This package implements a state-of-the-art version of the Particle Swarm Optimisation (PSO) algorithm (SPSO-2011 and SPSO-2007 capable), with a special focus on the calibration of environmental models. hydroPSO is model-independent, allowing the user to easily interface any model code with the calibration engine (PSO). It includes a series of controlling options and PSO variants to fine-tune the performance of the calibration engine to different calibration problems. An advanced sensitivity analysis function together with user-friendly plotting summaries facilitate the interpretation and assessment of the calibration results. Bugs reports/comments/questions are very welcomed.
Particle Swarm Optimisation algorithm to calibrate environmental models. It includes a series of controlling options and PSO variants to improve the performance of the algorithm and customize it to different calibration problems
Particle Swarm Optimisation algorithm to calibrate environmental models. It implements a state-of-the-art Particle Swarm Optimisation (PSO) algorithm (SPSO-2011 and SPSO-2007 capable), with several fine-tuning options and PSO variants available to customise the PSO engine to different calibration problems.
@@ -35,14 +35,15 @@ further arguments to be passed to \code{fn}.
...
@@ -35,14 +35,15 @@ further arguments to be passed to \code{fn}.
}
}
\item{method}{
\item{method}{
character, variant of the PSO algorithm to be used. Valid values are in \code{c('spso2011', 'spso2007', 'ipso', 'fips', 'wfips')}: \cr
character, variant of the PSO algorithm to be used. Valid values are in \code{c('spso2011', 'spso2007', 'ipso', 'fips', 'wfips')}: \cr
f
\kbd{pso}: at each iteration particles are attracted to its own best-known personal and to the best-known global position. Each particle is connected to a neighbourhood of particles depending on the \code{topology} value \cr
\kbd{spso2011}: At each iteration particles are attracted to its own best-known \sQuote{personal} and to the best-known position in its \sQuote{local} neighbourhood, which depens on the value of \code{topology}. In addition, values of the PSO engine are set to the values defined in the Standard PSO 2011 (SPSO 2011, see Clerc 2012) \cr
\kbd{spso2007}: As in \code{method='spso2011'}, but with values of the PSO engine set to the values defined in the Standard PSO 2007 (SPSO 2007, see Clerc 2012)
\kbd{ipso}: at each iteration particles in the swarm are rearranged in descending order according to their goodness-of-fit and the best \code{ngbest} particles are used to modify particles' position and velocity (see Zhao, 2006). Each particle is connected to a neighbourhood of particles depending on the \code{topology} value \cr
\kbd{ipso}: at each iteration particles in the swarm are rearranged in descending order according to their goodness-of-fit and the best \code{ngbest} particles are used to modify particles' position and velocity (see Zhao, 2006). Each particle is connected to a neighbourhood of particles depending on the \code{topology} value \cr
\kbd{fips}: at each iteration ALL particles contribute to modify the particles' position and velocity (see Mendes et al., 2004). Each particle is connected to a neighbourhood of particles depending on the \code{topology} value \cr
\kbd{fips}: at each iteration ALL particles contribute to modify the particles' position and velocity (see Mendes et al., 2004). Each particle is connected to a neighbourhood of particles depending on the \code{topology} value \cr
\kbd{wfips}: same implementation as \kbd{fips} method, but the contribution of each particle is weighted according to their goodness-of-fit value (see Mendes et al., 2004) \cr
\kbd{wfips}: same implementation as \kbd{fips} method, but the contribution of each particle is weighted according to their goodness-of-fit value (see Mendes et al., 2004) \cr
By default \code{method=pso}
By default \code{method=pso}
\kbd{spso2007}: As in \code{method='pso'}, but with values of the PSO engine set to the values defined in the Standard PSO 2007 (SPSO 2007, see Clerc 2011) \cr
\kbd{spso2011}: As in \code{method='pso'}, but with values of the PSO engine set to the values defined in the Standard PSO 2011 (SPSO 2011, see Clerc 2011)
}
}
\item{lower}{
\item{lower}{
numeric, lower boundary for each parameter \cr
numeric, lower boundary for each parameter \cr
...
@@ -149,12 +150,12 @@ By default \code{random.update=TRUE}
...
@@ -149,12 +150,12 @@ By default \code{random.update=TRUE}
}
}
\item{boundary.wall}{
\item{boundary.wall}{
character, indicates the type of boundary condition to be applied during optimisation. Valid values are in \code{c('absorbing', 'reflecting', 'damping', 'invisible')} \cr
character, indicates the type of boundary condition to be applied during optimisation. Valid values are in \code{c('absorbing', 'reflecting', 'damping', 'invisible')} \cr
By default \code{boundary.wall='absorbing'}
By default \code{boundary.wall='absorbing'}\cr
Experience has shown that Clerc's constriction factor and the inertia weights do not always confine the particles within the solution space. To address this problem, Robinson and Rahmat-Samii (2004) and Huang and Mohan (2005) propose different boundary conditions, namely, \kbd{reflecting}, \kbd{damping}, \kbd{absorbing} and \kbd{invisible} to define how particles are treated when reaching the boundary of the searching space (see Robinson and Rahmat-Samii (2004) and Huang and Mohan (2005) for further details)
Experience has shown that Clerc's constriction factor and the inertia weights do not always confine the particles within the solution space. To address this problem, Robinson and Rahmat-Samii (2004) and Huang and Mohan (2005) propose different boundary conditions, namely, \kbd{reflecting}, \kbd{damping}, \kbd{absorbing} and \kbd{invisible} to define how particles are treated when reaching the boundary of the searching space (see Robinson and Rahmat-Samii (2004) and Huang and Mohan (2005) for further details)
}
}
\item{topology}{
\item{topology}{
character, indicates the neighbourhood topology used in hydroPSO. Valid values are in \code{c('gbest', 'lbest', 'vonNeumann', 'random')}: \cr
character, indicates the neighbourhood topology used in hydroPSO. Valid values are in \code{c('random', 'gbest', 'lbest', 'vonNeumann')}: \cr
\kbd{gbest}: every particle is connected to each other and, hence the global best influences all particles in the swarm. This is also termed \samp{star} topology, and it is generally assumed to have a fast convergence but is more vulnerable to the attraction to sub-optimal solutions (see Kennedy, 1999; Kennedy and Mendes, 2002, Schor et al., 2010) \cr
\kbd{gbest}: every particle is connected to each other and, hence the global best influences all particles in the swarm. This is also termed \samp{star} topology, and it is generally assumed to have a fast convergence but is more vulnerable to the attraction to sub-optimal solutions (see Kennedy, 1999; Kennedy and Mendes, 2002, Schor et al., 2010) \cr
\kbd{lbest}: each particle is connected to its \code{K} immediate neighbours only. This is also termed \samp{circles} or \samp{ring} topology, and generally the swarm will converge slower than the \kbd{gbest} topology but it is less vulnerable to sub-optimal solutions (see Kennedy, 1999; Kennedy and Mendes, 2002) \cr
\kbd{lbest}: each particle is connected to its \code{K} immediate neighbours only. This is also termed \samp{circles} or \samp{ring} topology, and generally the swarm will converge slower than the \kbd{gbest} topology but it is less vulnerable to sub-optimal solutions (see Kennedy, 1999; Kennedy and Mendes, 2002) \cr
\kbd{vonNeumann}: each particle is connected to its \code{K=4} immediate neighbours only. This topology is more densely connected than \samp{lbest} but less densely than \samp{gbest}, thus, showing some parallelism with \samp{lbest} but benefiting from a bigger neighbourhood (see Kennedy and Mendes, 2003) \cr
\kbd{vonNeumann}: each particle is connected to its \code{K=4} immediate neighbours only. This topology is more densely connected than \samp{lbest} but less densely than \samp{gbest}, thus, showing some parallelism with \samp{lbest} but benefiting from a bigger neighbourhood (see Kennedy and Mendes, 2003) \cr
