SOFTWARE
SOPS: Stochastic Optimisation in Policy space
(a user-friendly alternative to stochastic dynamic programming)
A time limited trial version of the program can now be downloaded, see below.
SOPS is a user-friendly program for optimisation in stochastic, dynamic models. The underlying method is Stochastic Optimisation in Policy Space. As such it represents an alternative to stochastic dynamic programming: rather than simplifying the model structure in order to find solutions, the program allows for complex models and if necessary the use of simplified policies. The program is particularly useful for infinite horizon problems like investment problems, inventory management, renewable resource management etc. The program may also be used to address more complex problems involving measurement error and model uncertainty, and of course for simpler tasks like non-linear optimisation in static problems or model parameter estimation.
Models are formulated in the user-friendly and widely used simulation program Powersim Studio. Only releases of Studio after 2005 contains the special policy functions needed to do optimisation with SOPS. Policies can be formulated either by the use of built-in flexible grid functions or customised, algebraic policy functions. When the model has been formulated and tested in Studio, one command is sufficient to be transferred to the SOPS program. In this program one can make quick changes in model parameter assumptions, settings for the optimisation, initial assumptions about policies etc. The results section shows ranked criterion values and policy parameters for repeated policy searches as well as standard deviations to judge optimisation accuracy.
A pilot version of SOPS has been used in the following studies.
Moxnes, E. (2003). "Uncertain measurements of renewable resources: Approximations, harvest policies, and value of accuracy." Journal of Environmental Economics and Management 45(1): 85-108.
Measurement error (h) is more important than environmental variation for the optimal harvesting strategy (y=fish stock)
Moxnes, Erling (2003). Policy Sensitivity Analysis: simple versus complex fishery models. WPSD 3/03. Information Science, University of Bergen. (a revised version is published in System Dynamics Review in 2005, Vol 21, No 2).
An introduction to the use of SOPS can be found in the following publication:
Arne Krakenes and Erling Moxnes (2005). SOPS - A Tool to Find Optimal Policies in Stochastic Dynamic Systems. WPSD 2/05. System Dynamics Group, University of Bergen. (presented at the International System Dynamics Conference 2005, Boston)
After the trial period (summer 2006) you will find information about how to buy SOPS on this web page.
To download a time-limited beta version of the new version of SOPS, click SOPS. If you encounter problems please report these to Erling.Moxnes 'at' ifi.uib.no.
Note that to convert Powersim Studio models to SOPS you must install a C++ compiler on your machine. When converting from Studio to SOPS you must choose between two compilers CL (Microsoft) and GCC (GNU). The latter can be downloaded for free from http://www.bloodshed.net/devcpp.html (choose version 4, version 5 does not work in this case).
Some users may also need to install Microsoft's .NET program to run SOPS. Instructions for free downloading are available here.
To download the Powersim Studio fishery model used in the introduction, click Fishery.
If you do not currently have access to Powersim Studio 2005, you may download the ready made fishery model files that SOPS use. This allows you to test SOPS while you can make no changes to the original model structure (only parameter assumptions can be changed in SOPS). To download, click both Model and Settings.
A few general advice: Use the "eclectic" search method when you are much in doubt about what the optimal policy looks like or the problem is complex. The "gradient method" is usually the faster one but is more likely to fail. For both methods it is important to specify Variation for all policies as well as Iteration accuracy! To keep the files from each other: Studio models have the post script .sip, SOPS models .csim, and SOPS settings .sops.
Then some advice regarding the testing of the downloadable fishery model:
Besides changing whatever you want to, we suggest that you test the following two versions of the model:
1. In the Assumptions window set alfa=0, p0=8 and p1=0, and in the Custom Policy window unmark the policy parameter e0 (it will keep its fixed value of 0.5). With these parameters unit costs will be fixed and the harvest will not influence the fish price. The exact analytical solution is the so-called "target-escapement-policy" which implies that the harvest reduces the stock to the same level each year.
2. In the Assumptions window set alfa=1, p0=10 and p1=2, and in the Custom Policy window mark the policy parameter e0. Now costs will increase with the harvest rate and the fish price will decline. Intuitively this should make it more profitable to harvest more at low stock levels and less at higher levels. In this case SOPS will also find the optimal fishing capacity.
To visualise the harvesting strategies copy the data from the Optimization window to a spreadsheet.
Good luck with your testing!
Some key words to help you find this page:
optimization stochastic infinite horizon continuous discretized discretised stochastic programming complex systems nonlinear search nonlinear optimization nonlinear optimisation stochastic models deterministic models stochastic optimization algorithm dynamic models system dynamics nonlinear models stochastic dynamic optimization stochastic dynamic optimisation stochastic optimization markov chains markov process inventory control renewable resources management multi species measurement error assessment error optimal policy system dynamics stochastic dynamic programming optimal strategy practical strategy practical policy optimal investment strategy optimal investment strategies maximum principle user friendly program practical software user friendly optimisaton program user friendly optimization program practical optimisation program practical optimisaton program practical optimization program reinforcement learning curse of dimensionality numerous states genetic algorithm differential evolution simultaneous perturbation neuro-dynamic programming optimal control servo mechanisms state space stochastic system dynamics optimization in policy space stochastic dynamic programming stochastic dynamic optimization continuous and discrete models computer program computer software optimisation software optimization software optimizatioin program user friendly measurement error model uncertainty assessment error optimal policy optimal strategy response surface methods response surface method
