"# BarrOpt" %Barrier Optimization (BarrOpt.m) cookbook %Sam Roy, UMaine Mitchell Center, 3/13/20 %Place this older in your C:\ drive %This script contains all commands that are required to prepare and run a %simulation in BarrOpt.m, a Matlab program that accesses Matlab's %multi-objective optimization functions to identify pareto efficient %decisioin scenarios. These decision scenarios are presented as a series, each with %lists of decisions for multiple dams and culverts. The basic code is: %0 = keep dam and culvert as they are, %1 = remove dam, or replace culvert with upgraded safety/eco standards

%first, define the objective variables you want to optimize for (vars) and %the study region. For a 3D trade-off assesment we use river herring habitat (herring), road safety %risk (risk), and economic cost of decisions (cost). There are other %variables available, but these will do for now. For study region, 'All' %includes the entire region in Maine. fprintf('defining variables and study region\n') vars = {'herring','risk','cost'}; StudyRegion = 'All';

%Second, we need to retrieve and preprocess the culvert/dam data. %Specifically we need to access an index file (BarrIndex) for dams and culverts in %order to attribute variable values to the correct feature, but the index %is also important for ordering the dams and culverts based on their %position in river/stream networks. We also need to retrieve the variable values %for each culvert/dam (v). %But first, we need to direct the program to the directory where these data %are located. This WILL need to be altered to find where these files are on %your computer: fprintf('Preparing data for GA\n') data_directory = 'C:\BarrOpt_software'; [BarrIndex, v] = BarrOpt_prep(StudyRegion, data_directory, 'crh1',vars); %The 'crh1' code just means that we're only selecting existing dams/culverts

%Third, we need to prescribe an initial 'population' for the genetic %algorithm to start with. The genetic algorithm is designed based on a very %general interpretation of how evolution works. It will come up with %different 'populations', where each population offers a different decision %for each dam/culvert in the study region. The model iterates through %several populations and keeps members of the population (representing %decision scenarios) that are more 'fit' than others, and therefore offer %more efficiency when trading off between habitat, road safety, and cost. %But, you can help the algorithm out by giving it starting populations, and %the best way to do this (I have found by lots of trial and error) is to %give it 'extreme' cases to start with, and it will eventually fill in the %gradient between these. The extreme cases I use are: 'keep all dams and %culverts as they are' (all decisions = 0) and 'remove all dams and replace %all culverts' (all decisions = 1). Here is how to define these %populations: fprintf('initializing start population\n') if size(BarrIndex,2)==2 total_number_barrs = length(BarrIndex(BarrIndex(:,1)<1e9,1)); else total_number_barrs = length(BarrIndex(BarrIndex(:,1)<1e9,1))-1; end xx=zeros(2,total_number_barrs); %this is 'remove/replace all' xx(1,:) = 1; %this is 'keep all'

%Fourth, we want to prescribe a population size (pop) for the model. Larger %populations are better at finding efficient solutions, but if the %population is too big it will take to long to generate a solution/crash. %So, it's dependent on your processor speed, ram, etc. so we can fiddle %with this number. 2000 tends to be ideal, but we can start smaller: pop = 100;

%Fourth, we can now run the genetic algorithm (GA)! We do so using a 'shell' %program that feeds our data into the GA and calculates the 'fitness' of %each population, or in other words, the relative efficiency of each %scenario. %There are a lot of outputs from this (variables left of the %equal sign) but we really only care about x and f. %FYI, this simulation could take up to an hour, even more for a larger %population. fprintf('run 1...\n') [x, f, xu, fu, exitflag, output, population, score] = BarrOpt(BarrIndex,pop,xx,v); fprintf('complete\n')

%Fifth, I find that rerunning GA several times leads to more efficient results. %Sometimes this can lead to sudden improvements in efficiency. I run %BarrOpt again using the output population (x) as input. If run times %are long, take some reruns out. %Keep this commented out until confident about GA use %fprintf('run 2...\n') %[x, f, xu, fu, exitflag, output, population, score] = BarrOpt(BarrIndex,pop,x,v); %fprintf('complete\n') %fprintf('run 3...\n') %[x, f, xu, fu, exitflag, output, population, score] = BarrOpt(BarrIndex,pop,x,v); %fprintf('complete\n')

%Sixth, visualize the results. Because we have three objective variables, %we will genreate a 3D surface plot of points defining a production %possibility frontier 'surface' fprintf('plotting frontier\n') plot3(f(:,1),f(:,2),f(:,3),'.k')

%now save your results, using an easily identifiable name and date of %simulation. It's very useful to specify the objective variables you have %used and the population: fprintf('saving run\n') t=datestr(datetime('now'),'dd-mmm-yyyy_HH-MM-SS'); save(sprintf('%s\BarrOut\Scenarios_%s_herring_risk_cost_pop100_%s.mat',data_directory,StudyRegion,t))