CODEWITHSUNDEEP
montecarlobrightway
TB95
TB95

Reputation: 1

Montecarlo analysis in Brightway 2.5 using Beta-Pert distribution

This might be a beginner-type question:

I am trying to run a Monte Carlo analysis in Brightway 2.5 using the Beta-PERT distribution and ecoinvent 3.10 (cutoff). However, running even just 100 iterations takes a very long time. For instance, only 10 iterations take 7 minutes, while my goal is to run 10,000 iterations. I was able to pinpoint the issue: as long as my_mc.lci() is within the for loop, the process slows down significantly due to the repetitive recalculation of the inventory. However, as soon as I remove my_mc.lci() from the for loop, I keep getting the same score for every iteration. I've tried nearly every solution I could find online, but nothing seems to work. Does anyone have any advice? Thanks a lot! (I’m using Jupyter Notebook in Anaconda, Python v3.12.7)

#Presetting:

act2 = [x for x in eco_edit if 'carbon black production' in x ['name']
        ] [0]

heat_exchange= list(act2.technosphere())  [1]  
act2.heat_exchange = heat_exchange

#Set up beta-pert distribution since it is not available in stats_arrays
def beta_pert_random(min_val, mode_val, max_val, size=1):
    alpha = 1 + 4 * (mode_val - min_val) / (max_val - min_val)
    beta_param = 1 + 4 * (max_val - mode_val) / (max_val - min_val)
    return beta.rvs(alpha, beta_param, loc=min_val, scale=max_val - min_val, size=size)

#Defining Pert-parameters
min_heat = 0.1  # MJ
mode_heat = 1.5  # MJ (default)
max_heat = 10  # MJ
size = 1   

demand = {act2.id: 1}
     
list_of_methods= [AF, CC, EP, ...]  
CC_method = list_of_methods [1]     #Climate change

#Set up characterization matrix which is calculated once
def get_C_matrix(demand, list_of_methods):
    C_matrix = {}
    sacrificial_LCA = bc.LCA(demand)
    sacrificial_LCA.lci()
    
    CC_method = list_of_methods[1]
    sacrificial_LCA.switch_method(CC_method)
    C_matrix[CC_method] = sacrificial_LCA.characterization_matrix
    
    return C_matrix    


#Monte Carlo Simulation:
my_iterations = 100
mc_scores = np.empty(shape=[my_iterations])
my_C_matrix = get_C_matrix(demand, list_of_methods)[CC_method]

for iteration in range(my_iterations):
    random_heat = beta_pert_random(min_heat, mode_heat, max_heat, size)
    heat_exchange['amount'] = random_heat [0] 
    heat_exchange.save()  

    my_mc = bc.LCA(demand, method=CC_method)
    my_mc.lci()              
    my_mc.lcia()
    
    next(my_mc)

    mc_scores[iteration] = (my_C_matrix * my_mc.inventory).sum()

results = mc_scores.tolist()

Upvotes: 0

Views: 69

Answers (1)

mfastudillo
mfastudillo

Reputation: 1631

in each iteration you're modifying the sql database with your inventory when you do heat_exchange.save(). This is going to be mega slow. It is much better to use an interface with bw_processing. You can create one that in each iteration returns a value from your function and modify the precise value of your A matrix that you want to change.

An example here or here. You can combine that with the uncertainty coming from your background (ecoinvent) if you want.

Upvotes: 0

Related Questions