How to use M1 GPU or neural engine to decrease time taken for scipy.optimize.differential_evolution? - numpy

Question

Trying to use scipy.optimise.differential_evolution to optimise my code.
I set workers = -1 in my geneticalgo.py file to utilize all cores, and it always makes the CPU cores do all the work. Is there a way to force the GPU or neural engine to run the optimisation task instead? My last run where I got a result took just over 9 hours.
1st code block is the script I'm writing that needs to be optimised
2nd code block imports the code from the 1st and runs the differential_evolution on it.
Any answers appreciated!
# from decentralUncertainConstants import *
import pandas as pd
import numpy as np
from statistics import NormalDist
import random
import math
from scipy.stats import norm
import matplotlib.pyplot as plt;
demandyr0 = 0.0240232711111154
MDemLimit = 1.2042811301817
aTransParam = 49.1297731108936
bsharpParam = 0.20118446680713
annVol = 0.15
volyr0 = 0.5
volM = 0.5
volb = 0.7
scaleFactor = 1468000
BlueH2MarketShare = 0.4
SFMarketShare = 0.18
thou = 1000
mill = 1000000
daysyear = 365
cons1 = 8760
cons2 = 16.92
plantDesignCap = 190950;
plantsbuilt = 35
plantOperationalCap = 0.95;
CO2emissionRate = 1.06; # Kg of CO2 per Kg of H2
CO2captureRate = 8.60; # Kg of CO2 per Kg of H2
CO2emissionRatefeedstock = 0.28;
CO2emissionRatefuel = 1.18;
NatGasConsumption = 33411;
plantDesignCapBase = 190950;
plantDesignCapScaled = 190000;
lifetime = 25
workingHours = 8322
dailyprodrate = 190000;
yearlyprodrate = 65894.92
endcapreq = 65882.50;
singleModProdRate = 470.68
plantfixedoperationalcosts = 1.09
plantextensioncost = 0.54
discountrate = 0.1; # Nine percent per annum
th = 25
statetax = 0.0725;
fedtax = 0.21;
#Hprice stuff
Hpricelow = 8
Hpricehigh = 15
Hpriceave = 11.5
#H2 delivery cost stuff
H2dpave = 1.07
H2dplow = H2dpave*(1+0.3)
H2dohigh = H2dpave*(1-0.3)
#CO2 transport storage (ts) stuff
CO2low = 10
CO2high = 30
CO2ave = 15
#emissions
CO2emissionsCH4 = 0.185
CO2emissionrate = 15903257.28
CO2captureeff = 0.9
CO2captured = 14312931.55
annualCO2emissions = 1590325.73
#CO2 transport storage (ts) stuff
CO2low = 10
CO2high = 30
CO2ave = 15
#CO2 storage and compression (sc) stuff
CO2schigh = 50*(6000/500)**0.6
CO2sclow = CO2schigh*0.7
CO2scave = CO2schigh*1.3
#process specs
ngusage = 0.155797012;
ngusageannual = 85963552.88;
elecusage = 1.11;
industrialelec = 0.061
waterusage = 5.77;
processwater = 0.0024;
modsbuilt = 140
plantsbuilt = 35
modcapex = 1.57
setupcost = 3.24
capInvestment = 591730751/1000000
basecostFF = 67836966;
basecostWM = 104434;
basecostCC = 625712;
#process specs
ngusage = 0.155797012;
ngusageannual = 85963552.88;
elecusage = 1.11;
industrialelec = 0.061
waterusage = 5.77;
#conversion factors
btutokwh = 293.07;
ngtobtu = 58.36;
#CO2 gas prices constants
drift = 0.00234
volatility = 0.19
CO2captradeprice = 29.15
initialCapCostsFixed = 591.73
initialCapCostsPhased = 230.53
#Other Phased stuff
iniProdRateDay = 47500
iniProdRateYear = 16473.73
modsPerPlan = 4
plantDowntime = 0.15
iniMods = 35
expansionTimes = 3
expansionIncrement = 16470.63
expansionThresh = 0.75
reductionDuringExpansion = 0.5
othervariableoperating = 1800;
othervariableoperating2 = 5.32;
iniplantsbuilt = 9
#helpers to find CO2 prices
interval = 4
MACRSfixed = [22,43,40,37,34,31,29,27,26,26,26,26,26,26,26,26,26,26,26,26,13,0,0,0,0]
MACRSphased = [10,19,18,16,15,19,23,21,20,25,29,28,27,31,35,34,33,32,31,31,24,18,18,18,18]
MACRSflexible = [10,19,18,16,15,14,13,12,12,17,22,21,20,25,29,33,36,35,34,33,26,19,18,18,18]
q = [46.96, 50, 54.25, 58.86, 63.86, 69.29, 75.18, 81.57, 88.51, 96.03, 104.19, 113.05, 122.66, 133.08, 144.40, 156.67, 169.99, 184.44, 200.11, 217.12, 235.58, 255.60, 277.33, 300.90, 326.48]
nomyear = [1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25]
nomyear2 = [0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27]
nums = [0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24]
year = [2023, 2024, 2025, 2026, 2027, 2028, 2029, 2030, 2031, 2032, 2033, 2034, 2035, 2036, 2037, 2038, 2039, 2040, 2041, 2042, 2043, 2044, 2045, 2046, 2047, 2048, 2049, 2050]
ngprices = [130.48,109.52,118.95,225.84,207.50,177.11,286.63,308.64,455.36,352.65,365.23,464.26,206.46,228.99,209.60,144.10,195.45,228.99,137.29,132.05,156.68,165.06,134.14,106.37,203.84]
elecprices = [0.0727,0.0667,0.0681,0.0692,0.0688,0.0676,0.0691,0.0710,0.0689,0.0667,0.0682,0.0677,0.0683,0.0696,0.0639,0.0616,0.0573,0.0525,0.0511,0.0488,0.0505,0.0464,0.0443,0.0448,0.0453]
expansionChange = [0,0,0,1,0,0,0,1,0,0,0,1,0,0,0,0,0,0,0,0,0,0,0,0,0]
modulesbuilt = [0,0,0,0,35,0,0,0,35,0,0,0,35,0,0,0,0,0,0,0,0,0,0,0,0] #years 1-25
plantsbuilt = [9,9,9,9,18,18,18,18,27,27,27,27,35,35,35,35,35,35,35,35,35,35,35,35,35] #years 1-25
def test(help):
iniMods = help[0]
iniProdRateYear = help[1]
plantextensioncost = help[2]
endcapreq = help[3]
singleModProdRate = help[4]
def npv():
# demandCAL = pd.Series(index=nums, dtype='float64')
demandprojection = pd.Series(index=nomyear2, dtype='float64').array
normDemProjGrowth = pd.Series(index=nums, dtype='float64')
randomDraw = pd.Series(index=nums, dtype='float64')
realisedGrowth = pd.Series(index=nums, dtype='float64')
realisedNormalisedDemand = pd.Series(index=nums, dtype='float64')
realisedDemand = pd.Series(index=nums, dtype='float64')
demandSF = pd.Series(index=nums, dtype='float64')
availableRate = pd.Series(index=nums, dtype='float64')
H2AddedProdRate = pd.Series(index=nums, dtype='float64')
H2ReducedOutput = pd.Series(index=nums, dtype='float64')
dsbalance = pd.Series(index=nums, dtype='float64')
H2prodrate = pd.Series(index=nums, dtype='float64')
CO2emissions = pd.Series(index=nums, dtype='float64')
CO2capture = pd.Series(index=nums, dtype='float64')
modcapexrange = pd.Series(index=nums, dtype='float64')
plantcapexrange = pd.Series(index=nums, dtype='float64')
tcirange = pd.Series(index=nums, dtype='float64')
CO2prices = pd.Series(index=nums, dtype='float64')
CO2capture = pd.Series(index=nums, dtype='float64')
revenue = pd.Series(index=nums, dtype='float64')
fixedcost = pd.Series(index=nums, dtype='float64')
varcosts = pd.Series(index=nums, dtype='float64')
opcosts = pd.Series(index=nums, dtype='float64')
Depreciation = pd.Series(index=nums, dtype='float64')
plantsbuiltcounter = 0
sv = 0
cf = pd.Series(index=nums, dtype='float64')
dcf = pd.Series(index=nums, dtype='float64')
npv1 = 0
npv = 0
realisedDemandyr0 = (1-volyr0)*(demandyr0)+2*demandyr0*volyr0*random.uniform(0,1)
stoM = (1-volM)*MDemLimit+2*volM*MDemLimit*random.uniform(0,1)
stoa = stoM/realisedDemandyr0-1
stob = (1-volb)*bsharpParam+2*volb*bsharpParam*random.uniform(0,1)
for i in range(0,len(nomyear2)):
demandprojection[i] = stoM/(1+stoa*np.exp(-nomyear2[i]*stob))
for i in range(0,len(nomyear2)-1):
normDemProjGrowth[i] = (demandprojection[i+1]-demandprojection[i])/demandprojection[i]
for i in range(0,len(nomyear2)):
randomDraw[i] = NormalDist(mu=0, sigma=1).inv_cdf(random.uniform(0,1))
for i in range(0, len(nomyear2)-1):
realisedGrowth[i] = normDemProjGrowth[i]+randomDraw[i+1]*annVol
for i in range(0, len(nomyear2)-1):
realisedNormalisedDemand[i] = demandprojection[i+1]*(1+realisedGrowth[i])
for i in range(0, len(nomyear2)-1):
realisedDemand[i] = realisedNormalisedDemand[i]*(scaleFactor*BlueH2MarketShare*SFMarketShare)
demandSF = realisedDemand[0:th]
#Available Rate
for i in range(0, th-1):
availableRate[0] = iniProdRateYear
if expansionChange[i] == 1:
# availableRate[i+1] = expansionIncrement + availableRate[i]
availableRate[i+1] = (singleModProdRate*modulesbuilt[i+1]) + availableRate[i]
elif expansionChange[i] == 0:
availableRate[i+1] = availableRate[i]
for i in range(0,th):
dsbalance[i] = min(demandSF[i], endcapreq)-availableRate[i]
#Plants built
#Plant Output
for i in range(0,th):
H2prodrate[i] = min(demandSF[i], availableRate[i])
for i in range(0,th):
CO2emissions[i] = (ngusage*btutokwh*CO2emissionsCH4)*(1-CO2captureeff)*(H2prodrate[i]*thou)/thou
CO2capture[i] = (ngusage*btutokwh*CO2emissionsCH4)*(CO2captureeff)*(CO2emissions[i]*thou)/thou
#Hydrogen price
lowModeHighMode = (Hpriceave-Hpricelow)/(Hpricehigh-Hpricelow)
randomDraw2 = random.uniform(0, 1)
def findHprice():
if randomDraw2 < lowModeHighMode:
Hprice = Hpricelow+math.sqrt((Hpriceave-Hpricelow)*(Hpricehigh-Hpricelow)*randomDraw2)
else:
Hprice = Hpricehigh-math.sqrt((Hpricehigh-Hpricelow)*(Hpricehigh-Hpriceave)*(1-randomDraw2))
return Hprice
#Revenues
for i in range(0,th):
revenue[i] = findHprice()*(H2prodrate[i]*thou)/mill
#Total capital investment
modulecapex = modcapex * iniMods
plantcapex = iniplantsbuilt * setupcost
tci = modulecapex+plantcapex
for i in range(0,th):
modcapexrange[i] = modulesbuilt[i]*modcapex
for i in range(0,th):
if plantsbuilt[i] > plantsbuilt[i-1]:
plantcapexrange[i] = (plantsbuilt[i] - plantsbuilt[i-1])*setupcost
else:
plantcapexrange[i] = 0
for i in range(0, th):
tcirange[i] = plantcapexrange[i] + modcapexrange[i]
#Operational costs
#fixed costs
for i in range(0,th):
if nomyear[i] == 20:
fixedcost[i] = (plantfixedoperationalcosts+plantextensioncost)*plantsbuilt[i]
else:
fixedcost[i] = plantfixedoperationalcosts*plantsbuilt[i]
#CO2 prices for transport storage
lowModeHighMode3 = (CO2ave-CO2low)/(CO2high-CO2ave)
def priceCO2TS():
if randomDraw2 < lowModeHighMode3:
CO2ts = CO2low+math.sqrt((CO2ave-CO2low)*(CO2high-CO2low)*randomDraw2)
else:
CO2ts = CO2high-math.sqrt((CO2high-CO2low)*(CO2high-CO2low)*(1-randomDraw2))
return CO2ts
#CO2 prices for storage and compression
lowModeHighMode4 = (CO2scave-CO2sclow)/(CO2schigh-CO2scave)
def priceCO2SC():
if randomDraw2 < lowModeHighMode4:
CO2sc = CO2sclow+math.sqrt((CO2scave-CO2sclow)*(CO2schigh-CO2sclow)*randomDraw2)
else:
CO2sc = CO2schigh-math.sqrt((CO2schigh-CO2sclow)*(CO2schigh-CO2sclow)*(1-randomDraw2))
return CO2sc
#natural gas price
result = [];
for i in range(1000): #1000
result.append(random.choices(ngprices, weights=None, cum_weights=None, k = 25)) #k=25
samplemean = [];
for i in result:
samplemean.append(np.mean(i))
totalmean = [];
totalmean.append(np.mean(samplemean))
totalst = []
totalst.append(np.std(samplemean))
ngprice = norm.ppf(random.uniform(0,1), totalmean, totalst)
#Electricity price
result = [];
for i in range(1000): #1000
result.append(random.choices(elecprices, weights=None, cum_weights=None, k = 25)) #k=25
samplemean = [];
for i in result:
samplemean.append(np.mean(i))
totalmean = [];
totalmean.append(np.mean(samplemean))
totalst = []
totalst.append(np.std(samplemean))
elecprice = norm.ppf(random.uniform(0,1), totalmean, totalst)
#finding CO2 price
result2 = []
for i in range(111): #108
result2.append(drift+(norm.ppf(random.uniform(0,1), 0, 1))*volatility)
CO2tradeprices = []
CO2tradeprices.append(CO2captradeprice)
for i, v in enumerate(result2):
CO2tradeprices.append(CO2tradeprices[i]*(1+result2[i]))
result3 = [] #array used to get average price from 4 values from each year
for i in range(11, 111):
result3.append(CO2tradeprices[i])
for i in range(0, th):
CO2prices[i] = np.mean(result3[i:i+interval])
#variable costs
ng = pd.Series(index=nums, dtype='float64'); #natural gas
elec = pd.Series(index=nums, dtype='float64'); #electricity
ts = pd.Series(index=nums, dtype='float64'); #transport and storage
pw = pd.Series(index=nums, dtype='float64'); #process water
tax = pd.Series(index=nums, dtype='float64'); #CO2 tax
cc = pd.Series(index=nums, dtype='float64'); #CO2 capture and compression
ovoc = pd.Series(index=nums, dtype='float64'); #other variable operating costs
for i in range(0,th):
ng[i] = ngusage/ngtobtu*ngprice[0]*(H2prodrate[i]*thou)/mill
elec[i] = elecusage*(H2prodrate[i]*thou)*elecprice[0]/mill
pw[i] = waterusage*processwater*(H2prodrate[i]*thou)/mill
ovoc[i] = (othervariableoperating*othervariableoperating2)*plantsbuilt[i]/mill
for i in range(0,th):
ts[i] = CO2capture[i]*priceCO2TS()/mill
cc[i] = priceCO2SC()*CO2capture[i]/mill
tax[i] = CO2prices[i]*CO2emissions[i]/mill
for i in range(0,th):
varcosts[i] = ng[i] + elec[i] + pw[i] + ovoc[i] + ts[i] + cc[i] + tax[i]
for i in range(0,th):
opcosts[i] = fixedcost[i] + varcosts[i]
#Depreciation
for i in range(0, th):
Depreciation[i] = MACRSphased[i]*(statetax+fedtax);
sv = tci - sum(Depreciation)
decom = pd.Series(index=nums, dtype='float64')
for i in range(0, th):
if i <=24:
decom[i] = 0
else:
decom[i] = sv
#Cashflow
for i in range(0, th):
cf[i] = (revenue[i]-opcosts[i])*(1-statetax-fedtax)+Depreciation[i]
dcf[i] = (cf[i]-tcirange[i])/(1+discountrate)**nomyear[i]
npv = sum(dcf[:-5]) - tci
return -npv
counter = 0
for i in range(0, 2000): #2000 runs
counter += npv()
enpv = counter/2000 #2000 runs
return enpv
print(test([35, 16473.73, 0.54, 65882.50, 470.68]))
from scipy.optimize import differential_evolution
from decentralised.decentralPhased import *
if __name__ == "__main__":
... # Prepare all the arguments
# result = scipy.optimize.differential_evolution(minimize_me, bounds=function_bounds, args=extraargs,
# disp=True, polish=False, updating='deferred', workers=-1)
result = differential_evolution(test, bounds, workers=-1)
print("Optimal solution: ", result.x)
print("Objective function value: ", result.fun)
In 2nd code block, I set workers = -1 to utilise all cores but the scipy always defaults to the CPU cores which is only 8 on my machine (base model M1 mac mini). Is there a way to force the process to run on the apple silicon GPU or neural engine?