clean and update HEMS code

This commit is contained in:
Sadam93 2024-05-31 11:22:07 -04:00
parent 7cc3fd335c
commit 975e27741d

122
HEMS.py
View File

@ -7,15 +7,9 @@ import numpy as np
import pandas as pd
import pylab as pl
import pyomo.environ as py
# NOTE: model
from matplotlib import pyplot as plt, gridspec
from matplotlib.pyplot import legend
from pyomo.core import value, RangeSet
from matplotlib import pyplot as plt
# data = pd.ExcelFile('input_new_cluster.xlsx') #it will read the excel one time you do not need to read again and gain
# Parse the different tab
sheet_data = pd.read_excel("input_new_cluster.xlsx", sheet_name='data')
sheet_EV = pd.read_excel("input_new_cluster.xlsx", sheet_name='EV')
sheet_EWH = pd.read_excel("input_new_cluster.xlsx", sheet_name='EWH')
@ -28,9 +22,9 @@ sheet_ambient = pd.read_excel("input_new_cluster.xlsx", sheet_name='Ambient_Temp
sheet_load = pd.read_excel("input_new_cluster.xlsx", sheet_name='Base_Load')
# </editor-fold>
# </editor-fold>
model = py.ConcreteModel()
# <editor-fold desc="Reading the data from Excel file"> # i think it should be outside the loop
# <editor-fold desc="Sets: time intervals">
model.h = py.Set(initialize=sheet_home.home)
model.alpha = py.Param(initialize=39, doc='time of arrival ') # 6:30 = 39
@ -38,16 +32,16 @@ model.beta = py.Param(initialize=89, doc='time of departure ') # 7:00 = 89
model.t_final = py.Param(initialize=96, doc='final time of the day ')
model.second = py.Param(initialize=2, doc='second time of the day ')
model.t = py.Set(initialize=sheet_data.time, ordered=True, doc='time period')
model.EV_Dur = py.Set(initialize=RangeSet(model.alpha.value, model.beta.value),
model.EV_Dur = py.Set(initialize=py.RangeSet(model.alpha.value, model.beta.value),
doc='time interval for EV')
model.EV_Dur1 = py.Set(initialize=RangeSet(model.alpha.value + 1, model.beta.value),
model.EV_Dur1 = py.Set(initialize=py.RangeSet(model.alpha.value + 1, model.beta.value),
doc='time interval for EV after arrival')
model.t_second = py.Set(initialize=RangeSet(model.second.value, model.t_final.value),
model.t_second = py.Set(initialize=py.RangeSet(model.second.value, model.t_final.value),
doc='time greater then 1')
model.t_first = py.Param(initialize=1, doc='first time of the day ')
model.ist_interval = py.Set(initialize=RangeSet(model.t_first.value, model.alpha.value - 1),
model.ist_interval = py.Set(initialize=py.RangeSet(model.t_first.value, model.alpha.value - 1),
doc='time period before the arrival time')
model.second_interval = py.Set(initialize=RangeSet(model.beta.value + 1, model.t_final.value),
model.second_interval = py.Set(initialize=py.RangeSet(model.beta.value + 1, model.t_final.value),
doc='time interval after departure')
# </editor-fold>
@ -65,8 +59,9 @@ model.PV = py.Param(model.h, model.t, initialize=dict(zip(list(itertools.product
[i for x in sheet_PV.columns if "PV" in x for i in
sheet_PV.loc[:, x].values])), doc="PV Production")
model.max = py.Param(initialize=100, doc='maximum value for selling and buying power')
# <editor-fold desc="Electric Vehicle Parameters">
# todo this is need to be done for single value but tow home has seperate values
model.ChargeRate_EV = py.Param(model.h, initialize=dict(zip(list(itertools.product(model.h.data())),
[i for x in sheet_EV.columns if "charging_rate" in x for i
in
@ -125,10 +120,6 @@ model.DischRate_ESS = py.Param(model.h, initialize=dict(zip(list(itertools.produ
i in
sheet_ESS.loc[:, x].values])),
doc="Discharging rate of ESS")
# model.ChargeRate_ESS = py.Param(initialize=float(sheet_ESS.charging_rate1), doc='Charging rate of ESS ')
# model.DischRate_ESS = py.Param(initialize=float(sheet_ESS.discharging_rate1),
# doc='Discharging rate of ESS ')
model.Cap_ESS = py.Param(model.h, initialize=dict(zip((model.h.data()),
[i for x in sheet_ESS.columns if "capacity" in x for i in
sheet_ESS.loc[:, x].values])), doc="Capacity of ESS")
@ -217,14 +208,13 @@ model.water_use = py.Param(model.h, model.t,
sheet_wateruse.loc[:, x].values])), doc="Hot Water usage")
# </editor-fold2
print('this is first home')
# <editor-fold desc="Price, Base load, and time duration parameters">
model.Buy_price = py.Param(model.t, initialize=dict(zip(sheet_data.time, sheet_data.Buy_price)),
doc='Buying Price')
model.Sell_price = py.Param(model.t, initialize=dict(zip(sheet_data.time, sheet_data.Sell_price)), doc='Selling Price')
# Time duration: we took 15 mint granularity so in one hour it will be 1/4
model.time_d = py.Param(initialize=(1 / 4), doc='time duration ')
# model.DPT = py.Param(initialize=0.069, doc='Daily power tariff ') # .069
# </editor-fold>
# NOTE:Variable
@ -317,47 +307,6 @@ model.Const_3 = py.Constraint(model.h, model.t, rule=Status_Power,
# </editor-fold>
# <editor-fold desc="flexibility">
# model.P_max = py.Var(model.h, model.t, bounds=(0, None), doc=" Updated maximum demand of house")
# model.P_flex = py.Var(model.h, model.t, bounds=(0, None), doc=" Flexibility available in one house")
# model.P_flex_all = py.Var(model.h, model.t, bounds=(0, None), doc=" Sum of the Flexibility in all")
# model.S_device = py.Var(model.h, model.t, within=py.Binary, doc=" Status of the devices") # but i not using this
# model.flex_index = py.Param(doc='flexibility index ')
#
# #Todo need to verify these equation with omer:
#
# #
# def P_h_flex(model, h, i):
# return model.P_flex_all[i] == sum(
# model.base_load[h, i] * model.S_device[h, i] - model.base_load[h, i] * model.S_device[h, i] for h in model.h
# for i in model.t)
#
#
# model.Const_4a = py.Constraint(model.h, model.t, rule=P_h_flex, doc='sum of the flexibility')
#
#
# def P_flexibility(model, h, i):
# return model.P_flex_all[i] == sum([model.P_flex[h, i] for h in model.h for i in model.t])
#
#
# model.Const_4b = py.Constraint(model.h, model.t, rule=P_flexibility, doc='sum of the flexibility')
#
#
# def Flexibility_index(model, h, i):
# return model.flex_index[h, i] == model.P_flex[h, i] / model.P_flex_all[i]
#
#
# model.Const_4c = py.Constraint(model.h, model.t, rule=Flexibility_index, doc='sum of the flexibility')
#
#
# # this equation make the two way communication
# def P_maximum(model, h, i):
# return model.P_max[h, i] == model.P_max[h, i - 1] - model.P_flex[h, i]
#
#
# model.Const_4d = py.Constraint(model.h, model.t, rule=P_maximum, doc='updated maximum power')
# </editor-fold>
# <editor-fold desc="Electric vehicle Constraints">
@ -412,10 +361,6 @@ model.Const_EV_6 = py.Constraint(model.h, model.t, rule=Power_EV_Disch, doc='Dis
def SoC_EV1(model, h, i):
return model.Energy_EV[h, i] == model.Energy_EV_In[h]
# + (
# model.P_ESS_Charge[h, i] * model.Ch_Effc_ESS[h] - model.P_ESS_Disch[h, i]) * model.time_d
model.Const_EV_7a = py.Constraint(model.h, [model.alpha.value], rule=SoC_EV1, doc='SoC of the EV at arrival')
@ -426,14 +371,6 @@ def SoC_EV2(model, h, i):
model.Const_EV_7b = py.Constraint(model.h, model.EV_Dur1, rule=SoC_EV2, doc='SoC of the EV after arrival time')
# one equation of the SoC
# def SoC_EV(model, i):
# return model.Energy_EV[i] == model.Energy_EV_In + (model.P_EV_Charge[i] * model.Ch_Effc_EV - model.P_EV_Disch[i]) * model.time_d
#
#
# model.Const_EV_7 = py.Constraint(model.EV_Dur, rule=SoC_EV, doc='SoC of the EV')
def EV_availability1(model, h, i):
return model.Energy_EV[h, i] == 0
@ -490,15 +427,9 @@ model.Const_EV_12 = py.Constraint(model.h, [model.beta.value], rule=EV_final_SoC
# </editor-fold>
# <editor-fold desc="Energy Storage System Constraints">>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
# <editor-fold desc="Energy Storage System Constraints">
def SoC_ESS1(model, h, i):
return model.Energy_ESS[h, i] == model.Energy_ESS_In[h]
# #+ (
# model.P_ESS_Charge[h, i] * model.Ch_Effc_ESS[h] - model.P_ESS_Disch[h, i]) * model.time_d
model.Const_ESS_1a = py.Constraint(model.h, [model.t_first.value], rule=SoC_ESS1,
doc='SoC of ESS') # need to check wather we need to put in the brackets or not
@ -506,15 +437,11 @@ model.Const_ESS_1a = py.Constraint(model.h, [model.t_first.value], rule=SoC_ESS1
def SoC_ESS2(model, h, i):
return model.Energy_ESS[h, i] == model.Energy_ESS[h, i - 1] + (
model.P_ESS_Charge[h, i] * model.Ch_Effc_ESS[h] - model.P_ESS_Disch[h, i]) * model.time_d
model.Const_ESS_1b = py.Constraint(model.h, model.t_second, rule=SoC_ESS2, doc='SoC of ESS')
def Power_ESS_Charge(model, h, i):
return model.P_ESS_Charge[h, i] == model.P_ESS_Charge_Grid[h, i] + model.PV_Battery[h, i]
model.Const_ESS_2 = py.Constraint(model.h, model.t, rule=Power_ESS_Charge,
doc='Charging power of ESS ')
@ -614,8 +541,6 @@ def EWH_power(model, h, i):
model.Const_EWH_3 = py.Constraint(model.h, model.t, rule=EWH_power, doc='Electric water heater power')
#
# # NOTE: I write this function different from the GAMS
def EWH_temp_1(model, h, i):
return model.tetta_EWH_wat[h, i] == model.tetta_amb[h, i] + model.Q[h] * model.S_EWH[h, i] * model.R[
h] * model.time_d - (
@ -640,24 +565,14 @@ model.Const_EWH_5 = py.Constraint(model.h, model.t_second, rule=EWH_temp_2, doc=
# </editor-fold>
# <editor-fold desc="PV Constraints"> # need to include the PV equation of omer paper
# the below equation did not work got an error ( i dont know why!!!)
# def PV_production(model, h, i):
# return model.PV[ h,i] == model.PV_Grid[ h,i] + model.PV_Home[ h,i]
#
#
# model.Const_PV = py.Constraint(model.h, model.t, rule=PV_production, doc='PV Production')
def PV_production(model, h, i):
return model.PV_Grid[h, i] + model.PV_Home[h, i] + model.PV_Battery[h, i] == model.PV[h, i]
model.Const_PV = py.Constraint(model.h, model.t, rule=PV_production, doc='PV Production')
# </editor-fold>
# NOTE: I need to include the third term which is in the paper: daily peak demand * charge of the power
def objective_rule(model):
return sum((model.P_Buy_Grid[h, i] * model.Buy_price[i]) - (
model.P_Sell_Grid[h, i] * model.Sell_price[i]) for i in
@ -671,13 +586,10 @@ model.write('model2.lp', io_options={'symbolic_solver_labels': True})
opt = py.SolverFactory('cplex')
opt.options["mipgap"] = 0.09 # 0.155 for load and 0.8 for other load
result = opt.solve(model, tee=True)
# result = opt.solve(model, tee = True)
# model.pprint()
print(result)
# </editor-fold>
# <editor-fold desc="Check optimility">
if (result.solver.status == py.SolverStatus.ok) and (
result.solver.termination_condition == py.TerminationCondition.optimal):
print('optimal solution') # Do something when the solution in optimal and feasible
@ -799,10 +711,10 @@ for j in model.h:
SEWH[j - 1].append(py.value(v))
for i in model.Buy_price:
Buyprice.append(value(model.Buy_price[i]))
Buyprice.append(py.value(model.Buy_price[i]))
for i in model.Sell_price:
Sellprice.append(value(model.Sell_price[i]))
Sellprice.append(py.value(model.Sell_price[i]))
# alone
# for k in range(len(model.h)):
@ -895,8 +807,7 @@ for i in model.Sell_price:
# ax[2,0].set_ylabel('Power (kW)')
# ax[0, k].set_title(f"Home {k+1}")
#
# # pl.tight_layout()
#
# plt.show()
@ -914,7 +825,4 @@ for k in range(len(model.h)):
ax[1].set_ylabel('Electricity price ($/W/h)')
plt.suptitle(f"Home {k+1}")
plt.show()
# print('this is the end of code')
# # </editor-fold>