from flask import Flask, jsonify, request import numpy as np from datetime import datetime from modules.class_optimize import * # from modules.class_load import * # from modules.class_ems import * # from modules.class_pv_forecast import * # from modules.class_akku import * # from modules.class_strompreis import * # from modules.class_heatpump import * # from modules.class_load_container import * # from modules.class_eauto import * from modules.class_optimize import * from pprint import pprint import matplotlib.pyplot as plt from modules.visualize import * from deap import base, creator, tools, algorithms import numpy as np import random import os start_hour = 11 pv_forecast= [ 0, 0, 0, 0, 0, 0, 0, 35.4104640357043, 436.191574979506, 734.585613834398, 914.346108603927, 1019.5228214119, 1766.84136350058, 5980.60975052259, 6236.00681862336, 5893.38154543782, 4309.88538120413, 3338.29004915145, 2177.55660706753, 1091.00542545193, 437.819525591319, 44.2226537829726, 0, 0, 0, 0, -0.0269415125679914, 0, 0, 0, 0, 25.5745140893473, 494.188146846569, 943.821134036728, 1458.66413119635, 1819.46147983229, 2127.45430524539, 2267.78128099068, 5944.86706099518, 5337.1322153025, 4376.56125932204, 3020.00212091936, 2414.53994231359, 1373.626161377, 517.764497317134, 35.619750070296, 0, 0 ] temperature_forecast= [ 18.3, 17.8, 16.9, 16.2, 15.6, 15.1, 14.6, 14.2, 14.3, 14.8, 15.7, 16.7, 17.4, 18, 18.6, 19.2, 19.1, 18.7, 18.5, 17.7, 16.2, 14.6, 13.6, 13, 12.6, 12.2, 11.7, 11.6, 11.3, 11, 10.7, 10.2, 11.4, 14.4, 16.4, 18.3, 19.5, 20.7, 21.9, 22.7, 23.1, 23.1, 22.8, 21.8, 20.2, 19.1, 18, 17.4 ] strompreis_euro_pro_wh = [ 0.00033840228, 0.00033180228, 0.00032840228, 0.00032830228, 0.00032890228, 0.00033340228, 0.00032900228, 0.00033020228, 0.00030420228, 0.00024300228, 0.00022800228, 0.00022120228, 0.00020930228, 0.00018790228, 0.00018380228, 0.00020040228, 0.00021980228, 0.00022700228, 0.00029970228, 0.00031950228, 0.00030810228, 0.00029690228, 0.00029210228, 0.00027800228, 0.00033840228, 0.00033180228, 0.00032840228, 0.00032830228, 0.00032890228, 0.00033340228, 0.00032900228, 0.00033020228, 0.00030420228, 0.00024300228, 0.00022800228, 0.00022120228, 0.00020930228, 0.00018790228, 0.00018380228, 0.00020040228, 0.00021980228, 0.00022700228, 0.00029970228, 0.00031950228, 0.00030810228, 0.00029690228, 0.00029210228, 0.00027800228 ] gesamtlast= [ 546.16318964697, 893.072526185525, 448.7325491406, 460.696954446666, 497.688171532182, 468.186120420737, 424.440426628658, 454.341890696582, 1070.45287392313, 1096.46234344204, 1199.71317588613, 1294.39989535284, 1459.42631059004, 1295.23757474948, 1304.65748778424, 1187.47511606455, 1309.49984671163, 1106.60773651081, 1098.98136451936, 2112.82264661039, 1143.37118921705, 858.863135790621, 787.018517493612, 693.683533270357, 545.860858342826, 892.702692835489, 448.372058076642, 460.284228901714, 498.457870099476, 469.01826331988, 424.293997897019, 454.660942633609, 1070.71990586461, 1096.44557410693, 1199.72027861112, 1294.75410706442, 1459.61174223338, 1295.77661554687, 1304.95591385395, 1188.62778631227, 1310.43099742786, 1108.58589249073, 1101.73849714744, 2114.05576978017, 1143.68031998738, 858.607350786608, 786.574111043611, 693.463415886943 ] start_solution= [ 0, 1, 0, 1, 0, 1, 0, 0, 1, 1, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 ] parameter= {"preis_euro_pro_wh_akku": 30e-05,'pv_soc': 60.4052, 'pv_akku_cap': 30000, 'year_energy': 4100000, 'einspeiseverguetung_euro_pro_wh': 7e-05, 'max_heizleistung': 1000,"gesamtlast":gesamtlast, 'pv_forecast': pv_forecast, "temperature_forecast":temperature_forecast, "strompreis_euro_pro_wh":strompreis_euro_pro_wh, 'eauto_min_soc': 80, 'eauto_cap': 60000, 'eauto_charge_efficiency': 0.95, 'eauto_charge_power': 7590, 'eauto_soc': 53, 'pvpowernow': 211.137503624, 'start_solution': start_solution, 'haushaltsgeraet_wh': 937, 'haushaltsgeraet_dauer': 0} opt_class = optimization_problem(prediction_hours=48, strafe=10,optimization_hours=24) ergebnis = opt_class.optimierung_ems(parameter=parameter, start_hour=start_hour) # #Gesamtlast # ############# # gesamtlast = Gesamtlast() # # Load Forecast # ############### # lf = LoadForecast(filepath=r'load_profiles.npz', year_energy=year_energy) # #leistung_haushalt = lf.get_daily_stats(date)[0,...] # Datum anpassen # leistung_haushalt = lf.get_stats_for_date_range(date_now,date)[0,...].flatten() # # print(date_now," ",date) # # print(leistung_haushalt.shape) # gesamtlast.hinzufuegen("Haushalt", leistung_haushalt) # # PV Forecast # ############### # #PVforecast = PVForecast(filepath=os.path.join(r'test_data', r'pvprognose.json')) # PVforecast = PVForecast(prediction_hours = prediction_hours, url="https://api.akkudoktor.net/forecast?lat=50.8588&lon=7.3747&power=5000&azimuth=-10&tilt=7&powerInvertor=10000&horizont=20,27,22,20&power=4800&azimuth=-90&tilt=7&powerInvertor=10000&horizont=30,30,30,50&power=1400&azimuth=-40&tilt=60&powerInvertor=2000&horizont=60,30,0,30&power=1600&azimuth=5&tilt=45&powerInvertor=1400&horizont=45,25,30,60&past_days=5&cellCoEff=-0.36&inverterEfficiency=0.8&albedo=0.25&timezone=Europe%2FBerlin&hourly=relativehumidity_2m%2Cwindspeed_10m") # pv_forecast = PVforecast.get_pv_forecast_for_date_range(date_now,date) #get_forecast_for_date(date) # temperature_forecast = PVforecast.get_temperature_for_date_range(date_now,date) # # Strompreise # ############### # filepath = os.path.join (r'test_data', r'strompreise_akkudokAPI.json') # Pfad zur JSON-Datei anpassen # #price_forecast = HourlyElectricityPriceForecast(source=filepath) # price_forecast = HourlyElectricityPriceForecast(source="https://api.akkudoktor.net/prices?start="+date_now+"&end="+date+"") # specific_date_prices = price_forecast.get_price_for_daterange(date_now,date) # # print("13:",specific_date_prices[13]) # # print("14:",specific_date_prices[14]) # # print("15:",specific_date_prices[15]) # # sys.exit() # # WP # ############## # leistung_wp = wp.simulate_24h(temperature_forecast) # gesamtlast.hinzufuegen("Heatpump", leistung_wp) # # EAuto # ###################### # # leistung_eauto = eauto.get_stuendliche_last() # # soc_eauto = eauto.get_stuendlicher_soc() # # gesamtlast.hinzufuegen("eauto", leistung_eauto) # # print(gesamtlast.gesamtlast_berechnen()) # # EMS / Stromzähler Bilanz # #akku=None, pv_prognose_wh=None, strompreis_cent_pro_wh=None, einspeiseverguetung_cent_pro_wh=None, eauto=None, gesamtlast=None # ems = EnergieManagementSystem(akku=akku, gesamtlast = gesamtlast, pv_prognose_wh=pv_forecast, strompreis_cent_pro_wh=specific_date_prices, einspeiseverguetung_cent_pro_wh=einspeiseverguetung_cent_pro_wh, eauto=eauto) # o = ems.simuliere(start_hour)#ems.simuliere_ab_jetzt() # #pprint(o) # #pprint(o["Gesamtbilanz_Euro"]) # #visualisiere_ergebnisse(gesamtlast, pv_forecast, specific_date_prices, o,discharge_array,laden_moeglich, temperature_forecast, start_hour, prediction_hours) # # Optimierung # def evaluate_inner(individual): # #print(individual) # discharge_hours_bin = individual[0::2] # eautocharge_hours_float = individual[1::2] # #print(discharge_hours_bin) # #print(len(eautocharge_hours_float)) # ems.reset() # #eauto.reset() # ems.set_akku_discharge_hours(discharge_hours_bin) # ems.set_eauto_charge_hours(eautocharge_hours_float) # #eauto.set_laden_moeglich(eautocharge_hours_float) # #eauto.berechne_ladevorgang() # #leistung_eauto = eauto.get_stuendliche_last() # #gesamtlast.hinzufuegen("eauto", leistung_eauto) # #ems.set_gesamtlast(gesamtlast.gesamtlast_berechnen()) # o = ems.simuliere(start_hour) # return o, eauto # # Fitness-Funktion (muss Ihre EnergieManagementSystem-Logik integrieren) # def evaluate(individual): # o,eauto = evaluate_inner(individual) # gesamtbilanz = o["Gesamtbilanz_Euro"] # # Überprüfung, ob der Mindest-SoC erreicht wird # final_soc = eauto.ladezustand_in_prozent() # Nimmt den SoC am Ende des Optimierungszeitraums # strafe = 0.0 # #if final_soc < min_soc_eauto: # # Fügt eine Strafe hinzu, wenn der Mindest-SoC nicht erreicht wird # strafe = max(0,(min_soc_eauto - final_soc) * hohe_strafe ) # `hohe_strafe` ist ein vorher festgelegter Strafwert # gesamtbilanz += strafe # gesamtbilanz += o["Gesamt_Verluste"]/1000.0 # return (gesamtbilanz,) # # Werkzeug-Setup # creator.create("FitnessMin", base.Fitness, weights=(-1.0,)) # creator.create("Individual", list, fitness=creator.FitnessMin) # toolbox = base.Toolbox() # toolbox.register("attr_bool", random.randint, 0, 1) # toolbox.register("attr_bool", random.randint, 0, 1) # toolbox.register("individual", tools.initCycle, creator.Individual, (toolbox.attr_bool,toolbox.attr_bool), n=prediction_hours) # toolbox.register("population", tools.initRepeat, list, toolbox.individual) # toolbox.register("evaluate", evaluate) # toolbox.register("mate", tools.cxTwoPoint) # toolbox.register("mutate", tools.mutFlipBit, indpb=0.05) # toolbox.register("select", tools.selTournament, tournsize=3) # # Genetischer Algorithmus # def optimize(): # population = toolbox.population(n=500) # hof = tools.HallOfFame(1) # stats = tools.Statistics(lambda ind: ind.fitness.values) # stats.register("avg", np.mean) # stats.register("min", np.min) # stats.register("max", np.max) # algorithms.eaMuPlusLambda(population, toolbox, 50, 100, cxpb=0.5, mutpb=0.5, ngen=500, stats=stats, halloffame=hof, verbose=True) # #algorithms.eaSimple(population, toolbox, cxpb=0.2, mutpb=0.2, ngen=1000, stats=stats, halloffame=hof, verbose=True) # return hof[0] # start_solution = optimize() # print("Start Lösung:", start_solution) # # # Werkzeug-Setup # # creator.create("FitnessMin", base.Fitness, weights=(-1.0,)) # # creator.create("Individual", list, fitness=creator.FitnessMin) # # toolbox = base.Toolbox() # # toolbox.register("attr_bool", random.randint, 0, 1) # # toolbox.register("attr_float", random.uniform, 0.0, 1.0) # # toolbox.register("individual", tools.initCycle, creator.Individual, (toolbox.attr_bool,toolbox.attr_float), n=prediction_hours) # # start_individual = toolbox.individual() # # start_individual[:] = start_solution # # toolbox.register("population", tools.initRepeat, list, toolbox.individual) # # toolbox.register("evaluate", evaluate) # # toolbox.register("mate", tools.cxTwoPoint) # # toolbox.register("mutate", tools.mutFlipBit, indpb=0.05) # # toolbox.register("select", tools.selTournament, tournsize=3) # # # Genetischer Algorithmus # # def optimize(): # # population = toolbox.population(n=1000) # # population[0] = start_individual # # hof = tools.HallOfFame(1) # # stats = tools.Statistics(lambda ind: ind.fitness.values) # # stats.register("avg", np.mean) # # stats.register("min", np.min) # # stats.register("max", np.max) # # algorithms.eaMuPlusLambda(population, toolbox, 100, 200, cxpb=0.5, mutpb=0.2, ngen=1000, stats=stats, halloffame=hof, verbose=True) # # #algorithms.eaSimple(population, toolbox, cxpb=0.2, mutpb=0.2, ngen=1000, stats=stats, halloffame=hof, verbose=True) # # return hof[0] # # best_solution = optimize() # best_solution = start_solution # print("Beste Lösung:", best_solution) # #ems.set_akku_discharge_hours(best_solution) # o,eauto = evaluate_inner(best_solution) # # soc_eauto = eauto.get_stuendlicher_soc() # # print(soc_eauto) # # pprint(o) # # pprint(eauto.get_stuendlicher_soc()) # #visualisiere_ergebnisse(gesamtlast,leistung_haushalt,leistung_wp, pv_forecast, specific_date_prices, o,soc_eauto,best_solution[0::2],best_solution[1::2] , temperature_forecast) # visualisiere_ergebnisse(gesamtlast, pv_forecast, specific_date_prices, o,best_solution[0::2],best_solution[1::2] , temperature_forecast, start_hour, prediction_hours) # # for data in forecast.get_forecast_data(): # # print(data.get_date_time(), data.get_dc_power(), data.get_ac_power(), data.get_windspeed_10m(), data.get_temperature())for data in forecast.get_forecast_data(): # # app = Flask(__name__) # # @app.route('/getdata', methods=['GET']) # # def get_data(): # # # Hole das Datum aus den Query-Parametern # # date_str = request.args.get('date') # # year_energy = request.args.get('year_energy') # # try: # # # Konvertiere das Datum in ein datetime-Objekt # # date_obj = datetime.strptime(date_str, '%Y-%m-%d') # # filepath = r'.\load_profiles.npz' # Pfad zur JSON-Datei anpassen # # lf = cl.LoadForecast(filepath=filepath, year_energy=float(year_energy)) # # specific_date_prices = lf.get_daily_stats('2024-02-16') # # # Berechne den Tag des Jahres # # #day_of_year = date_obj.timetuple().tm_yday # # # Konvertiere den Tag des Jahres in einen String, falls die Schlüssel als Strings gespeichert sind # # #day_key = int(day_of_year) # # #print(day_key) # # # Überprüfe, ob der Tag im Jahr in den Daten vorhanden ist # # array_list = lf.get_daily_stats(date_str) # # pprint(array_list) # # pprint(array_list.shape) # # if array_list.shape == (2,24): # # #if day_key < len(load_profiles_exp): # # # Konvertiere das Array in eine Liste für die JSON-Antwort # # #((load_profiles_exp_l[day_key]).tolist(),(load_profiles_std_l)[day_key].tolist()) # # return jsonify({date_str: array_list.tolist()}) # # else: # # return jsonify({"error": "Datum nicht gefunden"}), 404 # # except ValueError: # # # Wenn das Datum nicht im richtigen Format ist oder ungültig ist # # return jsonify({"error": "Ungültiges Datum"}), 400 # # if __name__ == '__main__': # # app.run(debug=True)