import json import os import time import numpy as np import pandas as pd import torch from concurrent.futures import ThreadPoolExecutor from deap import base, creator, tools, algorithms def fitness_torch(individuals, price, load, temperature, irradiance, wind_speed, prev_soc, prev_Pg1, prev_Pg2, prev_Pg3, device): individuals_torch = torch.tensor(individuals, device=device) num = individuals_torch.shape[0] Ac, Ag1, Ag2, Ag3, Av = [individuals_torch[:, i * period:(i + 1) * period] for i in range(5)] # soc = torch.zeros((num, period), device=device) # Pg1, Pg2, Pg3 = [torch.zeros((num, period), device=device) for _ in range(3)] # Cb, Cg1, Cg2, Cg3, Cs, Cw, Rs, Cp, Pe, Ps, Ee, Es = [torch.zeros((num, period), device=device) for _ in range(12)] soc = torch.clamp(prev_soc + 0.2 * Ac * 0.9, 0.2, 0.8) Pg1 = torch.clamp(prev_Pg1 + 100 * Ag1, min=0, max=150) Pg2 = torch.clamp(prev_Pg2 + 100 * Ag2, min=0, max=375) Pg3 = torch.clamp(prev_Pg3 + 200 * Ag3, min=0, max=500) Pso = torch.clamp((0.2 * irradiance + 0.05 * temperature - 9.25) * (1 + Av), min=0) Pw = torch.where( (wind_speed >= 3) & (wind_speed < 8), wind_speed ** 3 * 172.2625 / 1000, torch.where( (wind_speed >= 8) & (wind_speed < 12), 64 * 172.2625 / 125, torch.zeros_like(wind_speed) ) ) P = Ac + Pg1 + Pg2 + Pg3 + Pso + Pw Ee = torch.where(P >= load, P - load, torch.zeros_like(P)) Es = torch.where(P < load, load - P, torch.zeros_like(P)) Cb = 0.01 * Ac + 0.1 * soc Cg1 = 0.0034 * Pg1 ** 2 + 3 * Pg1 + 30 Cg2 = 0.001 * Pg2 ** 2 + 10 * Pg2 + 40 Cg3 = 0.001 * Pg3 ** 2 + 15 * Pg3 + 70 Cs = 0.01 * Pso Cw = 0.01 * Pw Rs = 0.5 * price * Ee Cp = price * Es Pe = torch.where(Ee > 100, (Ee - 100) * 50, torch.zeros_like(Ee)) Ps = torch.where(Es > 100, (Es - 100) * 50, torch.zeros_like(Es)) total_cost = torch.sum(Cb + Cg1 + Cg2 + Cg3 + Cs + Cw + Pe + Ps - Rs + Cp, dim=1) reward = -total_cost / 1000 return reward.cpu().numpy(), soc.cpu().numpy(), Pg1.cpu().numpy(), Pg2.cpu().numpy(), Pg3.cpu().numpy() def save_decision_values(best_ind, period, index): decisions = { 'Ac': best_ind[0], 'Ag1': best_ind[1], 'Ag2': best_ind[2], 'Ag3': best_ind[3], 'Av': best_ind[4] } with open(f'decision_values_{period}_index_{index}.json', 'w') as f: json.dump(decisions, f) def save_progress(population, period): population_data = [ind.tolist() for ind in population] with open(f'population_{period}.json', 'w') as f: json.dump({'period': period, 'population': population_data}, f) def load_progress(): if os.path.exists('population_gen_499.json'): with open('population_gen_499.json', 'r') as f: data = json.load(f) return data['population'], data['period'] return None, 0 def check_bounds(func): def wrapper(*args, **kwargs): offspring = func(*args, **kwargs) if offspring[0] is None or offspring[1] is None: print("Error: One of the offspring is None", offspring) raise ValueError("Offspring cannot be None.") for child in offspring: for i in range(len(child)): if child[i] < -1: child[i] = -1 elif child[i] > 1: child[i] = 1 return offspring return wrapper def main(): period = 8760 NGEN = 500 CXPB, MUTPB = 0.7, 0.2 batch_size = 500 device = torch.device("cuda" if torch.cuda.is_available() else "cpu") data = pd.read_csv('./data.csv') price, load, temperature, irradiance, wind_speed = [data[col].values for col in ['price', 'load', 'temperature', 'irradiance', 'wind_speed']] creator.create("FitnessMin", base.Fitness, weights=(-1.0,)) creator.create("Individual", list, fitness=creator.FitnessMin) toolbox = base.Toolbox() toolbox.register("attr_float", np.random.uniform, -1, 1) toolbox.register("individual", tools.initRepeat, creator.Individual, toolbox.attr_float, 5) toolbox.register("population", tools.initRepeat, list, toolbox.individual) toolbox.register("mate", check_bounds(tools.cxBlend), alpha=0.5) toolbox.register("mutate", check_bounds(tools.mutGaussian), mu=0, sigma=0.1, indpb=0.2) toolbox.register("select", tools.selTournament, tournsize=3) toolbox.register("evaluate", fitness_torch) population = load_progress() if population is None: population = toolbox.population(n=NGEN) print("Initial population:", len(population), "and first individual:", population[0]) prev_soc = torch.tensor([0.4] * NGEN, device=device) prev_Pg1 = torch.zeros(NGEN, device=device) prev_Pg2 = torch.zeros(NGEN, device=device) prev_Pg3 = torch.zeros(NGEN, device=device) for index in range(period): for gen in range(NGEN): start_time = time.time() offspring = algorithms.varAnd(population, toolbox, cxpb=CXPB, mutpb=MUTPB) num_individuals = len(offspring) with ThreadPoolExecutor() as executor: futures = [] for i in range(0, num_individuals, batch_size): batch = offspring[i:i + batch_size] individuals = [ind[:] for ind in batch] futures.append( executor.submit(toolbox.evaluate, individuals, price[index], load[index], temperature[index], irradiance[index], wind_speed[index], prev_soc, prev_Pg1, prev_Pg2, prev_Pg3, device) ) for future in futures: fitnesses, socs, Pg1s, Pg2s, Pg3s = zip(*future.result()) for ind, fitness in zip(offspring, fitnesses): ind.fitness.values = (fitness,) prev_soc[:len(socs)] = torch.tensor(socs, device=device) prev_Pg1[:len(Pg1s)] = torch.tensor(Pg1s, device=device) prev_Pg2[:len(Pg2s)] = torch.tensor(Pg2s, device=device) prev_Pg3[:len(Pg3s)] = torch.tensor(Pg3s, device=device) population = toolbox.select(offspring, k=len(population)) print("Population after selection:", population[:5]) end_time = time.time() print(f"第 {index + 1}小时完成花费 {end_time - start_time:.2f} 秒") best_ind = tools.selBest(population, 1)[0] print('最佳个体:', best_ind) print('适应度:', best_ind.fitness.values) save_decision_values(best_ind, period, index) save_progress(population, period) prev_soc.fill_(0.4) prev_Pg1.fill_(0) prev_Pg2.fill_(0) prev_Pg3.fill_(0) prev_soc[:len(best_ind)] = torch.tensor(best_ind[:period], device=device) prev_Pg1[:len(best_ind)] = torch.tensor(best_ind[period:2 * period], device=device) prev_Pg2[:len(best_ind)] = torch.tensor(best_ind[2 * period:3 * period], device=device) prev_Pg3[:len(best_ind)] = torch.tensor(best_ind[3 * period:4 * period], device=device) if __name__ == "__main__": main()