Create app.py
Browse files
app.py
ADDED
|
@@ -0,0 +1,129 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import gradio as gr
|
| 2 |
+
import gymnasium as gym
|
| 3 |
+
from gymnasium import spaces
|
| 4 |
+
import numpy as np
|
| 5 |
+
import matplotlib.pyplot as plt
|
| 6 |
+
from stable_baselines3 import PPO
|
| 7 |
+
|
| 8 |
+
# --- 1. DEFINIMOS EL ENTORNO (Igual que en Colab) ---
|
| 9 |
+
class DataCenterEnv(gym.Env):
|
| 10 |
+
def __init__(self):
|
| 11 |
+
super(DataCenterEnv, self).__init__()
|
| 12 |
+
self.action_space = spaces.Discrete(5)
|
| 13 |
+
self.observation_space = spaces.Box(low=0, high=100, shape=(1,), dtype=np.float32)
|
| 14 |
+
self.state = 20
|
| 15 |
+
self.max_steps = 100
|
| 16 |
+
self.current_step = 0
|
| 17 |
+
self.temps = []
|
| 18 |
+
self.energies = []
|
| 19 |
+
|
| 20 |
+
def reset(self, seed=None, options=None):
|
| 21 |
+
super().reset(seed=seed)
|
| 22 |
+
self.state = 20 + np.random.rand()
|
| 23 |
+
self.current_step = 0
|
| 24 |
+
self.temps = [self.state]
|
| 25 |
+
self.energies = []
|
| 26 |
+
return np.array([self.state], dtype=np.float32), {}
|
| 27 |
+
|
| 28 |
+
def step(self, action):
|
| 29 |
+
# Usamos variables globales para simular la carga que el usuario elija en la web
|
| 30 |
+
global USER_CPU_LOAD_MIN, USER_CPU_LOAD_MAX
|
| 31 |
+
cpu_load = np.random.uniform(USER_CPU_LOAD_MIN, USER_CPU_LOAD_MAX)
|
| 32 |
+
|
| 33 |
+
cooling_effect = action * 2.5 # Potencia industrial
|
| 34 |
+
energy_cost = action ** 2
|
| 35 |
+
|
| 36 |
+
self.state = self.state + cpu_load - cooling_effect
|
| 37 |
+
self.state = np.clip(self.state, 10, 100)
|
| 38 |
+
self.current_step += 1
|
| 39 |
+
|
| 40 |
+
reward = 0
|
| 41 |
+
if 37 <= self.state <= 60: reward += 5
|
| 42 |
+
else: reward -= 2
|
| 43 |
+
if self.state > 80: reward -= 50
|
| 44 |
+
reward -= (energy_cost * 0.1)
|
| 45 |
+
|
| 46 |
+
terminated = False
|
| 47 |
+
if self.current_step >= self.max_steps: terminated = True
|
| 48 |
+
|
| 49 |
+
self.temps.append(self.state)
|
| 50 |
+
self.energies.append(energy_cost)
|
| 51 |
+
|
| 52 |
+
return np.array([self.state], dtype=np.float32), reward, terminated, False, {}
|
| 53 |
+
|
| 54 |
+
# Variables globales para controlar la dificultad desde la web
|
| 55 |
+
USER_CPU_LOAD_MIN = 1
|
| 56 |
+
USER_CPU_LOAD_MAX = 4
|
| 57 |
+
|
| 58 |
+
# --- 2. FUNCIÓN DE SIMULACIÓN (Lo que hace la web) ---
|
| 59 |
+
def run_simulation(load_min, load_max):
|
| 60 |
+
# Actualizamos las variables globales con lo que el usuario eligió
|
| 61 |
+
global USER_CPU_LOAD_MIN, USER_CPU_LOAD_MAX
|
| 62 |
+
USER_CPU_LOAD_MIN = load_min
|
| 63 |
+
USER_CPU_LOAD_MAX = load_max
|
| 64 |
+
|
| 65 |
+
# 1. Crear entorno y Cargar Modelo
|
| 66 |
+
env = DataCenterEnv()
|
| 67 |
+
|
| 68 |
+
# INTENTA CARGAR TU MODELO. Si no lo encuentra, usa uno aleatorio (para que no falle la demo)
|
| 69 |
+
try:
|
| 70 |
+
model = PPO.load("agente_aire_acondicionado.zip", env=env)
|
| 71 |
+
msg = "✅ Modelo Inteligente Cargado Correctamente."
|
| 72 |
+
except:
|
| 73 |
+
model = None
|
| 74 |
+
msg = "⚠️ ADVERTENCIA: No se encontró 'agente_aire_acondicionado.zip'. Usando acciones aleatorias."
|
| 75 |
+
|
| 76 |
+
# 2. Correr la simulación
|
| 77 |
+
obs, _ = env.reset()
|
| 78 |
+
done = False
|
| 79 |
+
|
| 80 |
+
while not done:
|
| 81 |
+
if model:
|
| 82 |
+
action, _ = model.predict(obs)
|
| 83 |
+
else:
|
| 84 |
+
action = env.action_space.sample() # Fallback aleatorio
|
| 85 |
+
|
| 86 |
+
obs, reward, done, truncated, info = env.step(action)
|
| 87 |
+
|
| 88 |
+
# 3. Generar Gráficos
|
| 89 |
+
fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(12, 5))
|
| 90 |
+
|
| 91 |
+
# Gráfico Temperatura
|
| 92 |
+
ax1.plot(env.temps, color="red", label="Temperatura")
|
| 93 |
+
ax1.axhline(y=37, color='green', linestyle='--', label="Mínimo (37°C)")
|
| 94 |
+
ax1.axhline(y=60, color='green', linestyle='--', label="Máximo (60°C)")
|
| 95 |
+
ax1.set_title("Control de Temperatura")
|
| 96 |
+
ax1.set_ylabel("°C")
|
| 97 |
+
ax1.set_ylim(10, 90)
|
| 98 |
+
ax1.legend()
|
| 99 |
+
ax1.grid(True)
|
| 100 |
+
|
| 101 |
+
# Gráfico Energía
|
| 102 |
+
ax2.plot(env.energies, color="blue", label="Energía")
|
| 103 |
+
ax2.set_title("Consumo de Energía")
|
| 104 |
+
ax2.set_ylabel("Watts")
|
| 105 |
+
ax2.grid(True)
|
| 106 |
+
|
| 107 |
+
plt.tight_layout()
|
| 108 |
+
return fig, msg
|
| 109 |
+
|
| 110 |
+
# --- 3. INTERFAZ GRÁFICA (Gradio) ---
|
| 111 |
+
with gr.Blocks() as demo:
|
| 112 |
+
gr.Markdown("# ❄️ AI Cooling System Optimization Agent")
|
| 113 |
+
gr.Markdown("Este agente de Aprendizaje por Refuerzo (RL) controla el aire acondicionado de un servidor para ahorrar energía sin que se sobrecaliente.")
|
| 114 |
+
|
| 115 |
+
with gr.Row():
|
| 116 |
+
with gr.Column():
|
| 117 |
+
gr.Markdown("### Configuración de Carga")
|
| 118 |
+
s_min = gr.Slider(1, 10, value=1, label="Carga Mínima de CPU (Calor)")
|
| 119 |
+
s_max = gr.Slider(1, 10, value=4, label="Carga Máxima de CPU (Calor)")
|
| 120 |
+
btn = gr.Button("🚀 Ejecutar Simulación")
|
| 121 |
+
status = gr.Textbox(label="Estado del Agente")
|
| 122 |
+
|
| 123 |
+
with gr.Column():
|
| 124 |
+
plot = gr.Plot(label="Resultados en Tiempo Real")
|
| 125 |
+
|
| 126 |
+
btn.click(fn=run_simulation, inputs=[s_min, s_max], outputs=[plot, status])
|
| 127 |
+
|
| 128 |
+
# Lanzar la app
|
| 129 |
+
demo.launch()
|