File size: 11,633 Bytes
47f1575 | 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 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 | {
"cells": [
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# ⏱️ TP-4 : Prédiction de Séries Temporelles avec LSTM\n",
"\n",
"**Objectif** : Prédire la consommation électrique avec des réseaux LSTM.\n",
"\n",
"**Compétences** :\n",
"- Préparation de données temporelles\n",
"- Fenêtres glissantes (windowing)\n",
"- Architecture LSTM avec Keras\n",
"- Early stopping et régularisation"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"import numpy as np\n",
"import pandas as pd\n",
"import matplotlib.pyplot as plt\n",
"from sklearn.preprocessing import MinMaxScaler\n",
"from sklearn.metrics import mean_squared_error, mean_absolute_error\n",
"\n",
"import tensorflow as tf\n",
"from tensorflow.keras.models import Sequential\n",
"from tensorflow.keras.layers import LSTM, Dense, Dropout\n",
"from tensorflow.keras.callbacks import EarlyStopping, ReduceLROnPlateau\n",
"\n",
"# Reproductibilité\n",
"np.random.seed(42)\n",
"tf.random.set_seed(42)\n",
"\n",
"print(f\"✅ TensorFlow version : {tf.__version__}\")"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# Génération de données synthétiques (consommation électrique)\n",
"# En pratique, remplacez par vos données réelles\n",
"\n",
"def generate_energy_data(n_days=365*2):\n",
" \"\"\"Génère des données de consommation électrique simulées\"\"\"\n",
" hours = np.arange(n_days * 24)\n",
" \n",
" # Tendance\n",
" trend = 0.001 * hours\n",
" \n",
" # Saisonnalité journalière\n",
" daily = 10 * np.sin(2 * np.pi * hours / 24)\n",
" \n",
" # Saisonnalité hebdomadaire\n",
" weekly = 5 * np.sin(2 * np.pi * hours / (24 * 7))\n",
" \n",
" # Saisonnalité annuelle\n",
" yearly = 15 * np.sin(2 * np.pi * hours / (24 * 365))\n",
" \n",
" # Bruit\n",
" noise = np.random.normal(0, 3, len(hours))\n",
" \n",
" # Consommation totale\n",
" consumption = 50 + trend + daily + weekly + yearly + noise\n",
" consumption = np.maximum(consumption, 0) # Pas de valeurs négatives\n",
" \n",
" return consumption\n",
"\n",
"# Génération des données\n",
"data = generate_energy_data(n_days=730) # 2 ans de données\n",
"\n",
"# Création du DataFrame\n",
"dates = pd.date_range(start='2022-01-01', periods=len(data), freq='H')\n",
"df = pd.DataFrame({'consumption': data}, index=dates)\n",
"\n",
"print(f\"📊 Période : {df.index[0]} à {df.index[-1]}\")\n",
"print(f\"📊 Total : {len(df)} heures de données\")\n",
"df.head()"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# Visualisation des données\n",
"fig, axes = plt.subplots(3, 1, figsize=(15, 10))\n",
"\n",
"# Vue complète\n",
"axes[0].plot(df.index, df['consumption'], alpha=0.7)\n",
"axes[0].set_title('Consommation électrique - Vue complète (2 ans)')\n",
"axes[0].set_ylabel('kWh')\n",
"\n",
"# Vue d'une semaine\n",
"one_week = df.iloc[:24*7]\n",
"axes[1].plot(one_week.index, one_week['consumption'], marker='o')\n",
"axes[1].set_title('Consommation - Vue hebdomadaire')\n",
"axes[1].set_ylabel('kWh')\n",
"\n",
"# Vue d'une journée\n",
"one_day = df.iloc[:24]\n",
"axes[2].plot(one_day.index.hour, one_day['consumption'], marker='o')\n",
"axes[2].set_title('Consommation - Vue journalière')\n",
"axes[2].set_xlabel('Heure')\n",
"axes[2].set_ylabel('kWh')\n",
"\n",
"plt.tight_layout()\n",
"plt.show()"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# Feature Engineering temporel\n",
"df['hour'] = df.index.hour\n",
"df['day_of_week'] = df.index.dayofweek\n",
"df['month'] = df.index.month\n",
"df['is_weekend'] = (df['day_of_week'] >= 5).astype(int)\n",
"\n",
"# Lags (valeurs précédentes)\n",
"for lag in [1, 2, 3, 24, 48]:\n",
" df[f'lag_{lag}'] = df['consumption'].shift(lag)\n",
"\n",
"# Rolling statistics\n",
"df['rolling_mean_24'] = df['consumption'].rolling(window=24).mean()\n",
"df['rolling_std_24'] = df['consumption'].rolling(window=24).std()\n",
"\n",
"# Suppression des NaN\n",
"df = df.dropna()\n",
"\n",
"print(f\"✅ Features créées : {df.shape[1]} colonnes\")\n",
"df.head()"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# Préparation des séquences pour LSTM\n",
"def create_sequences(data, target_col, sequence_length=24):\n",
" \"\"\"\n",
" Crée des séquences pour LSTM\n",
" data : DataFrame avec features\n",
" target_col : nom de la colonne cible\n",
" sequence_length : longueur de la séquence (ex: 24 heures)\n",
" \"\"\"\n",
" X, y = [], []\n",
" values = data.values\n",
" target_idx = data.columns.get_loc(target_col)\n",
" \n",
" for i in range(sequence_length, len(values)):\n",
" X.append(values[i-sequence_length:i])\n",
" y.append(values[i, target_idx])\n",
" \n",
" return np.array(X), np.array(y)\n",
"\n",
"# Séparation train/test\n",
"train_size = int(len(df) * 0.8)\n",
"train_df = df.iloc[:train_size]\n",
"test_df = df.iloc[train_size:]\n",
"\n",
"# Normalisation\n",
"scaler = MinMaxScaler()\n",
"train_scaled = scaler.fit_transform(train_df)\n",
"test_scaled = scaler.transform(test_df)\n",
"\n",
"# Conversion en DataFrame pour garder les noms de colonnes\n",
"train_scaled = pd.DataFrame(train_scaled, columns=df.columns, index=train_df.index)\n",
"test_scaled = pd.DataFrame(test_scaled, columns=df.columns, index=test_df.index)\n",
"\n",
"# Création des séquences\n",
"SEQUENCE_LENGTH = 24 # 24 heures d'historique\n",
"\n",
"X_train, y_train = create_sequences(train_scaled, 'consumption', SEQUENCE_LENGTH)\n",
"X_test, y_test = create_sequences(test_scaled, 'consumption', SEQUENCE_LENGTH)\n",
"\n",
"print(f\"📊 X_train shape : {X_train.shape}\")\n",
"print(f\"📊 y_train shape : {y_train.shape}\")\n",
"print(f\"📊 X_test shape : {X_test.shape}\")\n",
"print(f\"📊 y_test shape : {y_test.shape}\")"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# Construction du modèle LSTM\n",
"model = Sequential([\n",
" LSTM(64, return_sequences=True, input_shape=(SEQUENCE_LENGTH, X_train.shape[2])),\n",
" Dropout(0.2),\n",
" LSTM(32, return_sequences=False),\n",
" Dropout(0.2),\n",
" Dense(16, activation='relu'),\n",
" Dense(1)\n",
"])\n",
"\n",
"model.compile(\n",
" optimizer=tf.keras.optimizers.Adam(learning_rate=0.001),\n",
" loss='mse',\n",
" metrics=['mae']\n",
")\n",
"\n",
"model.summary()"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# Callbacks\n",
"callbacks = [\n",
" EarlyStopping(\n",
" monitor='val_loss',\n",
" patience=10,\n",
" restore_best_weights=True\n",
" ),\n",
" ReduceLROnPlateau(\n",
" monitor='val_loss',\n",
" factor=0.5,\n",
" patience=5,\n",
" min_lr=1e-6\n",
" )\n",
"]\n",
"\n",
"# Entraînement\n",
"history = model.fit(\n",
" X_train, y_train,\n",
" epochs=100,\n",
" batch_size=32,\n",
" validation_split=0.2,\n",
" callbacks=callbacks,\n",
" verbose=1\n",
")"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# Visualisation de l'entraînement\n",
"fig, axes = plt.subplots(1, 2, figsize=(14, 5))\n",
"\n",
"# Loss\n",
"axes[0].plot(history.history['loss'], label='Train')\n",
"axes[0].plot(history.history['val_loss'], label='Validation')\n",
"axes[0].set_title('Loss (MSE)')\n",
"axes[0].set_xlabel('Epoch')\n",
"axes[0].set_ylabel('Loss')\n",
"axes[0].legend()\n",
"\n",
"# MAE\n",
"axes[1].plot(history.history['mae'], label='Train')\n",
"axes[1].plot(history.history['val_mae'], label='Validation')\n",
"axes[1].set_title('MAE')\n",
"axes[1].set_xlabel('Epoch')\n",
"axes[1].set_ylabel('MAE')\n",
"axes[1].legend()\n",
"\n",
"plt.tight_layout()\n",
"plt.show()"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# Prédictions\n",
"y_pred = model.predict(X_test)\n",
"\n",
"# Métriques (sur données normalisées)\n",
"mse = mean_squared_error(y_test, y_pred)\n",
"mae = mean_absolute_error(y_test, y_pred)\n",
"rmse = np.sqrt(mse)\n",
"\n",
"print(f\"📊 MSE : {mse:.6f}\")\n",
"print(f\"📊 MAE : {mae:.6f}\")\n",
"print(f\"📊 RMSE : {rmse:.6f}\")"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# Visualisation des prédictions\n",
"plt.figure(figsize=(15, 6))\n",
"\n",
"# Plot des 500 premières prédictions\n",
"n_plot = 500\n",
"plt.plot(y_test[:n_plot], label='Réel', alpha=0.8)\n",
"plt.plot(y_pred[:n_plot], label='Prédit', alpha=0.8)\n",
"plt.title(f'Prédictions LSTM - {n_plot} premiers points de test')\n",
"plt.xlabel('Temps')\n",
"plt.ylabel('Consommation (normalisée)')\n",
"plt.legend()\n",
"plt.show()"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# Scatter plot : Réel vs Prédit\n",
"plt.figure(figsize=(8, 8))\n",
"plt.scatter(y_test, y_pred, alpha=0.5)\n",
"plt.plot([y_test.min(), y_test.max()], [y_test.min(), y_test.max()], 'r--', lw=2)\n",
"plt.xlabel('Valeurs réelles')\n",
"plt.ylabel('Valeurs prédites')\n",
"plt.title('Réel vs Prédit')\n",
"plt.show()"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## 🎓 Conclusion\n",
"\n",
"Dans ce TP, nous avons :\n",
"\n",
"1. ✅ **Généré** des données de consommation électrique avec patterns temporels\n",
"2. ✅ **Créé** des features temporelles (heure, jour, mois, lags, rolling stats)\n",
"3. ✅ **Préparé** les séquences pour LSTM avec windowing\n",
"4. ✅ **Construit** un modèle LSTM avec Dropout et Early Stopping\n",
"5. ✅ **Évalué** les performances sur l'ensemble de test\n",
"\n",
"**Améliorations possibles** :\n",
"- Utiliser des données météo comme features externes\n",
"- Tester des architectures plus complexes (Bidirectional LSTM, GRU)\n",
"- Faire du multi-step forecasting (prédire plusieurs heures en avance)"
]
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3",
"language": "python",
"name": "python3"
},
"language_info": {
"name": "python",
"version": "3.8.0"
}
},
"nbformat": 4,
"nbformat_minor": 4
}
|