- 🍨 本文为🔗365天深度学习训练营 中的学习记录博客
- 🍦 参考文章：第R2周：LSTM-火灾温度预测（训练营内部可读）
- 🍖 作者：K同学啊
任务说明：数据集中提供了火灾温度（Tem1）、一氧化碳浓度（CO 1）、烟雾浓度（Soot 1）随着时间变化数据，我们需要根据这些数据对未来某一时刻的火灾温度做出预测（本次任务仅供学习）
🍺要求：
了解LSTM是什么，并使用其构建一个完整的程序
R2达到0.83(完成)
🍻拔高：
使用第1-8个时刻的数据预测第9-10个时刻的温度数据(完成)
一句话介绍LSTM，它是RNN的进阶版，如果说RNN的最大限度是理解一句话，那么LSTM的最大限度则是理解一段话，详细介绍如下：
LSTM，全称为长短期记忆网络(Long Short Term Memory networks)，是一种特殊的RNN，能够学习到长期依赖关系。LSTM由Hochreiter & Schmidhuber (1997)提出，许多研究者进行了一系列的工作对其改进并使之发扬光大。LSTM在许多问题上效果非常好，现在被广泛使用。
一.前期准备工作
1.导入数据
import tensorflow as tf
import pandas as pd
import numpy as np
gpus = tf.config.list_physical_devices("GPU")
if gpus:
tf.config.experimental.set_memory_growth(gpus[0], True) #设置GPU显存用量按需使用
tf.config.set_visible_devices([gpus[0]],"GPU")
print(gpus)
[PhysicalDevice(name='/physical_device:GPU:0', device_type='GPU')]
df_1 = pd.read_csv("D:/woodpine2.csv")
df_1.head()
| Time | Tem1 | CO 1 | Soot 1 | |
|---|---|---|---|---|
| 0 | 0.000 | 25.0 | 0.0 | 0.0 |
| 1 | 0.228 | 25.0 | 0.0 | 0.0 |
| 2 | 0.456 | 25.0 | 0.0 | 0.0 |
| 3 | 0.685 | 25.0 | 0.0 | 0.0 |
| 4 | 0.913 | 25.0 | 0.0 | 0.0 |
2.数据可视化
import matplotlib.pyplot as plt
import seaborn as sns
plt.rcParams['savefig.dpi'] = 500 #图片像素
plt.rcParams['figure.dpi'] = 500 #分辨率
fig, ax =plt.subplots(1,3,constrained_layout=True, figsize=(14, 3))
sns.lineplot(data=df_1["Tem1"], ax=ax[0])
sns.lineplot(data=df_1["CO 1"], ax=ax[1])
sns.lineplot(data=df_1["Soot 1"], ax=ax[2])
plt.show()
[外链图片转存失败,源站可能有防盗链机制,建议将图片保存下来直接上传(img-qrzltiLi-1665993981350)(output_9_0.png)]
二.构建数据集
dataFrame=df_1.iloc[:,1:]
dataFrame.head()
| Tem1 | CO 1 | Soot 1 | |
|---|---|---|---|
| 0 | 25.0 | 0.0 | 0.0 |
| 1 | 25.0 | 0.0 | 0.0 |
| 2 | 25.0 | 0.0 | 0.0 |
| 3 | 25.0 | 0.0 | 0.0 |
| 4 | 25.0 | 0.0 | 0.0 |
1.设置X,y
width_X=8
width_y=2
取前8个时间段的Tem1、CO 1、Soot 1为X，第9,10个时间段的Tem1为y。
X = []
y = []
in_start = 0
for _, _ in df_1.iterrows():
in_end = in_start + width_X
out_end = in_end + width_y
if out_end < len(dataFrame):
X_ = np.array(dataFrame.iloc[in_start:in_end , ])
X_ = X_.reshape((len(X_)*3))
y_ = np.array(dataFrame.iloc[in_end :out_end, 0])
X.append(X_)
y.append(y_)
in_start += 1
X = np.array(X)
y = np.array(y)
X.shape, y.shape
((5938, 24), (5938, 2))
2.归一化
from sklearn.preprocessing import MinMaxScaler
#将数据归一化，范围是0到1
sc = MinMaxScaler(feature_range=(0, 1))
X_scaled = sc.fit_transform(X)
X_scaled.shape
(5938, 24)
X_scaled=X_scaled.reshape(len(X_scaled),width_X,3)
X_scaled.shape
(5938, 8, 3)
3.划分数据集
取5000之前的数据为训练集，5000之后的为验证集
X_train=X_scaled[:5000]
y_train=y[:5000]
X_test=X_scaled[5000:,]
y_test=y[5000:,]
X_train.shape
(5000, 8, 3)
三.构建模型
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense,LSTM,Bidirectional
from tensorflow.keras import Input
model_lstm = Sequential()
model_lstm.add(LSTM(units=64, activation='relu', return_sequences=True,
input_shape=(X_train.shape[1], 3)))
model_lstm.add(LSTM(units=64, activation='relu'))
model_lstm.add(Dense(width_y))
WARNING:tensorflow:Layer lstm_8 will not use cuDNN kernel since it doesn't meet the cuDNN kernel criteria. It will use generic GPU kernel as fallback when running on GPU
WARNING:tensorflow:Layer lstm_9 will not use cuDNN kernel since it doesn't meet the cuDNN kernel criteria. It will use generic GPU kernel as fallback when running on GPU
四.模型训练
1.模型编译
#只观察loss数值，不观察准确率，所以删去metrics选项
model_lstm.compile(optimizer=tf.keras.optimizers.Adam(1e-3),
loss='mean_squared_error')
from tensorflow.keras.callbacks import ModelCheckpoint
ModelCheckPointer=ModelCheckpoint('best_model.h5',
monitor='val_loss',
save_best_only=True,
save_weights_only=True,
)
X_train.shape,y_train.shape
((5000, 8, 3), (5000, 2))
history_lstm=model_lstm.fit(X_train,y_train,
batch_size=64,
epochs=50,
validation_data=(X_test,y_test),
validation_freq=1,
callbacks=[ModelCheckPointer])
Epoch 1/50
79/79 [==============================] - 1s 16ms/step - loss: 21.1715 - val_loss: 52.4376
Epoch 2/50
79/79 [==============================] - 1s 13ms/step - loss: 5.5628 - val_loss: 46.9862
Epoch 3/50
79/79 [==============================] - 1s 13ms/step - loss: 6.2596 - val_loss: 55.9755
Epoch 4/50
79/79 [==============================] - 1s 12ms/step - loss: 6.1024 - val_loss: 49.6590
Epoch 5/50
79/79 [==============================] - 1s 12ms/step - loss: 6.3087 - val_loss: 46.9177
Epoch 6/50
79/79 [==============================] - 1s 12ms/step - loss: 5.9893 - val_loss: 53.0820
Epoch 7/50
79/79 [==============================] - 1s 13ms/step - loss: 7.2166 - val_loss: 46.6139
Epoch 8/50
79/79 [==============================] - 1s 12ms/step - loss: 8.3203 - val_loss: 49.4651
Epoch 9/50
79/79 [==============================] - 1s 12ms/step - loss: 7.0174 - val_loss: 85.3318
Epoch 10/50
79/79 [==============================] - 1s 12ms/step - loss: 8.1098 - val_loss: 84.6306
Epoch 11/50
79/79 [==============================] - 1s 12ms/step - loss: 6.8533 - val_loss: 50.6454
Epoch 12/50
79/79 [==============================] - 1s 12ms/step - loss: 7.5618 - val_loss: 46.3002
Epoch 13/50
79/79 [==============================] - 1s 12ms/step - loss: 6.3271 - val_loss: 49.1259
Epoch 14/50
79/79 [==============================] - 1s 12ms/step - loss: 7.2708 - val_loss: 46.7672
Epoch 15/50
79/79 [==============================] - 1s 13ms/step - loss: 9.0250 - val_loss: 45.6259
Epoch 16/50
79/79 [==============================] - 1s 13ms/step - loss: 7.4962 - val_loss: 47.3104
Epoch 17/50
79/79 [==============================] - 1s 13ms/step - loss: 6.3288 - val_loss: 45.8246
Epoch 18/50
79/79 [==============================] - 1s 12ms/step - loss: 6.3963 - val_loss: 46.1642
Epoch 19/50
79/79 [==============================] - 1s 13ms/step - loss: 8.0644 - val_loss: 45.5498
Epoch 20/50
79/79 [==============================] - 1s 13ms/step - loss: 7.6460 - val_loss: 52.4149
Epoch 21/50
79/79 [==============================] - 1s 13ms/step - loss: 7.9108 - val_loss: 45.2785
Epoch 22/50
79/79 [==============================] - 1s 13ms/step - loss: 7.2452 - val_loss: 48.1493
Epoch 23/50
79/79 [==============================] - 1s 13ms/step - loss: 6.8818 - val_loss: 44.7868
Epoch 24/50
79/79 [==============================] - 1s 13ms/step - loss: 8.7648 - val_loss: 45.3644
Epoch 25/50
79/79 [==============================] - 1s 13ms/step - loss: 7.2923 - val_loss: 113.3569
Epoch 26/50
79/79 [==============================] - 1s 13ms/step - loss: 7.2891 - val_loss: 48.8831
Epoch 27/50
79/79 [==============================] - 1s 13ms/step - loss: 9.0204 - val_loss: 54.1173
Epoch 28/50
79/79 [==============================] - 1s 13ms/step - loss: 6.2687 - val_loss: 57.7011
Epoch 29/50
79/79 [==============================] - 1s 14ms/step - loss: 9.3178 - val_loss: 56.6076
Epoch 30/50
79/79 [==============================] - 1s 14ms/step - loss: 7.5619 - val_loss: 45.1185
Epoch 31/50
79/79 [==============================] - 1s 13ms/step - loss: 7.6877 - val_loss: 44.9177
Epoch 32/50
79/79 [==============================] - 1s 13ms/step - loss: 7.7795 - val_loss: 55.4915
Epoch 33/50
79/79 [==============================] - 1s 13ms/step - loss: 6.1562 - val_loss: 55.6177
Epoch 34/50
79/79 [==============================] - 1s 14ms/step - loss: 6.5898 - val_loss: 87.1372
Epoch 35/50
79/79 [==============================] - 1s 14ms/step - loss: 10.0048 - val_loss: 47.4471
Epoch 36/50
79/79 [==============================] - 1s 14ms/step - loss: 6.9753 - val_loss: 45.5101
Epoch 37/50
79/79 [==============================] - 1s 14ms/step - loss: 6.5068 - val_loss: 45.6193
Epoch 38/50
79/79 [==============================] - 1s 14ms/step - loss: 6.8584 - val_loss: 103.6446
Epoch 39/50
79/79 [==============================] - 1s 14ms/step - loss: 6.6473 - val_loss: 44.0263
Epoch 40/50
79/79 [==============================] - 1s 14ms/step - loss: 5.8646 - val_loss: 51.5055
Epoch 41/50
79/79 [==============================] - 1s 14ms/step - loss: 6.6104 - val_loss: 45.9765
Epoch 42/50
79/79 [==============================] - 1s 14ms/step - loss: 7.4796 - val_loss: 50.2358
Epoch 43/50
79/79 [==============================] - 1s 14ms/step - loss: 6.3155 - val_loss: 61.4754
Epoch 44/50
79/79 [==============================] - 1s 14ms/step - loss: 6.8296 - val_loss: 51.9304
Epoch 45/50
79/79 [==============================] - 1s 14ms/step - loss: 7.3016 - val_loss: 45.8511
Epoch 46/50
79/79 [==============================] - 1s 14ms/step - loss: 6.7835 - val_loss: 45.0044
Epoch 47/50
79/79 [==============================] - 1s 14ms/step - loss: 6.6329 - val_loss: 43.3930
Epoch 48/50
79/79 [==============================] - 1s 14ms/step - loss: 6.3484 - val_loss: 50.0709
Epoch 49/50
79/79 [==============================] - 1s 14ms/step - loss: 6.2621 - val_loss: 52.4607
Epoch 50/50
79/79 [==============================] - 1s 18ms/step - loss: 6.8254 - val_loss: 52.6669
五.评估
1.loss图
# 支持中文
plt.rcParams['font.sans-serif'] = ['SimHei'] # 用来正常显示中文标签
plt.rcParams['axes.unicode_minus'] = False # 用来正常显示负号
plt.figure(figsize=(5, 3),dpi=120)
plt.plot(history_lstm.history['loss'] , label='LSTM Training Loss')
plt.plot(history_lstm.history['val_loss'], label='LSTM Validation Loss')
plt.title('Training and Validation Loss')
plt.legend()
plt.show()
[外链图片转存失败,源站可能有防盗链机制,建议将图片保存下来直接上传(img-Pw5LoqXV-1665993981351)(output_31_0.png)]
2.调用方模型进行预测
model_lstm.load_weights('best_model.h5')
predicted_y_lstm=model_lstm.predict(X_test)
y_test_one=[i[0] for i in y_test]
predicted_y_lstm_one=[i[0] for i in predicted_y_lstm]
y_test_two=[i[1] for i in y_test]
predicted_y_lstm_two=[i[1] for i in predicted_y_lstm]
fig, ax =plt.subplots(1,2,constrained_layout=True, figsize=(14, 3))
#画出第9个时间段真实数据与预测数据的对比图
ax[0].plot(y_test_one[:1000],color='red',label='真实值')
ax[0].plot(predicted_y_lstm_one[:1000],color='blue',label='预测值')
#画出第10个时间段真实数据与预测数据的对比图
ax[1].plot(y_test_two[:1000],color='red',label='真实值')
ax[1].plot(predicted_y_lstm_two[:1000],color='blue',label='预测值')
ax[0].set(xlabel='X',ylabel='Y',title='第9个时间段')
ax[1].set(xlabel='X',ylabel='Y',title='第10个时间段')
[Text(0.5, 0, 'X'), Text(0, 0.5, 'Y'), Text(0.5, 1.0, '第10个时间段')]
[外链图片转存失败,源站可能有防盗链机制,建议将图片保存下来直接上传(img-fg11dSek-1665993981351)(output_34_1.png)]
from sklearn import metrics
"""
RMSE ：均方根误差 -----> 对均方误差开方
R2 ：决定系数，可以简单理解为反映模型拟合优度的重要的统计量
"""
RMSE_lstm = metrics.mean_squared_error(predicted_y_lstm, y_test)**0.5
R2_lstm = metrics.r2_score(predicted_y_lstm, y_test)
print('均方根误差: %.5f' % RMSE_lstm)
print('R2: %.5f' % R2_lstm)
均方根误差: 6.58734
R2: 0.85471
1万+
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
