-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathTE_FinalModel6.1_LSTM+NN.py
104 lines (89 loc) · 4.04 KB
/
TE_FinalModel6.1_LSTM+NN.py
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
# ********************* Final Model -- 6.LSTM&DAE降维+NN分类 **********************
"""
调试记录：
1. 不能在一张计算图上定义两个basic_LSTM_cell！！！刚开始代码采用一张默认计算图来
训练两个故障提取模型，但是一直报错；后来将正常样本特征提取模型、故障样本特征提取模型
分别绘制在两张计算图上进行训练，便解决了这个问题。
"""
# %% 导入包
import numpy as np
import matplotlib.pyplot as plt
import tensorflow as tf
from sklearn import preprocessing, manifold
import TE_function as TE
# %% 准备数据集
fault_classes = 21, # 输入故障类别
train_x, train_y = TE.get_data('train', fault_classes)
test_x, test_y = TE.get_data('test', fault_classes)
normalize = preprocessing.MinMaxScaler() # 特征归一化
train_x = normalize.fit_transform(train_x)
test_x = normalize.transform(test_x)
# 分配数据集
train_normal_x = train_x[:500, :]
train_fault_x = train_x[500:980, :]
test_normal_x = test_x[:160, :]
test_fault_x = test_x[160:960, :]
# %% 定义正常样本特征提取模型
# 注意，LSTM的训练结构要求：batch_size必须能够被time_step整除
NormalModel = TE.LSTM_DAE(input_num=52, hidden_num=20, output_num=52,
time_step=10, batch_size=100, learning_rate=0.05,
lamda=0.001, dropout_keep_prob=0.8, epochs=1000)
# 创建一张新的计算图
g1 = tf.Graph()
with g1.as_default():
# 绘制计算图
NormalModel.draw_ComputeMap()
# 模型保存路径
model_path1 = "D:\\Python Codes\\TE\\TE_SaveLSTM\\Feature/SaveLSTM_Normal.ckpt"
# 训练
NormalModel.train(train_normal_x, model_path1)
# 获取特征
train_normal_feature = NormalModel.get_feature(train_normal_x)
test_normal_feature = NormalModel.get_feature(test_normal_x)
# %% 定义故障样本特征提取模型
FaultModel = TE.LSTM_DAE(input_num=52, hidden_num=20, output_num=52,
time_step=10, batch_size=100, learning_rate=0.05,
lamda=0.001, dropout_keep_prob=0.8, epochs=1000)
# 创建一张新的计算图
g2 = tf.Graph()
with g2.as_default():
# 绘制计算图
FaultModel.draw_ComputeMap()
# 模型保存路径
model_path2 = "D:\\Python Codes\\TE\\TE_SaveLSTM\\Fault/SaveLSTM_Fault.ckpt"
# 训练
FaultModel.train(train_fault_x, model_path2)
# 获取特征
train_fault_feature = FaultModel.get_feature(train_fault_x)
test_fault_feature = FaultModel.get_feature(test_fault_x)
# %% 整合提取到的特征（正常特征、故障特征），并2D可视化
train_feature = np.row_stack((train_normal_feature, train_fault_feature))
test_feature = np.row_stack((test_normal_feature, test_fault_feature))
# 将训练集随机打乱（测试集保持原来的顺序，不打乱）
train_data = np.column_stack((train_feature, train_y))
np.random.shuffle(train_data)
train_feature = train_data[:, :20]
train_y = train_data[:, 20:22]
# t-SNE二维可视化高维特征
print('\n训练集t-SNE拟合... ...')
train_tsne = manifold.TSNE(n_components=2).fit_transform(train_feature)
title = 'TrainSet: Feature Visulization of Fault ' + str(fault_classes[0])
TE.visualize_feature(train_tsne, train_y, title)
print('测试集t-SNE拟合... ...\n')
test_tsne = manifold.TSNE(n_components=2).fit_transform(test_feature)
title = 'TestSet: Feature Visulization of Fault ' + str(fault_classes[0])
TE.visualize_feature(test_tsne, test_y, title)
# %% 训练分类器
ClassifyModel = TE.NeuralNetwork(input_num=20, hidden_num=10, output_num=2,
learning_rate=0.1, epochs=300, decay_steps=100, decay_rate=0.6,
lamda=0, batch_size=256)
# 模型保存路径
model_path3 = "D:\\Python Codes\\TE\\TE_SaveLSTM\\Classification/SaveLSTM_NeuralNetwork.ckpt"
# 训练
ClassifyModel.train(train_feature, train_y, model_path3)
# 测试
ClassifyModel.test(test_feature, test_y)
# %% 测试集的故障实时监控图
test_pred = ClassifyModel.predict(test_feature)
test_fault_prob = test_pred[:, 1]
TE.test_plot(test_fault_prob)