AndrewChung-GitHub / Alternator_analysis

依据发电机电流预测发电机故障以及故障时间

Alternator_analysis

This project is a commercial project v1,using generator current waveform to predict generator gault type and remaining working time.

​ The Network Architecture

part1. Paralycite

1. config.py

work_dir:   the work dir of this program
input_dim:  def the input sampling length of the sinusoid
hidden_dim: def the LSTM cell hidden dimension
model_dir:  def the dir to place the ckpt file
epoch:      def the training epoches
lables_num: def the num of error 
log_mode:   s:output message to screen f:output to file
log_dir:    

2. log.py

This file defines the format of the log, it can be shown on the screen and can also output to a file.
When output to file, just use the config --log_mode 'f'
the default --log_mode is 's'

3. lstm.py

The lstm network

4. train.py

定义网络输入：
inputs:[config.batchsize, None, config.input_dim]
lables:[config.batchsize, 2, config.labels_num]
sequence_length： 定义一个训练序列的的长度。
其中lables[:,0,:] 存储了故障类型的标签
lables[:,1,:]    存储了故障停用时间的标签

5. dataset.py

Network's training data input，just a interface, it defines the format of the net's input.
train_input():函数生成训练网络所需要的数据
test_input(): 函数生成测试网络所需要的数据

part2. requirements

tensorflow
numpy

part3. Running

$ python train.py
此函数何以直接运行，在网络中有50个样本用于训练，训练网络输出结果存放在log/train_log.txt

$ python test.py
此函数可以直接运行，网络中有10个测试样本，测试网络输出结果存放在log/test_log.txt

part4. Training Your Own Data

需要自己实现一个dataset的数据接口，其需继承自CNet_Input类。 在自定义类中，需要实现如下三个函数：

def train_input(self):
    返回值: train_sample, train_lable
    返回值格式: train_sample [config.batchsize,sequence_length config.input_dim]
                train_lable [config.batchsize, 2, config.lables_num]
    返回格式说明: sequence_length: 一个序列的长度，LSTM网络可以时序输入
                train_lable[:,0,:]:存储发电机故障类型，one-hot编码
                train_lable[:,1,:]:发电机故障类型对应位置，存储的其因故障而停用时间。
def val_input(self):
    同上
def test_input(self):
    同上

还需在类内定义三个变量:

train_sample_num:  训练集样本数量
test_sample_num:   验证集样本数量
eval_sample_num:   测试集样本数量