本Python笔记本显示和分析了如何处理NASA获得的电池充电/放电数据集。
对于这个模型的训练阶段,需要安装Python 3.x以及以下库:
Tensorflow 2.0
Numpy
Pandas
Scipy
Sci-kit learn
Matplot
Seaborn
对于该模型的预测阶段,除了Matplot和Seaborn之外,需要使用相同的库。
1.数据集的准备
需要下载数据集,然后将其解压缩到特定的目录中。
%tensorflow_version 2.x
%matplotlib inline
!pip show tensorflow
!wget -cq https://ti.arc.nasa.gov/c/5 -O naza.zip
!unzip -qqo naza.zip -d battery_data
在此部分中,所有处理数据集所需的库都很重要。
import datetime
import numpy as np
import pandas as pd
from scipy.io import loadmat
from sklearn.preprocessing import MinMaxScaler
from sklearn.metrics import mean_squared_error
from sklearn import metrics
import matplotlib.pyplot as plt
import seaborn as sns
2.将数据集加载到内存
数据存储在多个“.mat”文件中。每个文件对应于特定的电池,每个文件的数据结构如下:
在Python中创建了一个函数,负责从"mat"文件中读取这些数据,并将它们存储在内存中以供以后访问,加载数据集后,使用panda函数对数据进行描述,以验证数据加载是否正确。
def load_data(battery):
mat = loadmat('battery_data/' + battery + '.mat')
print('Total data in dataset: ', len(mat[battery][0, 0]['cycle'][0]))
counter = 0
dataset = []
capacity_data = []
for i in range(len(mat[battery][0, 0]['cycle'][0])):
row = mat[battery][0, 0]['cycle'][0, i]
if row['type'][0] == 'discharge':
ambient_temperature = row['ambient_temperature'][0][0]
date_time = datetime.datetime(int(row['time'][0][0]),
int(row['time'][0][1]),
int(row['time'][0][2]),
int(row['time'][0][3]),
int(row['time'][0][4])) + datetime.timedelta(seconds=int(row['time'][0][5]))
data = row['data']
capacity = data[0][0]['Capacity'][0][0]
for j in range(len(data[0][0]['Voltage_measured'][0])):
voltage_measured = data[0][0]['Voltage_measured'][0][j]
current_measured = data[0][0]['Current_measured'][0][j]
temperature_measured = data[0][0]['Temperature_measured'][0][j]
current_load = data[0][0]['Current_load'][0][j]
voltage_load = data[0][0]['Voltage_load'][0][j]
time = data[0][0]['Time'][0][j]
dataset.append([counter + 1, ambient_temperature, date_time, capacity,
voltage_measured, current_measured,
temperature_measured, current_load,
voltage_load, time])
capacity_data.append([counter + 1, ambient_temperature, date_time, capacity])
counter = counter + 1
print(dataset[0])
return [pd.DataFrame(data=dataset,
columns=['cycle', 'ambient_temperature', 'datetime',
'capacity', 'voltage_measured',
'current_measured', 'temperature_measured',
'current_load', 'voltage_load', 'time']),
pd.DataFrame(data=capacity_data,
columns=['cycle', 'ambient_temperature', 'datetime',
'capacity'])]
dataset, capacity = load_data('B0005')
pd.set_option('display.max_columns', 10)
print(dataset.head())
dataset.describe()
下图显示了随着充电周期的推进,电池的老化过程。水平线表示与电池生命周期结束相关的阈值。
plot_df = capacity.loc[(capacity['cycle']>=1),['cycle','capacity']]
sns.set_style("darkgrid")
plt.figure(figsize=(12, 8))
plt.plot(plot_df['cycle'], plot_df['capacity'])
#Draw threshold
plt.plot([0.,len(capacity)], [1.4, 1.4])
plt.ylabel('Capacity')
# make x-axis ticks legible
adf = plt.gca().get_xaxis().get_major_formatter()
plt.xlabel('cycle')
plt.title('Discharge B0005')
还需计算电池的SOH值:
attrib=['cycle', 'datetime', 'capacity']
dis_ele = capacity[attrib]
C = dis_ele['capacity'][0]
for i in range(len(dis_ele)):
dis_ele['SoH']=(dis_ele['capacity'])/C
print(dis_ele.head(5))
和以前所作的一样,每个周期都绘制一个SOH图表,水平线代表70%的阈值,即电池已经达到其使用寿命,因此建议进行更换。
plot_df = dis_ele.loc[(dis_ele['cycle']>=1),['cycle','SoH']]
sns.set_style("white")
plt.figure(figsize=(8, 5))
plt.plot(plot_df['cycle'], plot_df['SoH'])
#Draw threshold
plt.plot([0.,len(capacity)], [0.70, 0.70])
plt.ylabel('SOH')
# make x-axis ticks legible
adf = plt.gca().get_xaxis().get_major_formatter()
plt.xlabel('cycle')
plt.title('Discharge B0005')
3.SOH计算的训练阶段
准备了数据集,以便Tensorflow可以在训练阶段使用,为此创建两个结构,对应于预期的输入和输出。数据集的相关特征是:
电池容量、电压、电流、温度、负载电压、负载电流、时间。
对于输出数据,计算电池的SOH,以及在两种情况下的输入和输出,这些值被归一化到[0-1]之间的值。
C = dataset['capacity'][0]
soh = []
for i in range(len(dataset)):
soh.append([dataset['capacity'][i] / C])
soh = pd.DataFrame(data=soh, columns=['SoH'])
attribs=['capacity', 'voltage_measured', 'current_measured',
'temperature_measured', 'current_load', 'voltage_load', 'time']
train_dataset = dataset[attribs]
sc = MinMaxScaler(feature_range=(0,1))
train_dataset = sc.fit_transform(train_dataset)
print(train_dataset.shape)
print(soh.shape)
import tensorflow as tf
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense
from tensorflow.keras.layers import Dropout
from tensorflow.keras.layers import Flatten
from tensorflow.keras.layers import LSTM
from tensorflow.keras.optimizers import Adam
总训练参数:27;
可训练参数:27。
对该模型进行训练,epoch=50;
model.fit(x=train_dataset, y=soh.to_numpy(), batch_size=25, epochs=50)
第二节传送门:
深度学习模型的准备和使用教程,LSTM用于锂电池SOH预测(第二节)(附Python的jypter源代码)_新能源姥大的博客-CSDN博客