Pytorch实现：LSTM-火灾温度预测

骑猪玩狗

已于 2024-04-16 20:27:47 修改

阅读量195

点赞数 2

文章标签： pytorch lstm 人工智能

于 2024-04-16 20:15:17 首次发布

本文链接：https://blog.csdn.net/a536723241/article/details/137799486

版权

🍨 本文为🔗365天深度学习训练营中的学习记录博客
🍖 原作者：K同学啊

前期工作

语言环境：Python3.9.18
编译器：Jupyter Lab
深度学习环境：Pytorch 1.12.1

1.设置GPU

import torch
import torch.nn as nn
import torchvision
from torchvision import transforms, datasets

import os,PIL,pathlib

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

device

2. 导入数据

df = pd.read_csv("F:/365data/R2/woodpine2.csv")
df.head()

3. 数据可视化

import matplotlib.pyplot as plt
import seaborn as sns

plt.rcParams['savefig.dpi'] = 500
plt.rcParams['figure.dpi'] = 500

fig ,ax = plt.subplots(1,3,constrained_layout=True,figsize=(14,3))

sns.lineplot(data=df['Tem1'],ax=ax[0])
sns.lineplot(data=df['CO 1'],ax=ax[1])
sns.lineplot(data=df['Soot 1'],ax=ax[2])
plt.show()

data = df.iloc[:,1:]
data.head()

二、构建数据集

1、数据集预处理

width_X= 8 # 时间步，时间步特征为3
width_y = 1

X = []
y = []

in_start = 0

for _, _ in data.iterrows(): # 用于遍历data的每一行，返回的是每一行的index和每一行的值
    in_end = in_start + width_X
    out_end = in_end + width_y
    
    if out_end <= len(data):
        X_ = np.array(data.iloc[in_start:in_end,])
        X_ = X_.reshape((len(X_)*3))
        y_ = np.array(data.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

from sklearn.preprocessing import MinMaxScaler

# 归一化
sc = MinMaxScaler(feature_range=(0,1))
X_scaled = sc.fit_transform(X)

X_scaled = X_scaled.reshape((X_scaled.shape[0],width_X,3))
X_scaled.shape

3、划分数据集

# X_train = X_scaled[:5000].astype('float64') 直接修改原始数据
X_train = np.array(X_scaled[:5000]).astype('float64') # 切片创建新数组，不会修改原始数据
y_train = np.array(y[:5000]).astype('float64')

X_test = np.array(X_scaled[5000:]).astype('float64')
y_test = np.array(y[5000:]).astype('float64')

X_train.shape,y_train.shape,X_test.shape,y_test.shape

# Create datasets and data loaders
train_dataset = TensorDataset(X_train_tensor, y_train_tensor)
test_dataset = TensorDataset(X_test_tensor, y_test_tensor)

train_loader = DataLoader(dataset=train_dataset, batch_size=64, shuffle=True)
test_loader = DataLoader(dataset=test_dataset, batch_size=64, shuffle=False)

三、模型训练

1、构建模型

import torch.nn.functional as F
# 使用pytorch构建简单的LSTM模型
class LSTM(nn.Module): # 通常输入为3维张量，第一维为序列长度，第二维为批次大小，第三维为输入大小,这样的数据方式能并行过LSTM，从而一定程度上提高处理速度
    def __init__(self,input_size,hidden_size): # hidden_size为隐层的神经元个数
        super(LSTM,self).__init__()
        self.LSTM1 = nn.LSTM(input_size, hidden_size)# 总是返回所有时间步输出
        self.LSTM2 = nn.LSTM(hidden_size, hidden_size)
        self.fc = nn.Linear(64, 1)

    def forward(self,x):
        out,_ = self.LSTM1(x)
        out,_ = self.LSTM2(out)
        out = self.fc(out[:,-1,:]) # 最后一个时间步结果的集合
        return out # 一批批输出，最后为一个元组
    
model = LSTM(input_size=3,hidden_size=64).to(device)
print(model)

def train(dataloader,model,optimizer,loss_fn):
    num_batches = len(dataloader)

    train_loss = 0

    for X,y in dataloader:
        X,y = X.to(device),y.to(device) #[batch_size,seq_len,input_size]
        # 即使与LSTM的输入维度不同，也可以直接输入,pytorch会自动调整
        # 但是其batch_first参数的设置会影响LSTM层内部数据的处理方式，从而影响模型的训练效果
        # batch_first=True时，可能会导致LSTM在处理长序列时更有效
        # batch_first=False时，可能会导致LSTM在处理短序列时更有效

        pred = model(X)
        loss = loss_fn(pred,y)

        optimizer.zero_grad()
        loss.backward()
        # 梯度裁剪
        torch.nn.utils.clip_grad_norm_(model.parameters(), max_norm=1.68)
        optimizer.step()

        train_loss += loss.item()

    train_loss /= num_batches

    return train_loss

def test (dataloader, model, loss_fn):
    num_batches = len(dataloader)          # 批次数目, (size/batch_size，向上取整)
    test_loss= 0
    
    # 当不进行训练时，停止梯度更新，节省计算内存消耗
    with torch.no_grad():
        for imgs, target in dataloader:
            imgs, target = imgs.to(device), target.to(device)
            
            # 计算loss
            target_pred = model(imgs)
            loss        = loss_fn(target_pred, target)
            
            test_loss += loss.item()

    test_loss /= num_batches

    return test_loss

4、正式训练模型

# Define the loss function and optimizer
loss_fn = nn.MSELoss()
learn_rate = 1e-3
optimizer = torch.optim.Adam(model.parameters(),lr=learn_rate)

import copy 

epochs = 200

train_loss=[]
test_loss=[]

for epoch in range(epochs):

    model.train()
    epoch_train_loss = train(train_loader,model,optimizer,loss_fn)

    model.eval()
    epoch_test_loss = test(test_loader,model,loss_fn)

    train_loss.append(epoch_train_loss)
    test_loss.append(epoch_test_loss)

    lr = optimizer.state_dict()['param_groups'][0]['lr']

    template = ('Epoch:{:2d}, Train_loss:{:.3f}, Test_loss:{:.3f}, Lr:{:.2E}')
    print(template.format(epoch+1, epoch_train_loss, epoch_test_loss, lr))


print('Done')

Epoch: 1, Train_loss:20635.831, Test_loss:66297.108, Lr:1.00E-03
Epoch: 2, Train_loss:18829.208, Test_loss:63532.933, Lr:1.00E-03
Epoch: 3, Train_loss:17670.595, Test_loss:60912.115, Lr:1.00E-03
Epoch: 4, Train_loss:16463.469, Test_loss:58382.553, Lr:1.00E-03
Epoch: 5, Train_loss:15398.968, Test_loss:55925.933, Lr:1.00E-03
Epoch: 6, Train_loss:14261.022, Test_loss:53533.631, Lr:1.00E-03
Epoch: 7, Train_loss:13231.828, Test_loss:51205.521, Lr:1.00E-03
Epoch: 8, Train_loss:12314.483, Test_loss:48937.172, Lr:1.00E-03
Epoch: 9, Train_loss:11434.403, Test_loss:46726.929, Lr:1.00E-03
Epoch:10, Train_loss:10620.793, Test_loss:44571.675, Lr:1.00E-03
Epoch:11, Train_loss:9726.007, Test_loss:42470.488, Lr:1.00E-03
Epoch:12, Train_loss:9044.311, Test_loss:40422.014, Lr:1.00E-03
Epoch:13, Train_loss:8366.348, Test_loss:38425.752, Lr:1.00E-03
Epoch:14, Train_loss:7727.216, Test_loss:36481.379, Lr:1.00E-03
Epoch:15, Train_loss:7245.731, Test_loss:34473.627, Lr:1.00E-03
Epoch:16, Train_loss:6608.261, Test_loss:32571.641, Lr:1.00E-03
Epoch:17, Train_loss:6160.780, Test_loss:30739.927, Lr:1.00E-03
Epoch:18, Train_loss:5720.138, Test_loss:28968.774, Lr:1.00E-03
Epoch:19, Train_loss:5338.054, Test_loss:27254.684, Lr:1.00E-03
Epoch:20, Train_loss:5041.359, Test_loss:25596.053, Lr:1.00E-03
Epoch:21, Train_loss:4758.801, Test_loss:23995.714, Lr:1.00E-03
Epoch:22, Train_loss:4538.454, Test_loss:22509.405, Lr:1.00E-03
Epoch:23, Train_loss:4429.614, Test_loss:21245.318, Lr:1.00E-03
Epoch:24, Train_loss:4300.475, Test_loss:20142.770, Lr:1.00E-03
Epoch:25, Train_loss:4248.188, Test_loss:19199.139, Lr:1.00E-03
Epoch:26, Train_loss:4179.910, Test_loss:19461.267, Lr:1.00E-03
Epoch:27, Train_loss:2634.756, Test_loss:17759.148, Lr:1.00E-03
Epoch:28, Train_loss:1996.072, Test_loss:16587.582, Lr:1.00E-03
Epoch:29, Train_loss:1746.003, Test_loss:15413.238, Lr:1.00E-03
Epoch:30, Train_loss:1531.197, Test_loss:14248.107, Lr:1.00E-03
Epoch:31, Train_loss:1332.269, Test_loss:13128.375, Lr:1.00E-03
Epoch:32, Train_loss:1143.767, Test_loss:12075.063, Lr:1.00E-03
Epoch:33, Train_loss:1002.286, Test_loss:11076.224, Lr:1.00E-03
Epoch:34, Train_loss:850.997, Test_loss:10158.073, Lr:1.00E-03
Epoch:35, Train_loss:718.919, Test_loss:9335.331, Lr:1.00E-03
Epoch:36, Train_loss:601.860, Test_loss:8519.326, Lr:1.00E-03
Epoch:37, Train_loss:511.991, Test_loss:7720.966, Lr:1.00E-03
Epoch:38, Train_loss:428.137, Test_loss:7051.985, Lr:1.00E-03
Epoch:39, Train_loss:356.029, Test_loss:6445.943, Lr:1.00E-03
Epoch:40, Train_loss:304.774, Test_loss:5913.033, Lr:1.00E-03
Epoch:41, Train_loss:253.031, Test_loss:5382.602, Lr:1.00E-03
Epoch:42, Train_loss:205.955, Test_loss:4926.450, Lr:1.00E-03
Epoch:43, Train_loss:179.406, Test_loss:4532.605, Lr:1.00E-03
Epoch:44, Train_loss:149.479, Test_loss:4186.195, Lr:1.00E-03
Epoch:45, Train_loss:130.584, Test_loss:3811.634, Lr:1.00E-03
Epoch:46, Train_loss:106.846, Test_loss:3508.918, Lr:1.00E-03
Epoch:47, Train_loss:85.672, Test_loss:3221.941, Lr:1.00E-03
Epoch:48, Train_loss:70.628, Test_loss:2974.503, Lr:1.00E-03
Epoch:49, Train_loss:62.967, Test_loss:2765.389, Lr:1.00E-03
Epoch:50, Train_loss:54.679, Test_loss:2614.472, Lr:1.00E-03
Epoch:51, Train_loss:46.520, Test_loss:2457.512, Lr:1.00E-03
Epoch:52, Train_loss:46.700, Test_loss:2330.749, Lr:1.00E-03
Epoch:53, Train_loss:42.089, Test_loss:2187.785, Lr:1.00E-03
Epoch:54, Train_loss:39.379, Test_loss:2060.647, Lr:1.00E-03
Epoch:55, Train_loss:32.344, Test_loss:1976.932, Lr:1.00E-03
Epoch:56, Train_loss:28.527, Test_loss:1879.812, Lr:1.00E-03
Epoch:57, Train_loss:30.393, Test_loss:1791.903, Lr:1.00E-03
Epoch:58, Train_loss:19.974, Test_loss:1697.742, Lr:1.00E-03
Epoch:59, Train_loss:19.505, Test_loss:1627.917, Lr:1.00E-03
Epoch:60, Train_loss:24.052, Test_loss:1549.823, Lr:1.00E-03
Epoch:61, Train_loss:17.172, Test_loss:1471.113, Lr:1.00E-03
Epoch:62, Train_loss:22.835, Test_loss:1411.666, Lr:1.00E-03
Epoch:63, Train_loss:17.834, Test_loss:1358.213, Lr:1.00E-03
Epoch:64, Train_loss:16.190, Test_loss:1306.348, Lr:1.00E-03
Epoch:65, Train_loss:10.257, Test_loss:1267.215, Lr:1.00E-03
Epoch:66, Train_loss:11.787, Test_loss:1220.384, Lr:1.00E-03
Epoch:67, Train_loss:11.928, Test_loss:1173.339, Lr:1.00E-03
Epoch:68, Train_loss:10.096, Test_loss:1140.230, Lr:1.00E-03
Epoch:69, Train_loss:12.373, Test_loss:1109.372, Lr:1.00E-03
Epoch:70, Train_loss:8.956, Test_loss:1082.618, Lr:1.00E-03
Epoch:71, Train_loss:10.484, Test_loss:1053.763, Lr:1.00E-03
Epoch:72, Train_loss:9.285, Test_loss:1028.577, Lr:1.00E-03
Epoch:73, Train_loss:10.079, Test_loss:1005.900, Lr:1.00E-03
Epoch:74, Train_loss:8.466, Test_loss:984.480, Lr:1.00E-03
Epoch:75, Train_loss:11.057, Test_loss:958.131, Lr:1.00E-03
Epoch:76, Train_loss:9.518, Test_loss:935.754, Lr:1.00E-03
Epoch:77, Train_loss:7.988, Test_loss:917.656, Lr:1.00E-03
Epoch:78, Train_loss:12.262, Test_loss:893.840, Lr:1.00E-03
Epoch:79, Train_loss:10.581, Test_loss:876.436, Lr:1.00E-03
Epoch:80, Train_loss:8.653, Test_loss:857.843, Lr:1.00E-03
Epoch:81, Train_loss:7.925, Test_loss:841.759, Lr:1.00E-03
Epoch:82, Train_loss:8.069, Test_loss:829.125, Lr:1.00E-03
Epoch:83, Train_loss:11.105, Test_loss:811.239, Lr:1.00E-03
Epoch:84, Train_loss:6.694, Test_loss:796.871, Lr:1.00E-03
Epoch:85, Train_loss:9.938, Test_loss:782.340, Lr:1.00E-03
Epoch:86, Train_loss:6.365, Test_loss:767.827, Lr:1.00E-03
Epoch:87, Train_loss:8.319, Test_loss:752.654, Lr:1.00E-03
Epoch:88, Train_loss:7.610, Test_loss:732.174, Lr:1.00E-03
Epoch:89, Train_loss:9.572, Test_loss:720.828, Lr:1.00E-03
Epoch:90, Train_loss:8.744, Test_loss:701.906, Lr:1.00E-03
Epoch:91, Train_loss:8.735, Test_loss:679.722, Lr:1.00E-03
Epoch:92, Train_loss:6.629, Test_loss:674.427, Lr:1.00E-03
Epoch:93, Train_loss:7.194, Test_loss:656.849, Lr:1.00E-03
Epoch:94, Train_loss:6.989, Test_loss:636.975, Lr:1.00E-03
Epoch:95, Train_loss:7.029, Test_loss:620.471, Lr:1.00E-03
Epoch:96, Train_loss:8.883, Test_loss:607.685, Lr:1.00E-03
Epoch:97, Train_loss:9.923, Test_loss:581.866, Lr:1.00E-03
Epoch:98, Train_loss:6.538, Test_loss:572.428, Lr:1.00E-03
Epoch:99, Train_loss:8.349, Test_loss:554.513, Lr:1.00E-03
Epoch:100, Train_loss:7.702, Test_loss:544.101, Lr:1.00E-03
Epoch:101, Train_loss:5.517, Test_loss:519.929, Lr:1.00E-03
Epoch:102, Train_loss:5.773, Test_loss:512.505, Lr:1.00E-03
Epoch:103, Train_loss:6.067, Test_loss:495.514, Lr:1.00E-03
Epoch:104, Train_loss:4.833, Test_loss:491.224, Lr:1.00E-03
Epoch:105, Train_loss:7.961, Test_loss:473.603, Lr:1.00E-03
Epoch:106, Train_loss:5.927, Test_loss:452.194, Lr:1.00E-03
Epoch:107, Train_loss:5.463, Test_loss:442.340, Lr:1.00E-03
Epoch:108, Train_loss:5.609, Test_loss:417.340, Lr:1.00E-03
Epoch:109, Train_loss:9.296, Test_loss:395.328, Lr:1.00E-03
Epoch:110, Train_loss:5.264, Test_loss:382.694, Lr:1.00E-03
Epoch:111, Train_loss:4.928, Test_loss:377.233, Lr:1.00E-03
Epoch:112, Train_loss:4.505, Test_loss:376.018, Lr:1.00E-03
Epoch:113, Train_loss:4.843, Test_loss:365.014, Lr:1.00E-03
Epoch:114, Train_loss:4.353, Test_loss:355.474, Lr:1.00E-03
Epoch:115, Train_loss:4.379, Test_loss:331.388, Lr:1.00E-03
Epoch:116, Train_loss:4.028, Test_loss:326.193, Lr:1.00E-03
Epoch:117, Train_loss:3.886, Test_loss:320.433, Lr:1.00E-03
Epoch:118, Train_loss:4.402, Test_loss:317.491, Lr:1.00E-03
Epoch:119, Train_loss:5.047, Test_loss:311.838, Lr:1.00E-03
Epoch:120, Train_loss:5.796, Test_loss:288.153, Lr:1.00E-03
Epoch:121, Train_loss:4.249, Test_loss:294.147, Lr:1.00E-03
Epoch:122, Train_loss:4.291, Test_loss:271.253, Lr:1.00E-03
Epoch:123, Train_loss:3.870, Test_loss:259.648, Lr:1.00E-03
Epoch:124, Train_loss:5.571, Test_loss:255.828, Lr:1.00E-03
Epoch:125, Train_loss:7.542, Test_loss:249.424, Lr:1.00E-03
Epoch:126, Train_loss:4.281, Test_loss:228.701, Lr:1.00E-03
Epoch:127, Train_loss:3.593, Test_loss:228.517, Lr:1.00E-03
Epoch:128, Train_loss:4.849, Test_loss:233.302, Lr:1.00E-03
Epoch:129, Train_loss:2.678, Test_loss:214.361, Lr:1.00E-03
Epoch:130, Train_loss:3.843, Test_loss:229.420, Lr:1.00E-03
Epoch:131, Train_loss:5.235, Test_loss:227.116, Lr:1.00E-03
Epoch:132, Train_loss:3.353, Test_loss:210.958, Lr:1.00E-03
Epoch:133, Train_loss:3.567, Test_loss:201.080, Lr:1.00E-03
Epoch:134, Train_loss:3.009, Test_loss:200.930, Lr:1.00E-03
Epoch:135, Train_loss:2.630, Test_loss:216.823, Lr:1.00E-03
Epoch:136, Train_loss:4.214, Test_loss:221.735, Lr:1.00E-03
Epoch:137, Train_loss:3.353, Test_loss:186.374, Lr:1.00E-03
Epoch:138, Train_loss:2.607, Test_loss:214.251, Lr:1.00E-03
Epoch:139, Train_loss:2.212, Test_loss:196.602, Lr:1.00E-03
Epoch:140, Train_loss:3.035, Test_loss:186.465, Lr:1.00E-03
Epoch:141, Train_loss:2.845, Test_loss:210.534, Lr:1.00E-03
Epoch:142, Train_loss:3.348, Test_loss:199.425, Lr:1.00E-03
Epoch:143, Train_loss:2.547, Test_loss:201.113, Lr:1.00E-03
Epoch:144, Train_loss:1.970, Test_loss:193.439, Lr:1.00E-03
Epoch:145, Train_loss:3.030, Test_loss:191.147, Lr:1.00E-03
Epoch:146, Train_loss:2.216, Test_loss:192.023, Lr:1.00E-03
Epoch:147, Train_loss:2.169, Test_loss:192.685, Lr:1.00E-03
Epoch:148, Train_loss:1.994, Test_loss:172.968, Lr:1.00E-03
Epoch:149, Train_loss:2.770, Test_loss:167.034, Lr:1.00E-03
Epoch:150, Train_loss:2.130, Test_loss:176.286, Lr:1.00E-03
Epoch:151, Train_loss:2.506, Test_loss:172.341, Lr:1.00E-03
Epoch:152, Train_loss:1.665, Test_loss:180.329, Lr:1.00E-03
Epoch:153, Train_loss:1.732, Test_loss:163.748, Lr:1.00E-03
Epoch:154, Train_loss:2.366, Test_loss:176.554, Lr:1.00E-03
Epoch:155, Train_loss:1.583, Test_loss:187.248, Lr:1.00E-03
Epoch:156, Train_loss:2.278, Test_loss:172.116, Lr:1.00E-03
Epoch:157, Train_loss:2.383, Test_loss:173.419, Lr:1.00E-03
Epoch:158, Train_loss:1.524, Test_loss:152.646, Lr:1.00E-03
Epoch:159, Train_loss:1.687, Test_loss:169.870, Lr:1.00E-03
Epoch:160, Train_loss:1.767, Test_loss:165.409, Lr:1.00E-03
Epoch:161, Train_loss:1.922, Test_loss:179.328, Lr:1.00E-03
Epoch:162, Train_loss:1.458, Test_loss:167.239, Lr:1.00E-03
Epoch:163, Train_loss:1.961, Test_loss:141.942, Lr:1.00E-03
Epoch:164, Train_loss:1.508, Test_loss:153.061, Lr:1.00E-03
Epoch:165, Train_loss:1.582, Test_loss:146.338, Lr:1.00E-03
Epoch:166, Train_loss:1.552, Test_loss:135.282, Lr:1.00E-03
Epoch:167, Train_loss:1.197, Test_loss:142.602, Lr:1.00E-03
Epoch:168, Train_loss:1.311, Test_loss:156.572, Lr:1.00E-03
Epoch:169, Train_loss:1.642, Test_loss:144.998, Lr:1.00E-03
Epoch:170, Train_loss:1.879, Test_loss:134.414, Lr:1.00E-03
Epoch:171, Train_loss:2.269, Test_loss:138.463, Lr:1.00E-03
Epoch:172, Train_loss:1.781, Test_loss:136.314, Lr:1.00E-03
Epoch:173, Train_loss:1.795, Test_loss:135.326, Lr:1.00E-03
Epoch:174, Train_loss:1.487, Test_loss:122.718, Lr:1.00E-03
Epoch:175, Train_loss:1.544, Test_loss:155.864, Lr:1.00E-03
Epoch:176, Train_loss:1.739, Test_loss:130.433, Lr:1.00E-03
Epoch:177, Train_loss:1.075, Test_loss:144.390, Lr:1.00E-03
Epoch:178, Train_loss:1.438, Test_loss:147.833, Lr:1.00E-03
Epoch:179, Train_loss:1.276, Test_loss:116.244, Lr:1.00E-03
Epoch:180, Train_loss:1.161, Test_loss:114.795, Lr:1.00E-03
Epoch:181, Train_loss:1.565, Test_loss:125.655, Lr:1.00E-03
Epoch:182, Train_loss:1.339, Test_loss:111.128, Lr:1.00E-03
Epoch:183, Train_loss:1.753, Test_loss:113.491, Lr:1.00E-03
Epoch:184, Train_loss:1.373, Test_loss:120.987, Lr:1.00E-03
Epoch:185, Train_loss:1.343, Test_loss:113.305, Lr:1.00E-03
Epoch:186, Train_loss:0.947, Test_loss:105.335, Lr:1.00E-03
Epoch:187, Train_loss:1.363, Test_loss:92.125, Lr:1.00E-03
Epoch:188, Train_loss:1.680, Test_loss:103.017, Lr:1.00E-03
Epoch:189, Train_loss:0.978, Test_loss:96.335, Lr:1.00E-03
Epoch:190, Train_loss:0.987, Test_loss:94.008, Lr:1.00E-03
Epoch:191, Train_loss:0.949, Test_loss:119.769, Lr:1.00E-03
Epoch:192, Train_loss:1.196, Test_loss:82.924, Lr:1.00E-03
Epoch:193, Train_loss:0.978, Test_loss:90.593, Lr:1.00E-03
Epoch:194, Train_loss:1.298, Test_loss:92.257, Lr:1.00E-03
Epoch:195, Train_loss:0.925, Test_loss:99.024, Lr:1.00E-03
Epoch:196, Train_loss:1.341, Test_loss:94.766, Lr:1.00E-03
Epoch:197, Train_loss:1.194, Test_loss:88.215, Lr:1.00E-03
Epoch:198, Train_loss:1.547, Test_loss:98.698, Lr:1.00E-03
Epoch:199, Train_loss:0.989, Test_loss:93.121, Lr:1.00E-03
Epoch:200, Train_loss:0.902, Test_loss:80.574, Lr:1.00E-03
Done

四、模型评估

import matplotlib.pyplot as plt

plt.rcParams['font.sans-serif'] = ['SimHei']
plt.rcParams['axes.unicode_minus'] = False
plt.rcParams['figure.dpi'] = 100

epochs_range = range(epochs)
plt.plot(epochs_range,train_loss,label = 'Training Loss')
plt.plot(epochs_range,test_loss,label = 'Test Loss')
plt.legend(loc = 'upper right')
plt.title('Training and Test Loss')
plt.show()

在这里插入图片描述

调用模型进行预测

Xtest = X_test_tensor.to(device)
model.eval()
predicted_y_lstm = model(Xtest)
# 将预测结果和真实值都转移到 CPU 上，并转换为 numpy 数组
predicted_y_lstm = predicted_y_lstm.detach().cpu().numpy()
y_test_one = [i[0] for i in y_test_tensor]
predicted_y_lstm_one = [i[0] for i in predicted_y_lstm]
plt.figure(figsize=(5,3),dpi=120)
plt.plot(y_test_one,color='red',label='真实值')
plt.plot(predicted_y_lstm_one,color='blue',label='预测值')
plt.title('LSTM预测结果')
plt.xlabel('X')
plt.ylabel('Y')
plt.legend()
plt.show()

R2值评估

from sklearn import metrics
RESE_lstm = metrics.mean_squared_error(predicted_y_lstm_one,y_test_tensor)**0.5
R2_lstm = metrics.r2_score(predicted_y_lstm_one,y_test_tensor)

print('均方根误差:%.5f' % RESE_lstm)
print('R2:%.5f' % R2_lstm)

均方根误差:7.19652
R2:0.63034

总结

与RNN相比，LSTM的pytorch代码是将nn.RNN()变为了，nn.LSTM()，参数没有变化
从原理上讲，LSTM是RNN的进阶版，LSTM增加了多个状态
- 输入门输入靠近记忆单元的信息以及原始信息
- 遗忘门用来遗忘不需要的信息
- 输出门用来输出传递到下一单元的信息
LSTM相比其RNN可以记忆更长远的信息；并且可以有效解决梯度爆炸问题
我的输出模型预测结果在开始使较准确，到后面开始出现较大的偏差；比较K同学的代码之后，发现可能是模型的隐藏层大小设置的太小的缘故

骑猪玩狗

关注

2
点赞
踩
5

收藏

觉得还不错? 一键收藏
0
评论
Pytorch实现：LSTM-火灾温度预测

与RNN相比，LSTM的pytorch代码是将nn.RNN()变为了，nn.LSTM()，参数没有变化从原理上讲，LSTM是RNN的进阶版，LSTM增加了多个状态输入门输入靠近记忆单元的信息以及原始信息遗忘门用来遗忘不需要的信息输出门用来输出传递到下一单元的信息LSTM相比其RNN可以记忆更长远的信息；并且可以有效解决梯度爆炸问题我的输出模型预测结果在开始使较准确，到后面开始出现较大的偏差；比较K同学的代码之后，发现可能是模型的隐藏层大小设置的太小的缘故。
复制链接

扫一扫