Keras LSTM時間序列預測

使用Keras LSTM預測時間序列

參考文章：

Kesci： Keras 實現 LSTM——時間序列預測：https://www.cnblogs.com/mtcnn/p/9411597.html

讀取數據

data_path = "/mnt/X500/farmers/tongyao/機器學習項目練習/industry_timeseries/"

#查看其中一個地區的訓練數據
import pandas as pd
import numpy as np
from keras.models import Sequential
from keras.layers import Dense, LSTM, Dropout, Flatten
import matplotlib.pyplot as plt
import glob, os
import seaborn as sns
import sys
from sklearn.preprocessing import MinMaxScaler
%matplotlib inline

columns = ['YEAR','MONTH','DAY','TEMP_HIG','TEMP_COL','AVG_TEMP','AVG_WET','DATA_COL']
data = pd.read_csv(data_path + 'timeseries_train_data/2.csv', names=columns)
data.head()

	YEAR	MONTH	DAY	TEMP_HIG	TEMP_COL	AVG_TEMP	AVG_WET	DATA_COL
0	2015	2	1	1.9	-0.4	0.7875	75.000	160.477961
1	2015	2	2	6.2	-3.9	1.7625	77.250	129.268657
2	2015	2	3	7.8	2.0	4.2375	72.750	107.316539
3	2015	2	4	8.5	-1.2	3.0375	65.875	132.549075
4	2015	2	5	7.9	-3.6	1.8625	55.375	91.082841

數據預處理

# 數據預處理：將序列數據轉化爲監督問題數據
from pandas import DataFrame
from pandas import concat

def series_to_supervised(data, n_in=1, n_out=1, dropnan=True):
    n_vars = 1 if type(data) is list else data.shape[1]
    df = DataFrame(data)
    cols, names = [], []
    # i: n_in, n_in-1, ..., 1
    # 代表t-n_in, ... ,t-1
    for i in range(n_in, 0, -1):
        cols.append(df.shift(i))
        names += [('var%d(t-%d)' % (j+1, i)) for j in range(n_vars)]
    for i in range(0, n_out):
        cols.append(df.shift(-i))
        if i == 0:
            names += [('var%d(t)' % (j+1)) for j in range(n_vars)]
        else:
            names += [('var%d(t+%d)' % (j+1, i)) for j in range(n_vars)]
    agg = concat(cols, axis=1)
    agg.columns = names
    if dropnan:
        agg.dropna(inplace=True)
    return agg

# 將數據歸一化到0-1之間,無量綱化
scaler = MinMaxScaler(feature_range=(0,1))
scaled_data = scaler.fit_transform(data[['DATA_COL','TEMP_HIG','TEMP_COL','AVG_TEMP','AVG_WET']].values)

# 將時序數據轉換爲監督問題數據
reframed = series_to_supervised(scaled_data, 1, 1)

#刪除無用的label數據
reframed.drop(reframed.columns[[6,7,8,9]], axis=1, inplace=True)
print(reframed.info())
reframed.head()

運行結果

<class 'pandas.core.frame.DataFrame'>
Int64Index: 577 entries, 1 to 577
Data columns (total 6 columns):
var1(t-1)    577 non-null float64
var2(t-1)    577 non-null float64
var3(t-1)    577 non-null float64
var4(t-1)    577 non-null float64
var5(t-1)    577 non-null float64
var1(t)      577 non-null float64
dtypes: float64(6)
memory usage: 31.6 KB
None

	var1(t-1)	var2(t-1)	var3(t-1)	var4(t-1)	var5(t-1)	var1(t)
1	0.122335	0.049869	0.265823	0.212693	0.599165	0.096338
2	0.096338	0.162730	0.177215	0.236842	0.636743	0.078052
3	0.078052	0.204724	0.326582	0.298142	0.561587	0.099070
4	0.099070	0.223097	0.245570	0.268421	0.446764	0.064529
5	0.064529	0.207349	0.184810	0.239319	0.271399	0.085949

數據劃分

# 數據集劃分,選取前400天的數據作爲訓練集,中間150天作爲驗證集,其餘的作爲測試集
train_days = 400
valid_days = 150
values = reframed.values
train = values[:train_days, :]
valid = values[train_days:train_days+valid_days, :]
test = values[train_days+valid_days:, :]
train_X, train_y = train[:, :-1], train[:, -1]
valid_X, valid_y = valid[:, :-1], valid[:, -1]
test_X, test_y = test[:, :-1], test[:, -1]

# 將數據集重構爲符合LSTM要求的數據格式,即 [樣本，時間步，特徵]
train_X = train_X.reshape((train_X.shape[0], 1, train_X.shape[1]))
valid_X = valid_X.reshape((valid_X.shape[0], 1, valid_X.shape[1]))
test_X = test_X.reshape((test_X.shape[0], 1, test_X.shape[1]))
print(train_X.shape, train_y.shape, valid_X.shape, valid_y.shape, test_X.shape, test_y.shape)

(400, 1, 5) (400,) (150, 1, 5) (150,) (27, 1, 5) (27,)

訓練模型

使用Keras的Sequential搭建模型

model = Sequential()
model.add(LSTM(50, activation='relu',input_shape=(train_X.shape[1], train_X.shape[2]), return_sequences=True))
model.add(Flatten())
model.add(Dense(1, activation='linear'))
model.compile(loss='mean_squared_error', optimizer='adam') 
model.summary()

# fit network
hist = model.fit(train_X, train_y, epochs=50, batch_size=32, validation_data=(valid_X, valid_y), verbose=1, shuffle=False)

# plot history
plt.plot(hist.history['loss'], label='train')
plt.plot(hist.history['val_loss'], label='valid')
plt.legend()
plt.show()

輸出結果

_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
lstm_21 (LSTM)               (None, 1, 50)             11200     
_________________________________________________________________
flatten_3 (Flatten)          (None, 50)                0         
_________________________________________________________________
dense_19 (Dense)             (None, 1)                 51        
=================================================================
Total params: 11,251
Trainable params: 11,251
Non-trainable params: 0
_________________________________________________________________
Train on 400 samples, validate on 150 samples
Epoch 1/50
400/400 [==============================] - 3s 6ms/step - loss: 0.0325 - val_loss: 0.0644
Epoch 2/50
400/400 [==============================] - 0s 152us/step - loss: 0.0196 - val_loss: 0.0425
...
Epoch 50/50
400/400 [==============================] - 0s 176us/step - loss: 0.0019 - val_loss: 0.0031

看看預測結果如何

plt.figure(figsize=(24,8))
train_predict = model.predict(train_X)
valid_predict = model.predict(valid_X)
test_predict = model.predict(test_X)
plt.plot(values[:, -1], c='b', label="Train set (actual)")
plt.plot([x for x in train_predict], c='g', label="Train set (predict)")
plt.plot([None for _ in train_predict] + [x for x in valid_predict], c='y', label="Valid set (predict)")
plt.plot([None for _ in train_predict] + [None for _ in valid_predict] + [x for x in test_predict], c='r', label="Test set (predict)")
plt.legend(fontsize=16)
plt.show()

、