|
|
|
|
|
import keras
|
|
|
|
|
|
import matplotlib.pyplot as plt
|
|
|
|
|
|
import pandas as pd
|
|
|
|
|
|
import tensorflow as tf
|
|
|
|
|
|
import numpy as np
|
|
|
|
|
|
from sklearn.preprocessing import MinMaxScaler
|
|
|
|
|
|
from keras.layers import Dense, LSTM, Activation, Dropout
|
|
|
|
|
|
from keras.optimizers import SGD
|
|
|
|
|
|
|
|
|
|
|
|
colum = 13
|
|
|
|
|
|
step = 50
|
|
|
|
|
|
df = pd.read_csv('文件名')
|
|
|
|
|
|
df_for_training=df[:24000]
|
|
|
|
|
|
df_for_testing=df[24000:]
|
|
|
|
|
|
# print(df_for_testing.iloc[:, 2:])
|
|
|
|
|
|
|
|
|
|
|
|
scaler = MinMaxScaler(feature_range=(-1, 1))
|
|
|
|
|
|
df_for_training_scaled = scaler.fit_transform(df_for_training.iloc[:, 2:])
|
|
|
|
|
|
print("df_for_training_scaled shape:",df_for_training_scaled.shape)
|
|
|
|
|
|
#print(df_for_training_scaled.shape)
|
|
|
|
|
|
#print(df_for_testing_scaled.shape)
|
|
|
|
|
|
|
|
|
|
|
|
#抽取特征与标签列
|
|
|
|
|
|
def createXY(dataset, n_past):
|
|
|
|
|
|
dataX = []
|
|
|
|
|
|
dataY = []
|
|
|
|
|
|
for i in range(n_past, len(dataset)):
|
|
|
|
|
|
dataX.append(dataset[i - n_past:i, 1:dataset.shape[1]])
|
|
|
|
|
|
dataY.append(dataset[i, 0])
|
|
|
|
|
|
return np.array(dataX), np.array(dataY)
|
|
|
|
|
|
#trainX,testX数据的shape为[samples, steps, features]
|
|
|
|
|
|
trainX, trainY = createXY(df_for_training_scaled, step)
|
|
|
|
|
|
testX, testY = createXY(df_for_testing_scaled, step)
|
|
|
|
|
|
|
|
|
|
|
|
#构建模型
|
|
|
|
|
|
def build_model():
|
|
|
|
|
|
model = tf.keras.models.Sequential()
|
|
|
|
|
|
#经试验，如果下一层还是LSTM的话必须将return_sequences设置为True
|
|
|
|
|
|
model.add(LSTM(20, input_shape=(step, column), return_sequences=True))
|
|
|
|
|
|
model.add(LSTM(20))
|
|
|
|
|
|
model.add(Activation('relu'))
|
|
|
|
|
|
model.add(Dropout(0.2))
|
|
|
|
|
|
model.add(Dense(1))
|
|
|
|
|
|
model.compile(loss='mae', optimizer=SGD(lr=0.1,momentum=0.3))
|
|
|
|
|
|
return model
|
|
|
|
|
|
|
|
|
|
|
|
lstm_net = build_model()
|
|
|
|
|
|
lstm_net.fit(trainX, trainY, batch_size=8, epochs=100)
|
|
|
|
|
|
predictions = lstm_net.predict(testX)
|
|
|
|
|
|
|
|
|
|
|
|
#进行逆变换
|
|
|
|
|
|
test_Predict_copy = np.repeat(predictions,column+1,axis=-1)
|
|
|
|
|
|
test_Predict_inverse = scaler.inverse_transform(test_Predict_copy)[:,0]
|
|
|
|
|
|
#test_Predict_inverse = scaler.inverse_transform(test_Predict)
|
|
|
|
|
|
test_label = np.array(df_for_testing['omega_vsg_1'])[50:]
|
|
|
|
|
|
testY_copy = np.repeat(np.reshape(testY,(len(testY), 1)),column+1,axis=-1)
|
|
|
|
|
|
|
|
|
|
|
|
testY_inverse = scaler = scaler.inverse_transform(testY_copy)[:,0]
|
|
|
|
|
|
|
|
|
|
|
|
plt.plot(test_Predict_inverse,color='red')
|
|
|
|
|
|
plt.plot(test_label,color='green')
|
|
|
|
|
|
plt.plot(testY_inverse, color='blue')
|
|
|
|
|
|
plt.savefig('result_test.png')
|
