def Persistence_Method(data):
per = data[['DateTime', 'WS95']].copy()
per['V1'] = per['WS95'].copy()
per['F1'] = per['WS95'].shift(1)
per['F2'] = per['WS95'].shift(2)
per['F3'] = per['WS95'].shift(3)
per2 = DataRange(per, "2019-01-01 00:00", "2020-01-01 00:00")
per2.to_csv('WS_PER_Result.csv')
def Train_Model(Train_data, LB, LF):
start_date1, stop_date1 = "2017-01-01 00:00", "2018-01-01 00:01"
df_train = DataRange(Train_data, start_date1, stop_date1)
df_train.reset_index(inplace=True, drop=True)
x_train, y_train = shaping_one(df_train, LB, LF)
start_date2, stop_date2 = "2018-01-01 00:00", "2019-01-01 00:01"
df_test = DataRange(Train_data, start_date2, stop_date2)
df_test.reset_index(inplace=True, drop=True)
x_test, y_test = shaping_one(df_test, LB, LF)
dimension = x_train.shape[1]
timesteps = x_train.shape[2]
model = my_model(dimension, timesteps, output=int(len(LF)))
model.compile(loss='mean_squared_error', optimizer='adam')
model_ckpt = ModelCheckpoint(filepath="./tmp2.h5",
monitor="val_loss",
save_best_only=True)
history = model.fit(x_train,
y_train,
validation_data=(x_test, y_test),
batch_size=96,
epochs=50,
callbacks=[model_ckpt],
shuffle=True)
model = tf.keras.models.load_model("./tmp2.h5")
return model, history
def Step_Error_ndim(df,Start_date,Stop_date,i,variable='WS'):
df.fillna(method='ffill',inplace=True)
df.reset_index(inplace=True,drop=True)
df['V1_{}'.format(variable)].replace(0,0.001,inplace=True)
data = DataRange(df,Start_date,Stop_date)
WS_v = data['V1_{}'.format(variable)].copy()
WS_f = data['F{}_{}'.format(i+1,variable)]
MAE = Mean_Absolute_Error(WS_v,WS_f)
RMSE = Root_Mean_Square_Error(WS_v,WS_f)
NMSE = Nor_Mean_Square_Error(WS_v,WS_f)
print("{},{},{},{}".format(i+1,MAE,RMSE,NMSE))
return MAE,RMSE,NMSE
def Step_Error(df,Start_date,Stop_date,i):
df.dropna(how='any',inplace=True)
df.reset_index(inplace=True,drop=True)
df['V1'].replace(0,0.001,inplace=True)
data = DataRange(df,Start_date,Stop_date)
WS_v = data['V1'].copy()
WS_f = data['F{}'.format(i+1)]
MAE = Mean_Absolute_Error(WS_v,WS_f)
RMSE = Root_Mean_Square_Error(WS_v,WS_f)
NMSE = Nor_Mean_Square_Error(WS_v,WS_f)
print("{},{},{},{}".format(i+1,MAE,RMSE,NMSE))
return MAE,RMSE,NMSE
def Plot_Step_Prediction(data, steps):
date01 = '2019-02-07'
date02 = '2019-02-08'
df = DataRange(data, "{} 18:00".format(date01), "{} 03:01".format(date02))
WS_v = df['V1'].copy()
time = df['DateTime'].copy()
u1 = time.iloc[0]
u2 = time.iloc[-1]
#naming = {1:'1st',2:'2nd',3:'3rd'}
fig, (ax) = plt.subplots(steps,
1,
sharex=True,
figsize=(12, 9),
gridspec_kw={
'wspace': 0,
'hspace': 0.05
})
for step in range(steps):
WS_f = df['F{}'.format(step + 1)].copy()
ax[step].plot(time,
WS_v,
label='Real',
marker='h',
color='black',
linewidth=2)
ax[step].plot(time,
WS_f,
label='Predict',
linewidth=1,
linestyle=':',
marker='D',
color='blue')
ax[step].legend(loc=2, fontsize=12)
ax[step].tick_params(axis="x", labelsize=10, rotation=0)
ax[step].tick_params(axis="y", labelsize=15)
ax[step].set_xlim([u1, u2])
ax[step].set_ylim([0, 26])
ax[step].set_yticks([0, 5, 10, 15, 20, 25])
ax[step].grid(True)
ax[0].set_title("Wind Speed Prediction ({})".format(date01), fontsize=20)
ax[step - 1].set_ylabel("Wind Speed (m/s)", fontsize=20)
ax[step].set_xlabel("Time Series (Time Step=10min)", fontsize=20)
ax[step].xaxis.set_major_locator(mdates.HourLocator())
ax[step].xaxis.set_major_formatter(mdates.DateFormatter('%H00'))
props = dict(boxstyle='round', facecolor='wheat', alpha=0.5)
fig.text(0.80,
0.656,
'1st-Step Prediction',
ha='center',
size=16,
bbox=props)
fig.text(0.80,
0.400,
'2nd-Step Prediction',
ha='center',
size=16,
bbox=props)
fig.text(0.80,
0.145,
'3rd-Step Prediction',
ha='center',
size=16,
bbox=props)