testing = data[testing_start:testing_end]
print(f'iteration {index}/{length}')
print('--------------------------------')
print()
# arima
print(
f'fitting ARIMA model for location {location} and month {month} at {datetime.now()}'
)
with warnings.catch_warnings():
warnings.filterwarnings(
'error',
message='divide by zero encountered in double_scalars')
try:
arima = ARIMAModel(scaling=True)
arima.fit(training,
order=order,
seasonal_order=seasonal_order,
filter=filter,
use_exogenous=True)
prediction = arima.predict(testing_data=testing)
error_nrmse = nrmse(testing.power, prediction.power)
error_r2 = r2(testing.power, prediction.power)
df.loc[f'{location}_{month}'].nrmse_arima = error_nrmse
df.loc[f'{location}_{month}'].r2_arima = error_r2
print(f'nRMSE: {error_nrmse}; R2: {error_r2}')
except Exception as e:
print(
from importers import pvwatts
from predictors.arima_model import ARIMAModel
from evaluation.error_terms import mse, rmse, nrmse
import matplotlib.pyplot as plt
from pandas.plotting import register_matplotlib_converters; register_matplotlib_converters() # fix pandas future warnings
# load data
data = pw.load(lat=52.470001220703125, lon=13.399999618530273) # station in berlin
# prepare testing and training data
training = data['20190104':'20190131'] # use 4 weeks in january for training
testing = data['20190201':'20190202'] # use 2 days in february for testing
# prepare model
model = ARIMAModel(scaling=True)
model.fit(training, order=(3,0,1), seasonal_order=(3,0,3,24), filter=['tamb'], use_exogenous=True)
model.predict(testing_data=testing)
# evaluate model
print(f'MSE: {mse(testing.power, model.prediction.power)}')
print(f'RMSE: {rmse(testing.power, model.prediction.power)}')
print(f'nRMSE: {nrmse(testing.power, model.prediction.power)}')
# plot testing features
plt.plot(testing.tamb, color='red')
plt.show()
# plot prediction and expected power output
plt.plot(testing.power, color='red')
plt.plot(model.prediction.power, color='orange')
plt.show()
# one time without exogenous variables (arima only)
columns = {
'features': 'none',
'svr_jan': math.nan,
'svr_april': math.nan,
'svr_july': math.nan,
'svr_oct': math.nan
}
for month, dates in datestrings:
training_start, training_end, testing_start, testing_end = dates
training = data[training_start:training_end]
testing = data[testing_start:testing_end]
# arima
print(f'fitting arima model without exogenous variables for {month}...')
arima = ARIMAModel(scaling=True)
arima.fit_auto(training,
p=(1, 3),
q=(1, 3),
P=(1, 3),
Q=(1, 3),
d=0,
D=0,
use_exogenous=False)
print(
f'fitted arima model successfully with order={arima.model.order}, seasonal_order={arima.model.seasonal_order}'
)
prediction = arima.predict(hours=48)
error = rmse(testing.power, prediction.power)
print(f'RMSE: {error}')
print()
for location, [data, base_data] in {
'edwards': [edwards, edwards_base],
'car_park': [car_park, car_park_base]
}.items():
columns = {}
for month, dates in datestrings:
training_start, training_end, testing_start, testing_end = dates
training = data[training_start:training_end]
testing = data[testing_start:testing_end]
# arima
print(
f'fitting ARIMA model for location {location} and month {month} at {datetime.now()}'
)
try:
arima = ARIMAModel(scaling=True)
# arima.fit(training, order=order, seasonal_order=seasonal_order, filter=filter, use_exogenous=True)
arima.fit_auto(training,
p=p,
q=q,
P=P,
Q=Q,
d=0,
D=0,
filter=filter,
trace=False,
use_exogenous=True)
print(
f'successfully fitted ARIMA model with order={arima.model.order}; seasonal_order={arima.model.seasonal_order}'
)
prediction = arima.predict(testing_data=testing)