solcast['Hour'] = solcast.Date.dt.hour
solcast = solcast.drop(columns=['Date'])
solcast = solcast.reindex(index=np.roll(solcast.index, -4)).reset_index(
drop=True)
solcast['solar_output_kw'] = pv.solar_output_kw
solcast = solcast.drop(solcast.loc[(solcast.Dni == 0) & (solcast.Dhi == 0) &
(solcast.Ghi == 0)].index)
target_output = ['Dhi', 'Ghi', 'Dni']
#solcast.loc[solcast.Azimuth<=-1] = solcast.loc[solcast.Azimuth<=-1] + 360
est_dhi = solcast.Ghi - solcast.Dni * np.cos(solcast.Zenith)
model, batch, validation_data, callbacks, x_test_scaled, y_test_scaled, y_test, y_scaler = ml.model_generation(
solcast,
target_output,
batch_size=256,
sequence_length=24,
training_len=0.75,
validation_len=0.225)
model = ml.train_model(model,
batch,
validation_data,
x_test_scaled,
y_test_scaled,
callbacks,
epoch_size=25,
epoch_steps=100)
ml.save_model(model, 'model.h5')
columns={
'hour_index', 'timestamp', 'horizon_elevation_angle',
'optimizer_input_power', 'optimizer_output_power',
'dry_bulb_temperature', 'windspeed', 'albedo', 'nameplate_power',
'module_mpp_power', 'module_power', 'optimal_dc_power',
'optimal_dc_voltage', 'actual_dc_voltage',
'module_irradiance_derated_power', 'inverter_overpower_loss',
'inverter_underpower_loss', 'inverter_overvoltage_loss',
'inverter_undervoltage_loss'
})
df = df.div(1000)
target_output = ['actual_dc_power', 'ac_power', 'grid_power']
model, batch, validation_data, callbacks, x_test_scaled, y_test_scaled, y_test, y_scaler = ml.model_generation(
df, target_output)
model = ml.train_model(model,
batch,
validation_data,
x_test_scaled,
y_test_scaled,
callbacks,
epoch_size=15,
epoch_steps=100)
ml.save_model(model, 'watson_ibm_model.h5')
output = ml.load_models('watson_ibm_model.h5')
import importlib