grro / pv_forecast

machine learning based photovoltaic power forecast

photovoltaic power forecast

pv_forecast provides a set of artifacts to obtain PV solar power forecast. The library uses machine learning approaches to perform forecasts. To get appropriated results, real measured PV power values must be delivered, periodically. Internally, the library makes use of DWD weather forecast data.

Installing the library

To install this software you may use PIP package manager such as shown below

sudo pip install pvpower

Using the library

After this installation you should configure the library with your environment parameters. You have to set the closest DWD station id of our PV system location. Refer DWD station list to select the proper station id.

from pvpower.forecast import PvPowerForecast

dwd_station_id = 'L160'
pv_power_forecast = PvPowerForecast(dwd_station_id)

To get a power forecast, the predict method has to be called

tomorrow = datetime.now() + timedelta(days=1)
power_watt_tomorrow = forecast.predict(tomorrow)

Train the library with real measurements

It is essential that the PvPowerForecast library is provided with real measured PV values of our PV system. The provided real data is used to adapt the internal machine learning engine to your specific environment. Providing technical parameters of your PV system such as installed power or cardinal direction is not required. The library is self-learning.

# please provide the real measured PV power value periodically. 
# The period should be between 1 minute and 15 minutes.

while True:
    real_pv_power_watt = ...read real PV power ...
    pv_power_forecast.add_current_power_reading(real_pv_power_watt)
    time.sleep(60)

The provided real measurements will be stored internally on disc and be used to update the internal prediction model. Please consider, that a more accurate forecast requires collecting real PV measurements for at least 2-3 weeks, typically. Do not stop providing measurements, even though the predictions become better and better. You may use a periodic job to provide the real PV values

Energy management system support

The basic functionality of this library is to support photovoltaic power forecast. However, to maximize the yield of your PV system, your home appliances such as a dishwasher or laundry machine should operate only in periods when your PV system is delivering sufficient solar power. To manage this, the Next24hours convenience class can be used as shown below

from pvpower.forecast import PvPowerForecast
from pvpower.forecast_24h import Next24hours

power_forecast = PvPowerForecast('L160')
next24h = Next24hours.of(power_forecast)
peek_watt = next24h.peek()
peek_time = next24h.peek_time()
...

To start your home appliance such as a dishwasher at the right time you may query the available execution time frames. In the example below the frames will be filtered considering a hypothetical basic electricity consumption of 350 watt per hour. Time frames will be considered only, if the solar power is higher than the expected basic electricity consumption. Based on the resulting time frames the best one is used to start the home appliance in a delayed way.

...
pogram_duration_hours = 3
high_power_3h_frames = next24h.frames(width_hours=pogram_duration_hours).filter(min_watt_per_hour=350)
if high_power_3h_frames.empty():
    # start now (no sufficient solar power within next 24h)
    my_dishwasher.start_delayed(datetime.now())
else:
    # start delayed when best frame is reached
    best_3h_frame = high_power_3h_frames.best()
    my_dishwasher.start_delayed(best_frame.start_time)