0
$\begingroup$

I'm doing a machine learning time series forecast of electricity production shares by power plant (nuclear, coal-fired, gas, solar, wind, water etc.) in my country in 5 year horizon. I have historical data (weekly averages by power plants) since 2010.

Is there a way to include external factors in the forecast using ML methods such as LSTM, XGBoost, SVM, ARIMA and others? For example tell the model that certain percentage of solar power plants are expected to be installed in the future or that some coal-fired power plants will be shutdown in the next 2 years and thus this type of power plant is likely going to lose share in the overall electricity production?

I feel like the forecast can't be good if I'm relying only on historical data and not including other factors as well.

Is machine learning even good option for long term forecasts of electricity production shares?

Improve this question
$\endgroup$

1 Answer 1

Reset to default
0
$\begingroup$

Definitely yes. As starter, one approach combines time-series technique (e.g. ARIMA) to learn the historical trend, and leverage ML algorithms (e.g. random forest) to incorporate other external factor to refine the prediction.

Personally I found this Kaggle Learn's course very useful.

Improve this answer
$\endgroup$
5
  • $\begingroup$ Thank you for the recommendation. Does the external factor have to be some sort of dataset as well? Sorry if it's a silly question but is there a way to tell the model just something like "hey, just so you know, when you're making this 5 year forecast, in next 2 years 10% of the coal-fired power plants will shutdown. I'm having trouble to understand how would the external factor (like 10% of coal-fired powe plants shutting down) look like as a feature. $\endgroup$
    – Dan Jírovec
    Commented Apr 18, 2023 at 15:49
  • $\begingroup$ There are 2 options: 1st is to look for historical data of total power produced by coal-fired power plants, e.g. from 1990 to 2022 it was 1MW, 1.2MW, 2.1MW... 10MW and append the series as one feature to the model. Then in forecasting you supply an estimate on how it would decrease in next 5 years, e.g. 10MW, 9MW, 8MW... $\endgroup$
    – lpounng
    Commented Apr 19, 2023 at 1:32
  • $\begingroup$ Another way is to take a step back and ask, 'did anyone make any estimate on how the price would change given a change in electricity production?' Chances are there are already existing research which you can use. $\endgroup$
    – lpounng
    Commented Apr 19, 2023 at 1:42
  • $\begingroup$ Thank you for elaborating. In the first option, can I supply the estimate to pretty much every model or is it model specfic? I didnt know such option existed and can't find any literature on it. $\endgroup$
    – Dan Jírovec
    Commented Apr 20, 2023 at 6:29
  • $\begingroup$ There are a lot of ways to incorporate external factors in, I believe you can find a few ways in any approach you choose. Again, I recommend the Kaggle course as a starting point, especially the part how it put other factors into the model with random forest/NN etc.. $\endgroup$
    – lpounng
    Commented Apr 20, 2023 at 8:02

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.