Short-term load forecasting model for TSOs (LSTM)
A forecasting service for predicting the aggregated hourly net electrical load of the Portuguese transmission system operator (REN). The service makes use of an LSTM network.(AIExperiments Asset)
This is a forecasting service for predicting the aggregated hourly net electrical load of the Portuguese transmission system operator (REN). The core of the service is a totally recurrent LSTM deep neural network. The model has been trained on the REN load time series for the years 2018 and 2019 (except December 2019). The service is served as a docker container and a client script is also provided to help the user form their inference requests. The model is totally configurable in terms of:
- Provided ground truth data points: The client can update the existing model with the desired length of new data points that have been observed. The provided input should follow the format of the csv file history_sample.csv.
- Forecast horizons: The client can request a forecast horizon of their preference. It should be noted that large forecast horizons lead to worse results due to the error propagation caused by the LSTM recurrence.
This model has been developed within I-NERGY project. This asset has been published in AIExperiments.