this project is to implement different deep learning architectures and evaluate them based on their performance on the hour-ahead electricity price and load prediction task. More specifically, we will evaluate (i) Random Forest, (ii) CNN-Univariate, (iii) CNN-Multivariate, (iv) RNN-LSTM and (v) BiLSTM architectures, using the root mean squared error (RMSE). Furthermore, we will experiment on different task formulations and types of frameworks, alongside the two following dimensions: • We will compare the performance of univariate time series forecasting and multivariate time series forecasting. Univariate time series forecasting is a framework on which the predicted quantity (i.e. electricity price) is the sole feature that is used by the models, whereas the multivariate variant of the task also uses other features which may prove important for the prediction, such as the load of the energy grid, the temperature, etc. • We will compare the performance of using different time-steps (3, 10 and 25 time-lags) as a way of reframing the time-series prediction task into a supervised learning problem, i.e. using the past 3, 10 and 25 values of the features which are fed into our models.
xminxx / load-forecasting-using-different-deep-learning-architectures Goto Github PK
View Code? Open in Web Editor NEWThis project forked from kushwahavishal646/load-forecasting-using-different-deep-learning-architectures
this project is to implement different deep learning architectures and evaluate them based on their performance on the hour-ahead electricity price and load prediction task. More specifically, we will evaluate (i) Random Forest, (ii) CNN-Univariate, (iii) CNN-Multivariate, (iv) RNN-LSTM and (v) BiLSTM architectures, using the root mean squared erro