https://github.com/mathiasotnes/power-consumption-forecast

Forecasting power consumption to estimate bidding prices in the Norwegian Nord Pool bidding areas using different deep learning techniques.
https://github.com/mathiasotnes/power-consumption-forecast

Last synced: 21 days ago
JSON representation

Forecasting power consumption to estimate bidding prices in the Norwegian Nord Pool bidding areas using different deep learning techniques.

• Host: GitHub
• URL: https://github.com/mathiasotnes/power-consumption-forecast
• Owner: Mathiasotnes
• Created: 2024-03-06T18:16:28.000Z (10 months ago)
• Default Branch: main
• Last Pushed: 2024-03-14T19:25:54.000Z (10 months ago)
• Last Synced: 2024-03-15T15:05:56.810Z (10 months ago)
• Language: Jupyter Notebook
• Size: 3.75 MB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# Power-consumption-forecast

Forecasting power consumption to estimate bidding prices in the Norwegian Nord Pool bidding areas using different deep learning techniques.

# Notes

- Seasonal patters -> Daily, weekly and yearly

# Requirements

### General

- Must be able to forecast 24 hours into the future with a resolution of 1 hour.

- Must be able to predict on each individual bidding area separately (5 specific models or 1 generic model).

- Must be reproducable, and model should be pre-trained for demonstration.

- A model must be trained on a bidding area, and tested on another along with a reflection of the results. 

- At least three of four different forecasting models:

    - LSTM (Long Short-Term Memory) at least one layer

    - GRU (Gated Recurrent Unit) at least one layer

    - CNN (Convolutional Neural Network)

    - Transformer

    - Feed-forward regression model

- Must add reqularizers (e.g. dropout, L1 or L2)

### Feature engineering

Features must be implemented, visualized and discussed. Not all must be used in the final model, but it should be easy to add/remove them.

The features to include is:

- time_of_day

- time_of_week

- time_of_year

- comsumption_lag_24

- temperature_lag_24

- comsumption_mean_24

- temperature_mean_24

- previous_y (for RNN)

### Visualization

- Training progress

- Observed/forecasted comsumptions (48 hour plots where 24 last is observed/forecasted)

- An error plot showing the mean and standard deviation of the error for each hour of the forecast horizon over the test set. This plot will have forecast horizon (1-24 hours) on the x axis and absolute error on the y axis.

- A summary bar plot that compares the different models you have trained using metrics of your choice (e.g. test set RMSE, MAE, MAPE or similar).

### Checklist

- Training progress visualization √

- Forecast visualization (unscaled) √

- Error plot over forecast horizon √

- Multiple models √

- Comparison of models X

- Demonstration of model trained on seperate bidding areas compared √