Currently, the ESO produce a forecast of hourly interconnector flow at a lead time of day ahead to 1 week ahead. This forecast is based on the expected difference between electricity prices of the two interconnected countries.
To forecast the hourly electricity price in each country, the ESO have acquired future market price data for the peak and baseload periods from external data sources. This is transformed to hourly prices using a technique based on similar historical days. However, this method does not produce the level of accuracy required for a key input into the ESO interconnector flow model, published externally. This project therefore seeks to produce an improved method of converting future market prices data into a forecast of hourly electricity prices.
Benefits
- Improve predictions of Interconnector (IC) flows from a day +1 to day +7 timescale
- Facilitate more accurate decisions on interconnector trades and the volume of response required.
- Improve the system margin forecast.
- Improving the accuracy of interconnectors forecast flow would increase the confidence of the Electricity National Control Centre (ENCC) in reserve products, hence reducing the balancing cost and cost to consumers.
Learnings
Outcomes
The project has delivered a model which allows the ESO to forecast the Norwegian electricity price at an hourly resolution. A series of models were developed and the best modelling technique was the LSTM modelling package. Although, thorough testing is ongoing, indications suggest some improvements could be made by considering some periodic elements in the model.
The project highlighted that the relationship between forecast prices and the intra-day prices are greatly influenced by external events. These insights provide excellent opportunities for model improvement and future study.
The project helped us to narrow down the options for future models. The techniques described, used and analysis have allowed NGESO to rule out many modeling approaches, allowing resources to be efficiently deployed in productive areas.
Lessons Learnt
Working with cutting edge AI and Machine Learning datasets can bring fresh approaches to modeling problems. Using this external expertise to apply novel modeling methods allows a faster transition to alternative approaches in the ESO and the wider energy industry. For example, the LSTM PoC model is something we could apply to other time series forecasting projects within the ESO.
Additionally, working with market leaders in the field allowed ESO to learn key skills for ensuring the smooth running of data science projects. This includes new methods for logging data and how it can be efficiently stored.
The project also helped understand the limits of modelling in predicting price futures. Future projects may benefit from a greater understanding of the specific market and from stakeholder and SME interaction as there are limits to insights that can be gained from a data only approach.
