ICIS webinar - Price sensitivity analysis with neural networks
•
0 likes•677 views
This document summarizes a webinar on using neural networks to model German day-ahead spot electricity prices and analyze price sensitivity. It discusses modeling spot prices with neural networks, analyzing price expectations under varying market conditions, and answering participant questions. Key topics covered include using neural networks for price forecasting, considering different market inputs, approximating the merit order, and a new application that allows adjusting inputs to see how prices may change under different scenarios.
Recommended
Recommended
More Related Content
What's hot
What's hot (20)
Similar to ICIS webinar - Price sensitivity analysis with neural networks
Similar to ICIS webinar - Price sensitivity analysis with neural networks (20)
Recently uploaded
Recently uploaded (20)
ICIS webinar - Price sensitivity analysis with neural networks
- 1. Power price sensitivity analysis with Neural Networks for German day ahead spot trading Free webinar 10.07.2014 Jonathan Scelle Senior Analyst EU Power Markets Sebastian Stütz Lead Analyst Power
- 2. www.icis.com Content PART I – Concept of modeling spot prices with Neural Networks PART II – Sensitivity: Price expectations under varying conditions? Your Questions
- 3. www.icis.com Spot price forecast model – June, 14th & 15th 2014 June, 14th 2014 – 1-day-ahead: June, 15th 2014 – 2-day-ahead: Ø MAE hourly: Ø MAEbase:1.51 € 0.77 € Ø MAE hourly: Ø MAEbase:3.44€ 2.95 €
- 4. www.icis.com How to model power prices? Market Challenges Uncertainty in renewable gener- ation and power demand High day to day price volatility Negative prices Source: ICIS Power Portal
- 5. www.icis.com How to model power prices? Which prices to forecast? Spot vs. forward market (spot market highly volatile with renewable challenges) OTC, daily auction Specific power market problems to address State of information before auction gate closure How to model hours? Separate prediction / 24h prediction? (number of inputs in model / complexity / overfitting?) Multiple day ahead => feed same model / new model? Negative prices
- 6. www.icis.com Why use Neural Networks? NNs can learn from sample data NNs are data driven self-adaptive models which determine their function based on sample data No a-priori assumptions are needed NNs can generalize NNs can produce reasonable outputs for previously unseen data NNs are universal function approximators NNs can deal with non-linear relationships NNs are successfully used for a wide variety of tasks Facial Recognition Text analysis Technical process control Medical diagnosis Stock market forecasts
- 7. www.icis.com Price model key summary Data prepared with power market insight We aggregate raw input series to prepared series like residual demand We apply averages and self-developed indicators for key power factors, e.g. indicators for degree of utilization of merit order Considered inputs available capacities (EEX after scope correction with BNetzA figures) power demand forecast (own Neural Network based model for DE/AT) wind and solar production forecasts (own model) fuel and carbon price levels Import/export flows* multiple weather variables (based on high resolution GFS-WRF) efficiency factors of power plants * Import / Export explicit modeling is running project.
- 8. www.icis.com Implicit stack / merit order approximation Our model is not trained to forecast absolute prices but to learn price gradients in the merit order, visible through auction results (extension to bidding curves in plan) The trained “model” can be described as an experienced view on price gradients at the price setting parts of the merit order Hence, the model is capable of predicting price changes from drops/increases in e.g. residual load or available capacities In order to distil changes in historic data we normalize always based on each latest week. Our running forecasts take into account latest days and weeks and long-term trends. Source: Risø DTU
- 9. www.icis.com Advantages / disadvantages of Neural Networks as power price models Pros Decreases need for explicit assumptions How do you model in your stack… Actual efficiencies and capacities of each plant? Inland transportation costs Topping turbines Combined heat and process steam generation Must run conditions Transferable to other markets Constantly learning Cons Require long series, structural change of market mechanisms (e.g. capacity market) would be a problem Computationally expensive
- 10. www.icis.com Models in practise: ICIS Power Portal
- 11. www.icis.com Neural Network based price forecast model – backtesting 1-day-ahead: 2-day-ahead: Ø MAE hourly: Ø MAEbase:3.72 € 2.23 € Ø MAE hourly: Ø MAEbase:4.01€ 2.91 €
- 12. www.icis.com In a perfect world, inputs would be always right In Power markets: Most key inputs have to be estimated, too Estimations change over time and with more insight Sometimes, inputs are not even clear ex-post – What’s the German power demand?
- 13. www.icis.com Example for input forecast: ICIS Power Demand Forecast (DE/AT) Based on Neural Networks, trained to match the demand data supplied by entsoe.eu and apg.at Utilises a high resolution weather forecast data derived from the world-wide operational GFS (Global Forecasting System) model Considers the time of the year as well as a variety of date-depending factors Effects that directly affect the population – like weather and holidays – are weighted accordingly for each minor region within the countries Updated 4 times a day starting 3:04 [GMT] Period MAE Jan 2014 1741 MW Feb 2014 1914 MW Mar 2014 1501 MW Apr 2014 1846 MW Jan-Apr 2014 1750 MW30,000 40,000 50,000 60,000 70,000 80,000 90,000 actual_demand DE+AT (entsoe.eu, APG) forecasted_demand DE+AT Last week of April 2014 (latest actual publication by ENTSOE.EU)
- 14. www.icis.com Example I (28-06-2014) 0 1000 2000 3000 4000 5000 6000 1 3 5 7 9 11 13 15 17 19 21 23 MW Hour actual wind predicted wind 0 5000 10000 15000 20000 25000 1 3 5 7 9 11 13 15 17 19 21 23 MW Hour actual solar predicted solar Wind Solar 0 10000 20000 30000 40000 50000 60000 70000 1 2 3 4 5 6 7 8 9 101112131415161718192021222324 MW Hour actual demand predicted demand 0 10000 20000 30000 40000 50000 60000 1 2 3 4 5 6 7 8 9 101112131415161718192021222324 MW Hour actual residual demand predicted residual demand Demand Residual Demand
- 15. www.icis.com Content PART I – Concept of modeling spot prices with Neural Networks PART II – Sensitivity: Price expectations under varying conditions? Your Questions
- 16. www.icis.com Sensitivity: What would be the price under other conditions? “What you feed into is what you get” Single forecasts are tools, not final market views For a specific input, expectations vary (multiple models/state of information) Our aim as a data service provider Enable to widen methodology scope Enable to improve internal market views => Give the ability to adjust and see the changes in the price Enable to trade
- 17. www.icis.com Key power market drivers and what factors to adjust? Residual Load = Demand to be covered by conventional power plants Wind feed-in Solar feed-in Power demand Public behavior, holidays Outages Efficiencies Fuel prices Net interconnection flows Weather Marginal costs and supply bidding behaviourRamping costs Supply structure costs at volume Manual short-term inputs
- 18. www.icis.com The concept of residual load Power demand €/MWh Wind, Solar, Net import flow Residual load
- 19. www.icis.com Which input changes can be explained by shifting residual load against the stack? Price insight is generated by shifting the remaining consumption against the conventional (ANN: implicit) stack Change in power demand? Change in renewables? Change in net interconnection flow? Change in available capacity of baseload power plants (new builds)? Changes in the plant efficiencies and fuel prices? Source: Risø DTU Adjustable with residual load shift concept? No Approximated by ANN learning process
- 20. www.icis.com Residual load vs EPEX price
- 21. www.icis.com Example I (28-06-2014) 0 1000 2000 3000 4000 5000 6000 1 3 5 7 9 11 13 15 17 19 21 23 MW Hour actual wind predicted wind 0 5000 10000 15000 20000 25000 1 3 5 7 9 11 13 15 17 19 21 23 MW Hour actual solar predicted solar Wind Solar 0 10000 20000 30000 40000 50000 60000 70000 1 2 3 4 5 6 7 8 9 101112131415161718192021222324 MW Hour actual demand predicted demand 0 10000 20000 30000 40000 50000 60000 1 2 3 4 5 6 7 8 9 101112131415161718192021222324 MW Hour actual residual demand predicted residual demand Demand Residual Demand
- 22. www.icis.com 0 10000 20000 30000 40000 50000 60000 70000 1 2 3 4 5 6 7 8 9 101112131415161718192021222324 MW Hour actual residual demand predicted residual demand predicted residual demand: scenario RD +5GW 0 10 20 30 40 50 60 1 2 3 4 5 6 7 8 9 101112131415161718192021222324 Eur/MWh Hour actual auction price predicted auction price auction price: scenario RD +5GW Example I (28-06-2014) – Usage of input adjustments Residual Demand Auction Price
- 23. www.icis.com Summary Varying estimations require options to test changes in price models when inputs change Many key fundamental drivers changes can be modelled by left-right shifting of residual load against the stack Opportunity to gain confidence on expected price changes/risks for trading on changing fundamental expectations Source: Risø DTU
- 24. www.icis.com New application by ICIS / Tschach Solutions Input scenarios
- 26. www.icis.com
- 27. www.icis.com Result of full day +5GW shift in demand
- 28. www.icis.com Content PART I – Concept of modeling spot prices with Neural Networks PART II – Sensitivity: Price expectations under varying conditions? Your Questions