Core Concepts
Conformal prediction can be effectively combined with various bidding strategies to enhance the reliability and profitability of photovoltaic power participation in electricity markets.
Abstract
This paper proposes a framework that leverages conformal prediction (CP), an emerging probabilistic forecasting method, to enhance decision-making for photovoltaic (PV) power participation in electricity markets.
The framework consists of three main steps:
- Point predictions of PV power are generated using machine learning models such as simple linear regression, multiple linear regression, and random forest regression.
- Uncertainty of the point predictions is quantified using various CP methods, including basic CP, CP with k-nearest neighbors, CP with Mondrian binning, and conformal predictive systems (CPS).
- The quantified uncertainty is then used as input for different bidding strategies, including trust-the-forecast, worst-case, Newsvendor, and expected utility maximization (EUM), to determine the optimal quantity bids for the day-ahead electricity market.
The performance of the framework is evaluated using actual weather and energy market data from the Netherlands. The results show that:
- CP methods, especially CPS with k-nearest neighbors and Mondrian binning, outperform linear quantile regression in terms of forecasting performance.
- Combining CP methods with certain bidding strategies, such as EUM with conditional value at risk, can yield high profit with minimal energy imbalance.
- The best-performing approach is using CPS with k-nearest neighbors and Mondrian binning after random forest regression, which achieves up to 93% of the potential profit with minimal imbalance.
The proposed framework can help market participants increase their profit while being aware of the associated risk, and grid operators improve their insights into the expected grid loading.
Stats
The day-ahead market prices for the Netherlands in the considered time period were extracted from the ENTSO-E Transparency Platform.
The up- and down-regulation prices in the real-time market were obtained from TenneT, the Dutch TSO.
Quotes
"Conformal prediction (CP) is an emerging distribution-free and model-agnostic probabilistic forecasting method that offers a measure of confidence by transforming point predictions into prediction intervals."
"Combining CP in combination with certain bidding strategies can yield high profit with minimal energy imbalance."
"Using conformal predictive systems with k-nearest neighbors and Mondrian binning after random forest regression yields the best profit and imbalance regardless of the decision-making strategy."