According to the arXiv abstract for paper 2606.19118, authors Antoine Pesenti and Aidan O'Sullivan combine deep neural network models with explainable AI techniques to analyse electricity price formation across 39 European bidding zones. The paper reports use of SHAP for feature attribution and an extension called SSHAP to aggregate explanations in high-dimensional settings, per the arXiv submission. The authors identify renewable sources, particularly solar, as disproportionately influential for prices despite smaller generation shares, and report gas prices and cross-border interconnections as consistent drivers, according to the abstract. The paper also constructs a synthetic EU-wide electricity market to explore a counterfactual single-price scenario, as stated on arXiv.
What happened
According to the arXiv abstract for paper 2606.19118 (submitted 17 Jun 2026), authors Antoine Pesenti and Aidan O'Sullivan train deep neural networks to predict electricity prices and pair those models with explainable AI methods to analyse drivers across 39 European bidding zones. The paper reports using SHAP (SHapley Additive exPlanations) to quantify feature contributions and an aggregation framework called SSHAP to improve interpretability in high-dimensional settings, per the arXiv submission. The authors state that renewable energy sources, particularly solar, play a disproportionately important role in price formation, while gas prices and interconnections are identified as dominant and consistent drivers. The abstract also describes construction of a synthetic EU-wide electricity market to explore a counterfactual single-price scenario.
Technical details
Editorial analysis: The paper pairs black-box predictive models with post hoc attribution methods, following a common XAI workflow where predictive performance and interpretability are decoupled. Using SHAP for local and global attributions is standard; the reported use of an aggregation framework (SSHAP) addresses a frequent practitioner pain point when scaling SHAP to many zones and features. The creation of a synthetic, integrated-market counterfactual is a typical approach in energy-economics research for isolating structural effects from local noise.
Context and significance
Electricity markets exhibit strong nonlinearities and cross-region coupling, so applying DNNs plus XAI can surface multivariate dependencies that linear models miss. For practitioners, results that highlight the outsized role of solar and the continued dominance of gas prices underscore the value of combining high-capacity models with attribution tools to support scenario analysis and feature monitoring. The synthetic EU-market experiment provides a tractable counterfactual useful for policy simulation and stress testing models across integrated grids.
What to watch
Editorial analysis: Observers should look for the full paper and any released code or data to validate how SSHAP aggregates attributions, the model architectures and training regimes used, and whether the synthetic market construction includes realistic transmission constraints and bidding rules. Replication across different time windows and inclusion of additional exogenous drivers would determine robustness.
Scoring Rationale #
A domain-specific arXiv study that combines DNNs and XAI has practical relevance for energy-market modelling and interpretability workflows, but it is a single paper without broad release or field-changing claims. Recent submission date slightly reduces immediacy.
Practice interview problems based on real data
1,500+ SQL & Python problems across 15 industry datasets — the exact type of data you work with.