Energy Digest

The paper models Ireland's electricity system using a Mixed Integer Linear Programming (MILP) model to estimate renewable generation curtailing and constraints. A mathematical formulation is presented and tested on both simple and realistic networks, with results showing the model accurately captures curtailment effects while highlighting limitations. The proposed model identifies areas for future improvement in capturing constraint effects.

A new framework uses a Weather-to-Voltage (W2V) predictive model to identify weather conditions that can trigger high-voltage events in the power grid by maximizing a voltage criticality score. The model screens non-critical weather scenarios to determine potential drivers of grid-level voltage violations, promoting sparse and interpretable perturbations. This approach demonstrates that weather uncertainty is a key factor in triggering HV events.

A hybrid two-stage machine learning pipeline is proposed for fault detection and classification in power transmission systems, which improves end-to-end accuracy from 31.3% to 95.8% on the TLFaultDataset. The pipeline uses an Isolation Forest anomaly detector and a Random Forest multiclass classifier, with feature engineering based on zero-sequence symmetrical components derived from Fortescue's theorem. Ablation experiments show that using these features resolves ambiguity between three-phase and three-phase-to-ground faults, achieving high F1-scores across different classes.

Power outage prediction models' ability to generalize across novel conditions is inflated due to three common methodological choices that obscure their true performance. The predictive performance of these models degrades substantially when evaluated under spatial and temporal holdout experiments, with some failing to outperform a simple null baseline. Incorporating GeoAI foundation model embeddings does not consistently improve the models' generalizability.

The authors propose a framework for co-optimizing electric vehicle (EV) aggregator bidding and power allocation to improve profit and reduce degradation costs. They develop a stochastic programming model that considers the coupling between bidding and power allocation, and also provide an online power allocation model to meet solution time requirements. The proposed framework is verified in a case study to show its effectiveness in improving EV aggregator profits and reducing costs.

Electric vehicle owners can purchase "electric vehicle charging rights" (CRs) to reserve a predefined charging service, reducing waiting times and guiding them towards optimized charging behaviors. A trading mechanism (CRM) allows EVs to buy CRs in advance, estimating waiting times and updating pricing based on queue theory. The proposed CRM reduces waiting times and mitigates congestion in EV charging stations through accurate waiting time estimation for the first time.

A new data-driven approach to dimension reduction is proposed for industrial load modeling, which uses optimal energy usage data from industrial loads to train a reduced model fitting the original constraints. This method, called the adjustable load fleet model, outperforms traditional analytical methods on three industrial load datasets. It effectively addresses mixed-integer constraints and improves performance in economic dispatch or market clearing processes.