Search

This paper presents a novel, unified approach for generating high-quality datasets for training machine-learned models for real-time security assessment in power systems. Synthetic data generation methods that extrapolate beyond historical data can be inefficient in generating feasible and rare operating conditions (OCs). The proposed approach balances the trade-off between historically relevant OCs and rare but feasible OCs. Unlike conventional methods that rely on historical records or generic sampling, our approach results in datasets that generalise well beyond similar distributions. The proposed approach is validated through experiments on the IEEE 118-bus system, where a decision tree model trained on data generated using our approach achieved 97 % accuracy in predicting the security label of rare OCs, outperforming baseline approaches by 41 % and 20 %. This work is crucial for deploying reliable machine-learned models for real-time security assessment in power systems undergoing decarbonisation and integrating renewable energy sources. • Machine-learned models as surrogates to predict real-time security of power systems. • Quality datasets are crucial to train high performing Machine Learning (ML) models. • Method trades-off historically relevant operating conditions (OCs) and 'rare' feasible OCs. • Method generates datasets that generalise well beyond similar distributions. • Studies carried out on the IEEE 118-bus validate the proposed method. [ABSTRACT FROM AUTHOR]

With recent telemetric advancements, the real-time availability of power grid measurements has opened challenging opportunities for the design of advanced protection and control schemes. Artificial neural networks (ANN) are promising approaches for detecting and classifying disturbance events from measurement data. Numerous offline ANN-based classification algorithms were proposed in the past, which increased the interest for their real-world deployment. However, these algorithms are inadequate due to their conventional offline training procedures, model updating, and large backend computing requirements. Besides, most ANN-based algorithms require disturbance event samples to be collectively available during training. This availability may be uncommon in practice as disturbance events are rare, non-deterministic, and uncertain. Hence, an online training procedure where the model processes the events on-the-fly is required. However, ANNs may also suffer from catastrophic forgetting where the model may unintentionally unlearn an occurred disturbance under the learning of new event types; this means ANN may not detect very similar disturbances of the same type in the future. In this paper, we propose Dynamic Incremental Learning (IL) method for ANN models, which is updated in real-time when a new disturbance is detected. Our proposed method adopts a Replay-based IL strategy for designing long-term IL, balancing the accuracy with catastrophic forgetting of disturbance events. The method is designed in a way to learn efficiently for incoming disturbance data with minimized training time and the highest classification accuracy eliminating catastrophic forgetting. The results describe the methodology's performance regarding classification accuracy, training time, and storage memory. The findings demonstrate that the Dynamic IL method is promising for efficient learning and event classification. • We use Incremental Learning (IL) in power systems for online model training. • The new Dynamic IL method efficiently trains deep learning models in near real-time. • Our method adapts the model to learn new event types anticipated in a modern grid. • Learning accuracy, training time, and memory can be tuned during events learning. [ABSTRACT FROM AUTHOR]

Machine learning (ML) for real-time security assessment requires a diverse training database to be accurate for scenarios beyond historical records. Generating diverse operating conditions is highly relevant for the uncertain future of emerging power systems that are completely different to historical power systems. In response, for the first time, this work proposes a novel split-based sequential sampling approach based on optimisation that generates more diverse operation scenarios for training ML models than state-of-the-art approaches. This work also proposes a volume-based coverage metric, the convex hull volume (CHV), to quantify the quality of samplers based on the coverage of generated data. This metric accounts for the distribution of samples across multidimensional space to measure coverage within the physical network limits. Studies on IEEE test cases with 6, 68 and 118 buses demonstrate the efficiency of the approach. Samples generated using the proposed split-based sampling cover 37.5% more volume than random sampling in the IEEE 68-bus system. The proposed CHV metric can assess the quality of generated samples (standard deviation of 0.74) better than a distance-based coverage metric which outputs the same value (standard deviation of < 0. 001) for very different data distributions in the IEEE 68-bus system. As we demonstrate, the proposed split-based sampling is relevant as a pre-step for training ML models for critical tasks such as security assessment. • Database generation is pivotal in machine learning for Dynamic Security Assessment • GAPSPLIT ∗ approach sequentially samples pre-fault operating conditions • Convex hull volume (CHV) assesses the quality of a generic sampler • Studies carried out on the IEEE test systems with 6 , 68 , and 118 buses • Gapsplit samples cover 37.5 % more volume than random sampling [ABSTRACT FROM AUTHOR]

For optimal operation of power-intensive plants, production scheduling and electricity procurement have to be considered simultaneously. In addition, uncertainty needs to be taken into account. For this purpose, an integrated stochastic mixed-integer linear programming model is developed that considers the two most critical sources of uncertainty: spot electricity price, and product demand. Conditional value-at-risk is incorporated into the model as a measure of risk. Furthermore, scenario reduction and multicut Benders decomposition are implemented to solve large-scale real-world problems. The proposed model is applied to an illustrative example as well as an industrial air separation case. The results show the benefit from stochastic optimization and the effect of taking a risk-averse rather than a risk-neutral approach. An interesting insight from the analysis is that in risk-neutral optimization, accounting for electricity price uncertainty does not yield significant added value; however, in risk-averse optimization, modeling price uncertainty is crucial for obtaining good solutions. [ABSTRACT FROM AUTHOR]

• A metric to quantify the impact of topology changes on the accuracy of DSA models. • A construction method for training databases following high-impact topology changes. • Robustness against frequent changes in the network topology. • Study on the IEEE 68-bus system shows improvements in the accuracy up-to 52 %. • Study on the IEEE 68-bus system shows a reduction of up-to 85 % in the training data. Machine learning has been used in the past to construct predictors, also known as classifiers, for dynamic security assessment. Although accurate classifiers can be trained for a single topology, often they do not work for another. However, the power system topology can change frequently during operation due to maintenance and control actions. At one topological configuration, the system may have a different response to a fault than at another as the underlying distribution of power flows can be completely different. Quantifying the impact of changes in the topology on the predictive models' performance is an important step forward to minimize inaccurate predictions and improve their reliability. In this paper, for the first time, a metric for quantifying the impact of a topology change on the accuracy of the classification model is proposed. The key novelty is to first select a subset of power flow features with a physically informed feature selection technique and subsequently compute the metric with a novel convex hull-based analysis. In addition, the approach can advise to effectively constructing new training databases that improve the accuracy of new machines trained after high-impact topology changes. Through a case study using transient stability on the IEEE 68 -bus system, the use of the proposed metric in real-time operation was demonstrated. 17 high-impact topology changes were successfully detected among 42 studied topological changes. The subsequent effective construction of the training database improved the predictive accuracy by around 10 %. An interesting finding is the amount of newly generated data can be reduced by up to 85 % as often the generated data is the barrier for data-driven DSA. The proposed workflow significantly reduces data and trains robust classifiers against topological changes marking a fundamental step forward. [ABSTRACT FROM AUTHOR]

• A risk-metric for using machine learning in probabilistic security assessment. • A calibrated training process for accurate probability outputs of machine learning. • A probabilistic balance of machine learning with conventional security assessment. • Robustness against frequent changes in likelihood of contingencies. • Study on French system shows superiority in accuracy, robustness, and 90 % speed-up. Probabilistic security assessment and real-time dynamic security assessments (DSA) are promising to better handle the risks of system operations. The current methodologies of security assessments may require many time-domain simulations for some stability phenomena that are unpractical in real-time. Supervised machine learning is promising to predict DSA as their predictions are immediately available. Classifiers are offline trained on operating conditions and then used in real-time to identify operating conditions that are insecure. However, the predictions of classifiers can be sometimes wrong and hazardous if an alarm is missed for instance. A probabilistic output of the classifier is explored in more detail and proposed for probabilistic security assessment. An ensemble classifier is trained and calibrated offline by using Platt' scaling to provide accurate probability estimates of the output. Imbalances in the training database and a cost-skewness addressing strategy are proposed for considering that missed alarms are significantly worse than false alarms. Subsequently, risk-minimised predictions can be made in real-time operation by applying cost-sensitive learning. Through case studies on a real data-set of the French transmission grid and on the IEEE 6 bus system using static security metrics, it is showcased how the proposed approach reduces inaccurate predictions and risks. The sensitivity on the likelihood of contingency is studied as well as on expected outage costs. Finally, the scalability to several contingencies and operating conditions are showcased. [ABSTRACT FROM AUTHOR]