This study investigates outage statistics in the Swedish power system. More specifically, this paper delves into the critical analysis and enhancement of data quality, focusing on inconsistencies and missing values, i.e. unknown outage causes and unidentified faulty equipment. By carefully examining the data, noticeable gaps and deficiencies are revealed. Thus, a format for improving outage reporting using a database with 3 relations (outage summary, outage breakdown and customer breakdown) is proposed. In addition to a qualitative analysis of the data, various machine learning algorithms are explored and tested for their capability to predict the unknown values within the dataset, thereby offering a twofold solution: enhancing the accuracy of outage data and facilitating deeper, more accurate analytical capabilities. The findings and proposals within this work not only illuminate the current challenges within outage data management but also pave the way for more robust, data-driven decision-making in outage management and policy formation. 

This study investigates outage statistics in the Swedish power system. More specifically, this paper highlights the critical importance of addressing data quality issues such as inconsistencies and missing values, including unknown outage causes and unidentified faulty equipment. Existing research often overlooks the depth of these data quality challenges, leaving significant gaps in the reliability and utility of outage statistics. This paper reveals noticeable deficiencies in the current data and proposes a structured format for improving outage reporting through a database with three relations: outage summary, outage breakdown, and customer breakdown. To tackle these issues, a detailed qualitative analysis of the data is conducted, complemented by the exploration and testing of various machine learning algorithms. These algorithms are employed to predict unknown values within the dataset, thereby offering a twofold solution: enhancing the accuracy of outage data and enabling more precise analytical capabilities. Specifically, methods such as decision trees and random forests are utilized to address the data gaps. The findings and proposals within this work not only illuminate the current challenges in outage data management but also pave the way for more robust, data-driven decision-making in outage management and policy formation.

Decentralized energy generation, often in the form of industrial Combined Heat and Power (CHP) units, meets a significant part of the global energy demand. The power production from these units has to be forecast, both individually and collectively, to ensure balance and avoid congestion events on the power grid. However, despite its importance, forecasting CHP generation remains an under-studied problem. With increasing proliferation of renewable energy sources, large forecast errors for CHP generation are rapidly taking center-stage as well. In this paper, we propose a global, ensemble-based machine learning (ML) method to improve short-term forecasting accuracy. We demonstrate its efficacy using data from one hundred largest (industrial) CHP units in Belgium, showing a forecast error reduction of up to 27% compared to the baseline method. This can greatly reduce balancing needs and costs, as well as congestion issues. Our results also highlight several nuances that must be kept in mind by grid operators and market players alike, including fitting global models that leverage data from many CHPs which can lead to better forecast accuracy but limit interpretability of the results, and the fact that there is no single best forecast model for all CHPs.

In power systems, maintaining transient stability is crucial to avoid unanticipated blackouts. The role of Transient Stability Assessment (TSA) is vital for quickly identifying and promptly addressing instabilities. TSA facilitates rapid reactions to serious fault conditions. This paper pioneers the integrated comparison of two distinct methodologies-Maximal Lyapunov Exponent (MLE) methods and Convolutional Neural Networks (CNN)-in a single unified framework for transient stability assessment in power systems, uniquely evaluating their accuracy and reliability for TSA. The CNN-based method uses measured time series data from voltage magnitude, phase angle, and frequency measurements at generator buses, while the MLE approach utilizes both phase angles and frequency of generator buses. This paper provides a qualitative and quantitative comparison of the performance and accuracy of MLE and CNN. This research utilizes the same case studies conducted on the Nordic32 system for both MLE and CNN to ensure robust, unbiased comparisons and promote interdisciplinary research, aligning with current trends in AI integration in power systems.

Reliability indices like the System Average Interruption Duration Index (SAIDI) and System Average Interruption Frequency Index (SAIFI) serve as the Key Performance Indicators (KPIs) for Distribution System Operators (DSOs). They effectively measure the frequency and impact of outages on end users. Given the criticality of the electrical grid to many functions of the modern world, minimizing these values has been and continues to be a priority for DSOs. SAIDI and SAIFI can, however, be influenced by many factors including but not necessarily limited to the network topology, the type of installed components and the quantity of customers connected to the grid. In this work we thus attempt to predict the reliability indices of a DSO based on financial, customer and grid composition statistics reported to regulatory bodies by DSOs in Sweden between 2010 and 2021. By decomposing which features are the strongest predictors for SAIDI and SAIFI, DSOs can see how changes in their customer base and grid composition impact their reliability KPIs. In addition these indices can potentially be used to indicate which parts of the grid are most vulnerable to outages and thus prioritize mitigations at those locations.

Resilience in the electrical grid is of growing concern due to worsening weather patterns, heightened cyber-physical connectivity, and increased penetration of non-dispatchable generators. High-level strategic decisions about the commitment of flexibility resources must therefore be made between prioritizing short-term profits through ancillary services and minimizing the impacts of High Impact Low Probability (HILP) events. A Finite State Machine (FSM) is proposed to model these high level decisions in terms of Normal, Alert, Critical, Islanded and Blackout states. Transitions are triggered by the Transmission System Operator (TSO), Distribution System Operation (DSO) or the proposed Resilience Strategist (RS) depending on the expected vulnerability to a forecasted HILP event. This work therefore provides an overview of the operating conditions contained in the FSM as well as the role and functionality of the RS.

Resilience-enabling technologies (RETs), such as batteries, distributed generation and flexible demand, can be used to recover from, respond to, or anticipate high impact low probability (HILP) events. Increased investment in RETs is critical under the twin threats of climate change and cyber warfare since these - if unaddressed - increase the risk and severity of future HILP events. HILP events can lead to prolonged disruption of electricity supply to end users, which can reduce quality of life and undermine the operation of industries. Since Swedish grid users are entitled to compensation from the distribution system operator (DSO), there is a strong incentive for the grid owner to maximize resilience and minimize down time. To address the need to increase resilience at minimal cost, the interplay between the different energy markets, market rules and legislation in the Swedish energy grid for RETs is therefore mapped. A comparison is also made between the profitability of resilience-only, single-market and multi-market participation of a battery RET based on historical data and perfect information. Even when considering a 25% reserve capacity for resilience activation, the results show a greater return on investment for a battery RET when it is used in multiple markets, as opposed to pure grid reinforcements or single markets. They also highlight the need to update legislation in Sweden regarding islanded operation and energy arbitrage.

In this paper, we present a method for real-time transient stability prediction based on convolutional neural networks (CNN) using a novel CNN architecture compared to previous works on the topic. The method is based on monitoring voltage phasor and frequency measurements at the generator terminal buses, which are presented to the neural network in the form of three channel RGB images, taken as a sliding window. The sliding window consists of the most current set of measurements, as well as the four most recent historical measurements for a total of five time steps. When deployed, the neural network is continuously fed real-time measurements, thereby functioning as a real-time stability monitoring system. The neural network is trained and tested on simulated data of the Nordic32 system using five-fold cross-validation. The trained classifier is able to accurately predict instability in all but one case from the test set. In the successfully identified unstable cases, instability was predicted five cycles after fault clearance.

The electrical grid is a safety-critical system, since incorrect actions taken by a power system operator can result in grid failure and cause harm. For this reason, it is desirable to have an automated power system operator that can reliably take actions that avoid grid failure while fulfilling some objective. Given the existing and growing complexity of power system operation, the choice has often fallen on deep reinforcement learning (DRL) agents for automation, but these are neither explainable nor provably safe. Therefore in this work, the effect of shielding on DRL agent survivability, validation computational time, and convergence are explored. To do this, shielded and unshielded DRL agents are evaluated on a standard IEEE 14-bus network. Agents are tasked with balancing generation and demand through redispatch and topology changing actions at a human timescale of 5 minutes. To test survivability under controlled conditions, varying degrees of scheduled unavailability events are introduced which could cause grid failure if unaddressed. Results show improved convergence and generally greater survivability of shielded agents compared with unshielded agents. However, the safety assurances provided by the shield increase computational time. This will require trade-offs or optimizations to make real-time deployment more feasible.

Distribution level outages generally affect fewer customers than regional or transmission level outages. However, as global temperatures continue to rise, the radial topology and overhead lines typical at this level make it particularly vulnerable to High Impact Low Probability weather events. A Machine Learning model is therefore proposed that uses Multinomial Logistic Regression (MLR) to predict the likelihood of an outage given the weather conditions and the composition of the Distribution System Operator (DSO). The model is tuned by using a traditional binary classification problem as ground truth, but is evaluated based on its probability distributions near outage events. Results show a greater classification confidence for true outages than false outages as well as a probability distribution that is skewed towards actual outage events.