This work addresses the challenge of accurately forecasting electricity prices within the volatile Australian market, especially during extreme conditions. It leverages advanced generative pre-trained Large Language Models (LLMs) to analyze the content of electricity market notices with the goal of identifying the drivers behind extreme price fluctuations. Additionally, this approach employs LLMs for an in-depth time-series analysis of electricity prices, providing Australian electricity company traders with insights to refine their trading strategies. To enhance forecasting accuracy this study adopts the QLoRA method for fine-tuning open access LLMs, enabling the analysis of market notices to generate a time series event dataset. A CNN-LSTM network architecture is designed to process both electricity price data and market notice information, thereby improving forecast precision in periods of extreme price volatility. The proposed decision support framework undergoes simulation and evaluation using data from the Australian electricity market, demonstrating its potential to significantly benefit traders in navigating the complexities of the energy sector.

SCADA alarm management poses many challenges for power grid control room operators. Vast quantities of information must be managed and filtered by human operators to identify the specific alarm of interest so that system issues can be promptly addressed. The use of knowledge based systems and decision support systems to assist in industrial decision making is well established but comes with heavy cost burdens for development and maintenance. This work demonstrates a possible approach to addressing this cost issue by utilising large language models (LLMs) to parse SCADA alarms and extract information to construct a knowledge base in the form of a knowledge graph. This knowledge base can then be used to construct a decision support system suitable for control room operators to make plain language queries that assist in their problem solving. The decision support approach facilitates better and faster decision making assisting the effective management of electrical distribution networks by control room operators. The case study shown in this work demonstrates that LLMs show promising capacity for knowledge base constructions and have the potential to unlock easy access and insights from large volumes of SCADA alarm data.

The operation of power grids is increasingly complex and there is a growing need for data-driven fault diagnosis in smart grid dispatching. This study presents an innovative real-time machine learning framework that significantly enhances the detection and management of alarm events in power grid SCADA systems. At the heart of this framework is a novel deep neural network (DNN) architecture designed to efficiently identify alarm events. Additionally, an algorithm prioritizing the identification of historically relevant alarm events is developed, providing robust decision support. The proposed framework is constructed and underwent thorough testing with a comprehensive power grid dataset. The results demonstrate that the framework not only meets but exceeds the operational demands for real-time alarm detection, particularly during periods of alarm floodpeak. Moreover, the alarm event searching and decision support algorithm is proven to be a critical tool for power grid control room experts and operators, facilitating rapid decision-making by providing actionable insights based on historical and current data analysis. The successful integration of this machine learning framework into SCADA systems marks a significant step forward in the technological evolution of energy systems, leading to smarter, more efficient, and reliable power grid management.

Smart grid development and evolution requires tools that facilitate the use of data collected from smart grid devices. Smart metering analytics allows stakeholders to gain insights from smart metering data that aids in network management and regulatory compliance. Rule extraction through interim summarisation provides a viable approach to classifying daily electricity consumption load profiles for the solar PV detection problem. This work explains how this method can be applied to smart metering data and shows how the rules generated by this approach can be interpreted to provide insight into the solar PV detection problem. By providing an example approach to smart metering analytics that is interpretable this work addresses the need for white box classification models in the smart grid domain. It is shown that the application achieved up to 88.19% classification accuracy when classifying daily load profiles from our industry supplied dataset.

With the increasing popularity of domestic solar PV systems there is a need for smart grid network operators to be able to identify solar PV systems attached to their networks. This need is driven by human safety, equipment safety, and regulatory compliance concerns. Given the implementation of smart metering as part of the evolution toward smart grids and the availability of smart metering data, methods that automate the identification of solar PV systems from consumption data are needed to address these concerns. This paper proposes an optimal template approach with genetic algorithm for solar PV detection, which successfully classifies solar PV and non-solar PV customers by utilising genetic algorithm optimisation to find optimal template pairs and matching observations to the closest template. This is done by using domain knowledge to specify a template parameterisation specific to the problem and using genetic algorithm optimisation to find template pairs that are optimised for accuracy.