This survey paper explores the cybersecurity certification requirements defined by the SunSpec Alliance for Distributed Energy Resource (DER) devices, focusing on aspects such as software updates, device communications, authentication mechanisms, device security, logging, and test procedures. The SunSpec cybersecurity standards mandate support for remote and automated software updates, secure communication protocols, stringent authentication practices, and robust logging mechanisms to ensure operational integrity. Furthermore, the paper discusses the implementation of the SAE J3072 standard using the IEEE 2030.5 protocol, emphasizing the secure interactions between electric vehicle supply equipment (EVSE) and plug-in electric vehicles (PEVs) for functionalities like vehicle-to-grid (V2G) capabilities. This research also examines the SunSpec Modbus standard, which enhances the interoperability among DER system components, facilitating compliance with grid interconnection standards. This paper also analyzes the existing SunSpec Device Information Models, which standardize data exchange formats for DER systems across communication interfaces. Finally, this paper concludes with a detailed discussion of the energy storage cybersecurity specification and the blockchain cybersecurity requirements as proposed by SunSpec Alliance.

Personalized device-level energy consumption recommendations towards energy efficiency can have a notable impact both on electricity bills and on the overall energy supply-demand balance. End-user behavior regarding device activation is usually unknown a priori, thus giving rise to a highly dynamic environment. Hence, Reinforcement Learning (RL) can be utilized for device scheduling and consumption recommendations since it constitutes an Artificial Intelligence (AI) framework that learns a control policy in a dynamic environment through trying actions and observing incurred rewards. However, existing works on energy consumption recommendations do not explicitly take into account human feedback and preferences regarding the issued recommendations, and they train a single RL agent per device, hence missing the human behavior interdependencies in using different devices. In addition, a flexible open-source RL environment model that integrates user behavior in a Markov Decision Process (MDP) model is missing. In this paper, we propose an MDP-driven RL framework for energy efficiency recommendations that jointly learns the user’s behavior for multiple devices. The proposed model is wrapped as an open-source customizable Gymnasium environment, named EMS-env, for multi-device energy efficiency recommendations. EMS-env can simulate different types of consumer behavior profiles based on the MDP model and supports different device types as well as user feedback. Validation experiments demonstrate the framework’s merits and hyperparameters for diverse use cases in terms of user simulation models and RL training policies, resulting in decreased energy costs while maintaining end-user satisfaction.

The project energy Optimization of building Internet Of Things Infrastructures in a Stratified way presented a holistic approach for an AI-enabled EMS for building-level energy management and max- imizing renewable energy sources utilization. The solution integrated a Photovoltaics (PV) generation forecasting module, a building energy demand forecasting module, and an energy task scheduling and optimization component. This paper presents the methodology and preliminary results regarding the PV generation forecasting and energy task scheduling modules tested for the premises of Institute Mihajlo Pupin.

Standard Full-Data classifiers in NLP demand thousands of labeled examples, which is impractical in data-limited domains. Fewshot methods offer an alternative, utilizing contrastive learning techniques that can be effective with as little as 20 examples per class. Similarly, Large Language Models (LLMs) like GPT-4 can perform effectively with just 1-5 examples per class. However, the performance-cost trade-offs of these methods remain underexplored, a critical concern for budget-limited organizations. Our work addresses this gap by studying the aforementioned approaches over the Banking77 financial intent detection dataset, including the evaluation of cutting-edge LLMs by OpenAI, Cohere, and Anthropic in a comprehensive set of few-shot scenarios. We complete the picture with two additional methods. First, a cost-effective querying method for LLMs based on retrieval-augmented generation (RAG), able to reduce operational costs multiple times compared to classic few-shot approaches, and second, a data augmentation method using GPT-4, able to improve performance in data-limited scenarios. Finally, to inspire future research, we provide a human expert’s curated subset of Banking77, along with extensive error analysis.

With buildings accounting for a significant portion of the grid’s total energy demand, it is evident that consumers should be engaged in energy efficiency. In this work, we propose a reinforcement learning approach that conducts energy efficiency recommendations for buildings, in the form of load-shift suggestions for different devices/assets. The adopted methodology can continuously learn consumer energy behavior and preferences to minimize energy costs, while preserving comfort by jointly training a single agent for all the building assets. The agent utilizes user feedback on the recommendations and integrates it in the reward function. Preliminary experiment results with simulated data show that the agent’s reward is increasing throughout time.