The presence of intermittent sources of renewable energy in power systems requires ESSs to manage temporal imbalance in energy supply and demand. In this study, we introduce a hybrid energy storage system (HESS) coupled with an AI energy management system (EMS) that uses deep reinforcement learning (DRL) for optimal scheduling of renewable energy utilization within grid-connected and islanded microgrids. AI-enabled EMS utilizes a DRL agent with proximal policy optimization (PPO) to make optimal decisions regarding energy generation based on state space and economic considerations, while accounting for SoC constraints of batteries. An important aspect of the proposed system is the design of HESS architecture and reward function based on DRL. Further improvements are made via analysing the PPO clipping sensitivity, Pearson correlation analysis on the relationship between the intermittency of renewables and response latency, and Monte Carlo uncertainty analysis with a 95% confidence interval. For a 24-hour simulation period, the developed system is able to cut down on grid power imports by 43.2%, have an 87.3% renewable energy utilization rate, extend the lifespan of the battery from 8.1 to 12.5 years, and have 91.4% peak shaving efficiency through 100 Monte Carlo runs and without any SoC violations (25%–90%). Net benefit analysis is estimated to be $56,000–$66,000 for 15 years at a 6% discount rate, while a 120 ms response time and one-way ANOVA with Tukey's HSD confirm statistical significance.
Lalit Sachdeva, U. Anand· Energy Storage and Conversio...· 0 citations
The rapid advancement of intelligent networking technologies, such as 6G systems, the Internet of Things (IoT), and edge-enabled communication technologies, raises multiple challenges regarding user data privacy, network reliability, and automated self-healing mechanisms for quickly fixing network failures. A classical paradigm of centralized machine learning (ML) algorithms requires the collection and aggregation of data, resulting in significant data privacy challenges and increased network delays. This research proposes the first framework of Privacy-Preserving Federated Learning (F-PF-ML) for decentralized edge intelligence, specifically designed for self-healing autonomous networks. This model allows the distributed edge nodes to train the global network intelligence models collaboratively while preserving the privacy of the data and enhancing the adaptability of the network. The framework's innovative design combines federated learning (FL), differential privacy (DP), and an adaptive anomaly detection module integrated across decentralized edge nodes. Each node conducts local training using the network traffic data in real-time and updates its model with encrypted or compressed data, which is sent to the federated aggregation server. The self-healing control module uses the aggregated data to determine the presence of an anomaly and automatically triggers an action to recover the network, such as rerouting and reallocating the network resources dynamically. The proposed model was evaluated in a large-scale edge network environment consisting of 200 nodes and 15,000 network transactions to simulate diverse traffic patterns and potential attack conditions. Based on the findings, it can be stated that the suggested model using PPFL-based architecture helps achieve 94.3% of accuracy with anomaly detection and helps with reduced data transmission overhead by 38%, and also helps with reducing the network recovery time by 27% when compared with the conventional methods of networking management systems with AI systems in the background. This also helps in bridging the gap that exists for the practical application of the proposed federated learning models on the self-operating networks.
Lalit Sachdeva, Utkarsh Anand· International Conference Com...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
