Self-Optimizing Power Converters using Hybrid DRL and LSTM-Driven MPPT For IoT Energy Harvesting

A hybrid CNN–LSTM-attention network for adaptive energy management of battery–supercapacitor systems in electric vehicles

Integrating renewable energy and advanced storage and management systems are vital solutions to overcome variability, reliability, and efficiency issues in modern power systems. This article introduces the intelligent energy management system (IEMS) which integrates photovoltaic (PV) generation, lithium-ion batteries (Li-ion) and supercapacitors (SCs) are controlled by a CNN–LSTM-attention mechanism. The hybrid configuration leverages complimentary qualities: batteries have a high density of energy for sustained supply, while SCs deliver high-power density with millisecond charge–discharge response, ideal for handling sudden load variations. The CNN–LSTM-attention mechanism is trained with historical and real-time data of solar power, temperature, load demand, and state of charge (SoC). It predicts power generation and utilization to optimize allocation: PV energy is prioritized during peak availability, SCs manage transient surges up to 300% of rated current, and batteries stabilize long-term operation. This strategy reduces rapid battery cycling, extending lifespan, while enhancing system efficiency and minimizing losses. The system is tested with MATLAB/Simulink simulation of PV array, Li-ion, and SC with their performance. The results have shown that the proposed system is more efficient than the traditional rule-based control system, it has reduced the peak current stress on battery, and the voltage regulation is achieved during dynamic loads. The results highlight the resilience and adaptability of the CNN–LSTM-attention-based IEMS to the application in solar powered electric vehicles. An experimental prototype of 100 W also developed to check the feasibility of the proposed system.

Intelligent Task Offloading and Energy Management for Battery-Less 6G Industrial IoT Using Multi-Agent Deep Reinforcement Learning

A constrained, risk-sensitive multi-agent reinforcement learning framework is presented for joint task offloading and EH scheduling in battery-less 6G industrial networks, yielding a Pareto-non-dominated, statistically validated policy.

Techno-economic sizing of constrained microgrids using multivariate deep learning and Dc-coupled solar-plus-storage integration

The transition toward reliable and sustainable microgrids in constrained power systems requires both accurate load demand forecasting and the intelligent resource coordination. The primary objective of this study is to optimize the capacity sizing and operational resilience of such systems by proposing an Artificial Intelligence-driven Load Forecasting and Battery-Integrated Energy Management approach, structured around a DC-coupled solar-plus-storage (DC-CS-P-S) architecture. To accurately capture the complex thermodynamic and behavioural patterns driving electricity consumption, a multivariate long short-term memory (LSTM) network was developed. The proposed strategy physically decouples energy storage recovery from the grid by dedicating solar photovoltaic generation strictly to charging the battery energy storage system to ensure availability for peak shaving and essential load protection. Evaluated over an 8760-hour simulation, the LSTM forecasting achieved high predictive accuracy with a mean absolute percentage error of 3.31% and an RMSE of 12.39 kW. Under severe bottleneck constraints, the DC-coupled battery actively contributed 93.64 kW of peak shaving power, successfully diverting 100% of the energy deficit (135 020 kWh) entirely to non-essential infrastructure. Furthermore, a comprehensive techno-economic parametric analysis, incorporating a value of lost load reliability penalty, and a $50/ton carbon price, identified 330 kW as the absolute optimal grid capacity. At this threshold, the microgrid achieves complete reliability with zero energy shedding while maintaining a minimized baseline operational costs of $92 033 USD. Ultimately, the results demonstrate that intelligently maximizing zero-emission resources actively suppresses compounding carbon liabilities (averting a 21.6% financial premium), providing system planners with a resilient, data-driven methodology to optimize infrastructure investments without over-sizing centralized grid connections.

A Hybrid Framework for Energy-Efficient Sensor Management in RF-Powered IoMT Systems

A Hybrid SCO–BiLSTM–DRL Framework for Intelligent Smart Grid Energy Management and Energy Demand Forecasting

A hybrid Sine Cosine Optimization–Bidirectional Long Short-Term Memory (BiLSTM)–Deep Reinforcement Learning (DRL) framework for smart grid energy management that improves energy utilization efficiency, reduces operating costs, and enhances electrical grid stability is proposed.

AI-driven optimization: revolutionizing energy efficiency in modern buildings.

This work establishes a novel continuous-time dynamic-policy learning paradigm that integrates predictive modeling with real-time adaptive control, advancing data-driven intelligent building operation toward sustainable and autonomous energy management.