On November 27, Sergio Ramos, Associate Professor of GECAD at Polytechnic Institute of Porto, and Joao Soares, a GECAD researcher at the same institution, participated in our university's "Frontiers in Intelligent Construction and Smart City Research" forum via online platform. They delivered two insightful lectures titled "Mini-hydro Power Plants" and "Towards Explainable AI-driven Optimization for Complex Energy Problems," respectively. The presentations delved into the pivotal role of artificial intelligence (AI) in addressing modern energy system optimization challenges, systematically outlining current obstacles, innovative solutions, and future trends, which garnered significant attention from the industry.
To tackle energy optimization challenges, Professor Sergio identified AI as the game-changer. His lecture introduced the widely recognized "no free lunch theorem" in energy systems—no algorithm can achieve optimal performance across all optimization scenarios. Significant performance variations across different contexts pose major challenges for smart grids that rely on efficient scheduling. With the increasing share of renewable energy and heightened electricity demand fluctuations, traditional optimization methods struggle to handle complex energy systems characterized by high dimensions, nonlinearity, and strong uncertainty. The current energy challenges are: numerous variables, high uncertainty, time-sensitive requirements, and pronounced nonlinearity.
To tackle these challenges, Professor Sergio developed an innovative solution: AI + probabilistic modeling + stochastic optimization. The research team proposed a new paradigm called "AI-driven probabilistic-stochastic collaborative optimization" to build a next-generation framework. This approach uses probability distributions to characterize uncertainties, incorporates wind/solar forecasting errors and load fluctuations into the modeling process, leverages AI's powerful fitting capabilities to handle high-dimensional nonlinear relationships, and achieves rapid solution under complex constraints. Additionally, it introduces a stochastic search mechanism to dynamically select optimal strategies from over 500 optimization algorithms, enhancing global optimization performance.
Subsequently, Professor Sergio's research team compared multiple AI approaches including genetic algorithms, particle swarm optimization, and reinforcement learning. The results demonstrated that in scenarios with severe price volatility, the reinforcement learning model based on Deep Q Network (DQN) performed optimally. Conversely, during periods of relatively stable supply and demand, the improved particle swarm optimization algorithm exhibited faster convergence and higher stability. This finding validates the core principle that' no algorithm is inherently superior—success depends on problem matching, 'providing a scientific basis for the industry to' select the right tool for the task.'
Notably, the HIDE (Hybrid Intelligent Differential Evolution) algorithm independently developed by João Soares 'team has demonstrated exceptional performance across multiple tests. This innovative approach combines differential evolution's global search with local reinforcement learning mechanisms, delivering: rapid convergence (40% fewer iterations than traditional methods) and balanced exploration-utilization to avoid local optima, making it suitable for optimization problems in power grid dispatching, energy storage configuration, and microgrid planning. By incorporating techniques like SHAP value analysis and attention mechanism visualization, the AI model transforms from a "black box" into a comprehensible and verifiable decision-support system.
Looking ahead, João Soares' research team has proposed a vision for "LLM-driven automated optimization modeling." While still in its early stages, this approach has already achieved initial success: users can describe optimization goals and constraints in natural language, and the LLM automatically generates corresponding mathematical models and solution code. By integrating knowledge graphs and historical cases, it enables intelligent recommendations across "model-algorithm-data" ecosystems. João Soares advocates for collaborative efforts through open datasets, international algorithm competitions, and strengthened industry-academia-research collaboration to advance AI-powered energy optimization from lab research to large-scale engineering applications.
In his concluding remarks, Joao Soares emphasized that AI is fundamentally transforming the operational logic of energy systems. The paradigm shift—from "experience-based dispatch" to "intelligent optimization," and from "passive response" to "proactive forecasting" —not only offers innovative solutions to complex energy challenges but also injects robust momentum into building a secure, efficient, and sustainable energy future. As technology continues to advance, an AI-powered era of smart energy is rapidly approaching.
