AI-assisted Energy Portfolio Optimization Toolkit for Renewable Energy-rich Grid Management

• Auteur / Autrice : Aditya narayan Sankaran
• Direction : Badii Jouaber, Reza Farahbakhsh
• Type : Projet de thèse
• Discipline(s) : Informatique, données, IA
• Date : Inscription en doctorat le 15/10/2024
• Etablissement(s) : Institut polytechnique de Paris
• Ecole(s) doctorale(s) : École doctorale de l'Institut polytechnique de Paris
• Partenaire(s) de recherche : Laboratoire : SAMOVAR - Services répartis, Architectures, Modélisation, Validation, Administration des Réseaux

Mots clés

Mots clés libres

Artificial Intelligence Energy Forecasting

Time-Series Data; Predictive Models

Grid Management

Résumé

Context As per the International Energy Agency (IEA), the energy sector accounts for more than three-quarters of greenhouse gas emissions globally. Replacing coal, gas, and oil-fired power with renewable energy (RE) sources, such as wind and solar, would be key to dramatically reduce carbon emissions and meet the net-zero targets. Therefore, it is imperative to bolster the operations of the energy sector with RE-rich generation resources. However, the RE-rich grids are currently facing significant challenges in ensuring a stable and resilient energy supply due to the drastic and unpredictable variations in the energy supply from RE sources and also incurring significant penalties due to real-time deviations exceeding the permissible errors on the demand and supply of the energy. The rise in unpredictability on both supply and demand sides calls for the intervention of Artificial Intelligence (AI) based modeling and intelligent energy analytics. Abstract of the Research Project Our aim is to build the energy portfolio optimization toolkit for the RE-rich grid operators that enable accurate energy forecasting using AI/ML-based models, data-driven energy scheduling, and efficient power procurement decisions and augment with the blockchain technology to manage the energy operations database. Overall, we envision such a framework will help improve grid energy balancing, reliability, and maximize the cost-effectiveness, and transparency of operations to enable successful transition toward the cleaner and sustainable power system. Accordingly, in this project, we will first build an ensemble of RE forecasting models that provide more reliable and accurate energy forecast data. Second, we will develop a multi-objective stochastic optimization-based scheduling platform to optimally plan the power procurement decisions based on the prevailing power generation and load demand. Third, we will build an immutable energy database to realize transparent, traceable, and secure decision-making for RE grid operations. Scientific Objectives of the PhD To assist the grid operators in matching the demand and supply of power in the grid, the accurate forecasting of the real-time and day-ahead renewable energy generation, the load demand to be supplied and prevailing electricity market prices is required. However, renewable energy generation processes, load and price variations can be very hard to model due to their inherent natures of intermittency, stochasticity and fluctuation. A fundamental step in this process involves the creation of a data standardization pipeline to unify large-scale time-series data from disparate sources and thereby facilitating intelligent data analysis and the subsequent development of predictive models. The development of models should encompass the investigation of a wide range of methodologies, including shallow models, deep learning models, and hybrid approaches. Given the time-series nature of the data, it is also essential to implement a continuous learning pipeline and establish monitoring metrics to ensure that the models remain up-to-date to deliver robust forecasting results. Keywords: Methodology and Timeline This project will be jointly worked by the India and French Academia and Industry partners for the duration of 3 years. Accordingly, there are four key work packages in this project. We plan to develop the key AI models and energy scheduling toolkit by the end of second year with preliminary testing at the Institutional infrastructure. Further, the Industry partners in India will help us take the prototype developments for the deployment, testing and validation at the Industry scale testbed for the period of third year to ensure the quality of our deliverables to reach the desired level TRL-4. By the end of the project, we aim to have impactful publications in relevant top tier conference and journals, and deliver the i) Ensemble of AI-models ii) Algorithm for Grid Scheduling iii) Blockchain database and all the necessary transfer learning reports.