Modelos Generativos Explicáveis para Dados Sintéticos de Mercados Locais de Energia Elétrica

New LEM models are being proposed to enable the active participation of consumers, as means to balance the increasing generation stochasticity brought by the widespread of renewable generation [T2], [T3], [6]. The analysis of future LEM potential impacts is, however, constrained by: (i) the lack of specific LEM simulators that enable the experimentation and comparison of different LEM models and structures; (ii) the existing data and scenarios, which is limited concerning future evolution of the multiple energy resources and privacy/security constraints. Emergent AI models for synthetic data generation are promising solutions to overcome this problem [2], [9]. Motivated by the recent advances in image and language generation models (e.g. GPT-based models), the development of synthetic time-series data generation models should incorporate the latest advances in the field in order to reach representative synthetic datasets. The lack of trust in computer-generated data is a significant issue that arises from the use of synthetic-generated data. XAI [14] has the potential to play an important part in endowing the data generation process with explanations that reflect the reasons behind the creation of specific data. GENESIS will take a significant step further by proposing new models for i) generating synthetic time-series data, representative of real data sets, encompassing a contextual dimension and surpassing the ethical limitations of real data; ii) enlarging the available data sets to enable the analysis of new LEM models; (iii) explaining the data generation process in a flexible and adaptive way according to different types of users, of explanation types and of execution constraints; (iv) simulating LEM in a coherent and user-friendly way, bringing together a comprehensive and representative set of the most relevant LEM models proposed in the literature and in related projects. The GENESIS general concept is depicted in Figure Concept.pdf in annex. Accordingly, the GENESIS specific objectives are: - To conceive and develop novel contextual synthetic time-series data generation models, departing from the latest advances in the generative AI domain - To generate new synthetic datasets, representative of real data, regarding different energy resources, as means to enable the simulation of future LEM scenarios and diverse PES related studies - To develop adaptive and flexible XAI models to support the understanding of the data generation process by users with different know-how and characteristics - To collect, adapt, refine and consolidate the most relevant and promising alternative LEM models and structures - To develop a LEM simulator that makes available the different LEM models in a way to facilitate the experimentation and comparison of alternative market models - To assess the performance of the developed generative models by analysing the quality of the generated synthetic data, and assess the impact of the LEM models using scenarios based on the generated datasets - To validate the suitability of the provided explanations, of the generated datasets and of the overall quality of the LEM simulator through the feedback of different types of users, including experts and industry players active in the field - To disseminate the project results among the scientific community, the industry, the general public and other relevant stakeholders with potential interest in the project subjects