It integrates multi-agent systems, predictive models, and reinforcement learning to optimize energy management in rural prosumer communities. | DemIA International Chair in Trustworthy AI and the Demographic ChallengeInternational Chair in Trustworthy AI and the Demographic Challenge

ARISE

Why is it important?

The energy transition drives prosumer communities that generate, store, and trade electricity. However, their smart management remains a challenge, particularly in rural environments.

Project objective

To create AI models capable of optimising energy management in rural prosumer communities through reinforcement learning and predictive models, improving consumption planning, photovoltaic production, and PVPC electricity tariff forecasting.

How does it work?

ARISE recreates the operation of a rural energy community through a simulation environment where each household is represented as a smart agent, integrating demographic information, residential consumption profiles, photovoltaic production, and weather data retrieved from various sources.

Reinforcement learning models capable of learning optimal energy management strategies are trained within this environment. In parallel, PVPC tariff forecasting models are developed using time-series techniques, providing data that enhances decision-making for the agents.

The platform combines specialised tools such as PyTorch, PyTorch Forecasting, RLLib, Gymnasium, PettingZoo, LoadProfileGenerator, SAM, PVGIS, and NASA POWER, enabling the construction of a comprehensive ecosystem for the simulation and optimisation of rural energy communities.

Research lines:

Simulated energy community: A model representing rural communities, taking into account the number of households and their demographic profile.

Energy simulation: Tools that generate realistic consumption and photovoltaic production profiles using historical and weather data.

Predictive models: Development of advanced models (TFT, LSTM) for forecasting the PVPC tariff and other relevant variables.

Artificial Intelligence: Single-agent and multi-agent reinforcement learning systems to optimise energy decisions in dynamic environments.

Technological ecosystem: Integration of Deep Learning libraries and tools, energy simulation, and RL into a reproducible experimental platform.

Project innovations:

Incorporation of rural demographic variables into reinforcement learning systems.
Integration of predictive models within the state of smart agents to improve their performance.
Development of action masking mechanisms that enhance the convergence of RL models.
Expansion of the action space to represent more realistic and complex energy scenarios.
Pioneering research in forecasting the PVPC electricity tariff using AI.

Link to the Chair and the Demographic Challenge:

ARISE addresses the demographic challenge by developing AI solutions to optimise energy management in rural communities in south-western Castile and León. Its approach promotes a more efficient and sustainable energy model adapted to the territory, boosting energy self-sufficiency and improving the quality of life and resilience of areas affected by depopulation.