Muhammad Akram

• Machine learning and deep learning models for electricity demand forecasting in Renewable Energy Communities
• Uncertainty quantification and explainability of energy forecasts
• Operational value of forecasting for battery dispatch and energy sharing
• Optimal sizing of photovoltaic and battery systems and fair allocation of costs and benefits among community members
• Time-dependent CO₂ emission factors and environmental optimization of energy consumption
• Digital twins for the operational monitoring of energy communities

The research activity lies in the field of energy systems and concerns the thermoeconomic optimization of Renewable Energy Communities (RECs). Artificial intelligence methods are combined with energy, economic and environmental analysis, and the models are applied to real case studies in Italy, Portugal and Ireland. The activity is carried out in collaboration with KOALA S.R.L.S. and is oriented towards operational deployment.

A first strand concerns short-term electricity demand forecasting at household and community level. Hybrid models combining deep learning (LSTM) with gradient-boosting ensembles (XGBoost, LightGBM, CatBoost) have been developed, together with recursive and adaptive forecasting schemes. SHAP-based explainability is used to make the forecasts interpretable, and tariff-aware conformal prediction is used to quantify their uncertainty.

A second strand assesses the operational value of forecasts. The forecasts are coupled with battery dispatch and energy-sharing strategies, and the effect of forecast quality on the self-consumption, costs and emissions of the community is evaluated.

A third strand concerns the design and economic management of RECs. A 4E (energy, economic, environmental and equity) framework has been developed for the optimal sizing of photovoltaic and battery systems. Fair cost-allocation mechanisms among members are compared across different tariff and incentive regimes.

The environmental dimension is addressed through optimization based on time-dependent CO₂ emission factors of the power grid, applied to residential and commercial end users. Ongoing work includes digital twins based on smart-meter data for operational monitoring and DSO flexibility signalling, and the application of large language models to power and energy systems.