How predictive analytics can unlock smarter positive energy districts

Positive energy districts represent one of the most promising pathways towards climate-neutral cities.

By integrating renewable energy generation, smart buildings, energy storage, electric mobility and active consumers, positive energy districts can contribute significantly to decarbonisation goals. However, managing these interconnected energy assets requires more than real-time monitoring. It requires the ability to anticipate future conditions and make informed decisions before challenges arise.

Demand side flexibility has emerged as a critical mechanism for balancing energy systems with growing shares of renewable generation. Flexible demand allows buildings, communities and energy users to adapt consumption patterns according to system needs, helping reduce peak loads, improve renewable energy utilisation, and support grid stability.

Within the InterPED project, forecasting services have been developed as part of the demand side flexibility forecasters framework. These forecasting models provide predictions of electrical demand, thermal demand, renewable energy production and electric vehicle charging behaviour across the project's four pilot sites. The forecasts serve as a foundation for intelligent energy management services that require accurate predictions to optimise operations and unlock flexibility potential.

InterPED methodology

The forecasting framework developed in InterPED adopts a hybrid data-driven approach tailored to the characteristics of different energy vectors and pilot environments.

A key challenge in forecasting flexibility is the diversity of energy assets and user behaviours. Residential electricity demand, thermal consumption, photovoltaic generation, wind energy production and EV charging patterns all exhibit different temporal characteristics and levels of uncertainty. To address this challenge, InterPED developed specialised forecasting models for each domain while maintaining a common architecture and forecasting specification across pilots.

For electrical demand forecasting, the methodology combines multiple seasonal-trend decomposition (MSTL), machine learning regression models and neural prophet forecasting techniques. This approach separates demand signals into trend, seasonal and residual components, allowing each component to be modelled more effectively. Weather forecasts, historical consumption data and temporal variables are incorporated to improve prediction accuracy.

Thermal demand forecasting models were developed for both heating and cooling applications. Historical building consumption data were combined with weather information to capture the strong relationship between climate conditions and thermal energy demand. The project evaluated both recursive and direct forecasting strategies to identify the most suitable approach for different building types.

For renewable energy forecasting, machine learning techniques were applied to predict PV and wind power generation. Since detailed technical specifications were not always available, the models relied on historical production data and meteorological forecasts to estimate future renewable output.

The framework also includes EV charging forecasting capabilities, enabling predictions of charging duration, energy consumption, and aggregated charging behaviour. These forecasts support flexibility oriented charging management and contribute to more efficient integration of electric mobility within positive energy district environments.

An important feature of the InterPED forecasting framework is its probabilistic nature. Rather than providing only a single prediction, the models generate confidence intervals that quantify uncertainty and support risk aware decision-making. This is particularly important in energy systems where forecasting errors can have operational and economic consequences.

Results and discussion

The forecasting services were applied across four diverse European pilot sites representing different energy contexts, including healthcare facilities, residential communities, energy communities and district-scale environments.

Several important findings emerged from the development and validation process.

First, data quality and historical data availability were found to be major determinants of forecasting performance. Pilots with longer historical datasets and fewer missing values generally achieved higher forecasting accuracy. Conversely, shorter datasets and higher levels of missing information increased uncertainty and reduced predictive performance.

Second, aggregated demand forecasting consistently outperformed individual household forecasting. Community-level demand profiles tend to smooth out random fluctuations associated with individual user behaviour, producing more predictable consumption patterns. This finding is particularly relevant for energy communities and positive energy districts, where flexibility services are often deployed at aggregation level rather than at individual building level.

Third, the results demonstrated that there is no universally optimal forecasting configuration. Different pilots required different model configurations depending on building characteristics, climate conditions, occupancy patterns and available datasets. This highlights the importance of adaptable forecasting frameworks capable of supporting a wide variety of local energy contexts.

One of the most significant outcomes of the work is the development of forecasting services capable of operating in real time. The forecasting workflow automatically retrieves historical energy data and weather forecasts, processes missing values, performs signal decomposition, generates predictions and delivers forecasts for a 48-hour horizon. This enables forecasting services to function as operational components within broader energy management platforms rather than remaining purely research-oriented tools.

Beyond forecasting accuracy, the broader significance of this work lies in its contribution to flexibility activation. Accurate forecasts allow optimisation services to anticipate demand peaks, identify flexibility opportunities, coordinate energy assets and support more effective use of renewable energy resources.

Within InterPED, these predictions provide essential inputs to several operational services. The model predictive controllers (MPCs) use forecasts of demand, renewable generation and thermal loads to determine optimal control actions while maintaining user comfort and system constraints. The EV flexibility orchestrator relies on charging demand forecasts to schedule electric vehicle charging in a way that supports both user needs and grid flexibility objectives. Similarly, the cross vector optimiser uses forecasts across electricity, heating, cooling, storage and mobility systems to identify optimal energy management strategies at district level. 

Without the ability to anticipate future conditions, these services would be limited to reactive operation rather than proactive optimisation.

In this way, predictive analytics becomes a foundational capability for intelligent energy management in positive energy districts.

Conclusion

The transition towards climate-neutral districts requires energy systems that are not only connected and digitalised, but also predictive. InterPED’s demand side flexibility forecasters demonstrate how advanced forecasting techniques can support this transition by providing reliable predictions of energy demand, thermal loads, renewable generation and EV charging behaviour.

The project shows that predictive analytics can significantly improve the understanding and management of flexibility resources at both building and community levels. By combining machine learning, weather information, time series analysis and probabilistic forecasting, InterPED has developed forecasting services capable of supporting real-world operational decision-making.

As the project progresses, these forecasting services will continue to be integrated within the broader InterPED ecosystem, helping unlock flexibility, improve renewable energy integration and support the development of scalable positive energy district solutions across Europe.

References

1. InterPED Consortium, 2026. D6.1 Demand-side Flexibility Forecasters. 
2. European Commission. Positive Energy Districts and Neighbourhoods Initiative.
3. InterPED Grant Agreement No. 101138047, Horizon Europe Programme.

About the authors