New Probabilistic Model from TU Delft Enhances Forecasting for Urban Residential Solar Fleets

In a significant leap forward for solar forecasting and urban energy planning, researchers at Delft University of Technology (TU Delft) have developed a novel probabilistic framework that accurately predicts the energy output of residential solar photovoltaic (PV) systems in densely built urban environments. Notably, the model operates without needing sensitive installation details—offering a privacy-preserving solution for solar fleet prediction and grid planning.

Understanding the Urban Solar Landscape

Urban landscapes, with their complex geometries, varied roof structures, and dense building configurations, present formidable challenges for estimating solar energy yields. Traditional PV yield assessment tools typically require high-resolution, installation-specific data such as panel tilt, azimuth, and location—information that is often unavailable or raises privacy concerns.

Moreover, these traditional methods rely heavily on computationally intensive processes like full horizon scanning and detailed geometric modeling, making them impractical for fleet-wide simulations or real-time applications.

A New Probabilistic Approach

Led by Dr. Hesan Ziar, the TU Delft research team has introduced a mathematical framework that addresses these limitations. The model estimates solar yields based on the probability mass function (PMF) of sunlight access, accounting for key parameters such as shading, surface roughness, and sky visibility.

Central to this approach is the Sky View Factor (SVF)—a measure of how much of the sky is visible from a given point, influencing the receipt of diffuse solar radiation. The team derived a closed-form analytical expression for SVF in complex urban geometries, allowing for swift, scalable irradiance estimation without the need for time-consuming ray-tracing calculations.

This enables the model to work with aggregated data from PV fleets (such as total installed capacity in a neighborhood) instead of detailed information on each system. The result is a fast, lightweight, and privacy-protecting solution that is well-suited for urban-scale applications.

Model Validation and Real-World Applications

The framework was validated using Digital Surface Models (DSMs) and real-world PV system data from the Netherlands. The research team tested the model in two regions of Eindhoven—one urban, one rural—comparing predictions against actual anonymized production data. In both cases, the model’s estimates closely aligned with observed results, confirming its effectiveness for fleet-level forecasting.

By reducing yield calculation times from hours to mere seconds, the model opens the door to real-time or near-real-time PV output predictions—a powerful tool for grid operators, urban planners, and energy analysts.

Transforming Urban Energy Planning

This research offers transformative potential for smart city development and grid management. Utilities and grid operators can now better anticipate solar generation at the fleet level without infringing on user privacy. In congested grid areas, this can aid in load balancing, storage planning, and proactive congestion mitigation.

Furthermore, by leveraging only limited and coarse data, the model allows for rapid assessment of solar potential in areas where comprehensive PV system information is not available or not willingly shared by owners.

Looking Ahead: AI and Beyond

Looking to the future, Dr. Ziar and his team aim to integrate this model with artificial intelligence-based irradiance forecasting tools, enabling a two-tiered prediction approach. First, AI would estimate solar irradiance under open-sky conditions; then, the probabilistic model would predict energy yields within urban environments.

Beyond energy forecasting, the researchers also envision applications in urban heat island studies, sunlight exposure assessments for public health, and automated city modeling through rapid SVF-based terrain analysis.

TU Delft’s new model represents a vital advancement in the way we forecast and manage solar PV performance across cities. With its combination of speed, accuracy, and data privacy, it sets a new benchmark for how urban solar infrastructure can be planned and integrated into tomorrow’s energy systems.