Revolutionizing Heating Planning: A Data-Driven Approach for Accurate and Scalable Energy Consumption Predictions

Accurately predicting heating energy consump-tion for buildings is crucial for optimizing com-munal heating planning. However, the lack of detailed building information hinders precise estimations necessary for city decarbonization and efficient use of local renewable or waste energy. To address this challenge, this study utilizes 3D geometries and open data of Berlin's half a million buildings to perform building-specific simulations and trains AI-models to predict heating consumption.A physics library was developed using statisti-cal data from 16 building types and 6 refur-bishment scenarios, providing building model components with physical properties. Addition-ally, a usage library comprising 24 different usage types and a weather database spanning 13 recent years, and 3 future climate scenarios for the next 8 decades were created. Simula-tions on a subset of Berlin's buildings were conducted using these libraries.To extend results city-wide, an Artificial Neural Network (ANN) was trained using input data and simulation outcomes. This approach re-places traditional simulation software, offering a more efficient alternative. An intuitive web-application with a fast AI surrogate model at its core allows individuals to access and modify results for their buildings as well as correcting values to reflect actual demands. Corrected data is incorporated into neural network train-ing, improving accuracy for various climate scenarios and potential refurbishments.This iterative process not only streamlines communal heating planning in Berlin but also establishes a flexible framework for continuous improvement. This emphasizes the adaptability and scalability of our proposed methodology, contributing to enhanced predictions and a more efficient approach to city-wide heating planning.