DySENSE
A multimodal generative-AI framework for urban energy consumption data generation that decouples static urban morphology from dynamic monthly energy consumption and jointly generates spatially aligned satellite imagery and energy consumption maps, supporting downstream tasks such as energy-efficient urban planning and building performance simulation.
Timeline (last 90 days)
The DySENSE framework decouples urban problems into long-term stable urban form (static layer) and monthly energy consumption that fluctuates with seasonal climate (dynamic layer).
DySENSE supports two conditional generation settings: conditioned on static land use, generating realistic and spatially aligned satellite imagery and energy consumption maps; and conditioned on monthly and climate prompts, keeping urban form unchanged and jointly generating only dynamic monthly energy consumption maps.
DySENSE adopts multimodal joint generation, outputting multiple spatially aligned results (satellite imagery and energy consumption maps) in a single forward pass rather than staged serial generation.
The main technical difficulty of DySENSE is the inter-modality alignment constraint: building outlines in satellite imagery must be pixel-level consistent with high-value regions in energy consumption maps.
DySENSE was validated on four cities: New York, Washington, Norfolk, and Busan.
Using less than 25% of labeled data, DySENSE can generate labeled high-fidelity synthetic energy consumption data and improve downstream prediction performance by 8% mIoU.
Compared with existing methods, DySENSE improves prediction performance by 8.7% (accuracy) and reduces prediction error by 9%–18% (CVRMSE, ASHRAE standard).
DySENSE's application directions include energy-efficient urban planning, peak load analysis, building performance simulation, and city-scale dynamic demand response.
The value of synthetic data ultimately depends on whether its deviation from the real world is controllable, and performance under cross-city generalization and extreme climate scenarios still requires more independent validation.
All Facts (9)
The DySENSE framework decouples urban problems into long-term stable urban form (static layer) and monthly energy consumption that fluctuates with seasonal climate (dynamic layer).
50%UnverifiedDySENSE supports two conditional generation settings: conditioned on static land use, generating realistic and spatially aligned satellite imagery and energy consumption maps; and conditioned on monthly and climate prompts, keeping urban form unchanged and jointly generating only dynamic monthly energy consumption maps.
50%UnverifiedDySENSE adopts multimodal joint generation, outputting multiple spatially aligned results (satellite imagery and energy consumption maps) in a single forward pass rather than staged serial generation.
50%UnverifiedThe main technical difficulty of DySENSE is the inter-modality alignment constraint: building outlines in satellite imagery must be pixel-level consistent with high-value regions in energy consumption maps.
50%UnverifiedDySENSE was validated on four cities: New York, Washington, Norfolk, and Busan.
50%UnverifiedUsing less than 25% of labeled data, DySENSE can generate labeled high-fidelity synthetic energy consumption data and improve downstream prediction performance by 8% mIoU.
50%UnverifiedCompared with existing methods, DySENSE improves prediction performance by 8.7% (accuracy) and reduces prediction error by 9%–18% (CVRMSE, ASHRAE standard).
50%UnverifiedDySENSE's application directions include energy-efficient urban planning, peak load analysis, building performance simulation, and city-scale dynamic demand response.
50%UnverifiedThe value of synthetic data ultimately depends on whether its deviation from the real world is controllable, and performance under cross-city generalization and extreme climate scenarios still requires more independent validation.
50%