AI Building EnergyEvidence Atlas
Home / Topics
Research theme

Smart Energy Systems, EVs and Load Uncertainty

The selected papers address uncertainty at three levels: missing EV charging data, uncertain day-ahead market and demand conditions, and uncertain building design loads. Their methods span residual sequence models, distributionally robust optimization, and graph convolutional networks.

Smart building energy system connecting buildings, electric vehicles, solar PV, thermal systems, and energy networks.
Figure 1. Architecture of a smart building energy system integrating electric loads, EVs, PV, thermal equipment, and external grids.
Figure excerpt from Zhao et al. (2025), Building Simulation. DOI 10.1007/s12273-024-1219-1
Source-attributed paper figure
Graph convolutional network workflow for building design-load uncertainty quantification.
Figure 1. GCN-based design-load uncertainty-quantification method, including graph construction, uncertain-factor sampling, and CAM interpretation.
Figure excerpt from Lu et al. (2025), Building Simulation. DOI 10.1007/s12273-024-1212-8
Source-attributed paper figure
Residual dual BiLSTM-CNN network for missing electric-vehicle charging-data imputation.
Figure 5. Workflow of the proposed ResiDualNet model for electric-vehicle charging-data imputation.
Figure excerpt from Fahim et al. (2025), Building Simulation. DOI 10.1007/s12273-025-1236-8
Source-attributed paper figure

Questions addressed

  • How can EV data quality improve load forecasting?
  • How can buildings and EVs be coordinated under uncertainty?
  • How can GCNs accelerate design-load uncertainty quantification?
Source figures

Methods and applications represented in this theme

Smart building energy system connecting buildings, electric vehicles, solar PV, thermal systems, and energy networks.
Figure 1. Architecture of a smart building energy system integrating electric loads, EVs, PV, thermal equipment, and external grids.
Figure excerpt from Zhao et al. (2025), Building Simulation. DOI 10.1007/s12273-024-1219-1
Source-attributed paper figure
Graph convolutional network workflow for building design-load uncertainty quantification.
Figure 1. GCN-based design-load uncertainty-quantification method, including graph construction, uncertain-factor sampling, and CAM interpretation.
Figure excerpt from Lu et al. (2025), Building Simulation. DOI 10.1007/s12273-024-1212-8
Source-attributed paper figure
Residual dual BiLSTM-CNN network for missing electric-vehicle charging-data imputation.
Figure 5. Workflow of the proposed ResiDualNet model for electric-vehicle charging-data imputation.
Figure excerpt from Fahim et al. (2025), Building Simulation. DOI 10.1007/s12273-025-1236-8
Source-attributed paper figure

Comparison of selected papers

PaperResearch problemMethodReported evidence
Fahim et al. (2025)Missing EV charging records caused by communication or equipment failures can substantially degrade charging-load forecasts.ResiDualNet uses a residual dual BiLSTM-CNN sequence-to-sequence architecture for imputation and is evaluated before several downstream forecasting models.Across four charging datasets, the reported RMSE improvement over comparison imputation methods averages 82%; SCINet performs best among the tested forecasting models.
Zhao et al. (2025)Day-ahead scheduling of buildings and EVs must manage demand response, thermal inertia, uncertainty, and carbon-related objectives.The study formulates price-based demand response and thermal storage, then solves a Wasserstein distributionally robust Stackelberg game through KKT and duality transformations.The proposed strategy reduces total building-energy-system operating cost by 26.37% in the reported simulations and smooths the load curve.
Lu et al. (2025)Data-driven design-load models may not generalize across buildings with diverse and complex geometries.Buildings are represented as graphs of blocks, nodes, and spatial edges; a GCN predicts uncertainty statistics and a class-activation map identifies influential uncertain factors.Reported MAPE is below 6% for cooling-load statistics and below 4% for heating-load statistics, while computation for a 20-floor building drops from 331 hours to 30 seconds relative to the physics-based workflow.

Paper pages