EP3 Solutions

AI’s rapid success in arts and communication has been striking—but as someone who’s worked in building energy simulation for over 15 years, I’m finding that energy modeling is still figuring out where AI fits in.

The difference is fundamental and worth examining closely.

The Creative vs. Technical AI Divide

AI has gained rapid adoption in creative fields because validation is straightforward: does the image convey the right message? Does the copy resonate? While AI can’t truly innovate and carries inherent biases from its training data, creative creative applications often succeed on whether they ‘look right’—and that intuitive standard is frequently enough.

Energy modeling operates under different constraints. We’re not just seeking plausible results; we need verifiable, defensible outcomes that can withstand peer review, regulatory scrutiny, and real-world performance validation. The stakes are higher—these models inform expensive building decisions and energy code compliance.

The Persistence of Established Simulation Engines

This is why traditional, peer-reviewed simulation engines like EnergyPlus aren’t going anywhere. These tools have decades of validation behind them, extensive documentation, peer-review, and proven track records. They represent the engineering rigor our industry requires.

But here’s what’s changing rapidly: how we interact with these engines.

The Hybrid Approach: AI as Interface, Not Engine

We’re observing an interesting evolution in our EP3 user base. Many are now using our IDF import functionality to validate, edit, and understand AI-generated building models. They’re leveraging AI for initial model creation, then building on those models—modifying, applying efficiency measures, comparing alternatives and applying traditional validation methods to ensure accuracy.

This hybrid workflow acknowledges both AI’s strengths and limitations. AI excels at rapid prototyping, pattern matching, and handling routine modeling tasks. Traditional simulation engines provide the verification and precision our industry demands, and the integration with EP3 ensures both that modelers can understand and verify the appropriateness of the AI generated model, and that they can build on the model—adding efficiency measures, fine-tuning and iterating.

Early Results from AI Integration

Our internal experiments with AI integration in EP3 have been eye-opening. We’ve developed an early-stage AI chatbot that can modify energy models through natural language commands, and the results have exceeded our expectations.

The system currently handles several key modeling tasks effectively:

• Schedule Generation: Creating complex operational patterns from simple descriptions. Example prompt: “typical office occupancy with extended operation for weekend cleaning”
• HVAC System Modifications: Adjusting system parameters, control sequences, and operational strategies. Example prompt: “change the operation of this heat pump so that it can cycle on all of the time, but fans run continuously from 8am to 6pm on weekdays”
• Construction and Material Creation: Creation of building assemblies and associated materials through descriptions. Example prompt: “create a construction and all required materials for an exterior wall with 2-wythe brick cladding, an air gap, R-20 interior insulation (given in IP units) and gypsum board”
• Daylighting Control Integration: Adding automated lighting controls with appropriate sensor placement. Example prompt: “add daylighting controls to the Space SPACE3-1”
• Power Density Modifications: For example, making changes to one or more lighting object. Example Prompt: “reduce the lighting power density of all selected lights by 20%”

After each prompt, the EP3 chatbot suggests modifcations, which the uesr can accept, edit or reject before they’re integrated into the model.

What’s particularly valuable is that these modifications can either update existing Energy Efficiency Measures (EEMs) or generate new comparative scenarios—all while maintaining the underlying simulation integrity that EnergyPlus provides.

The main goal of the AI interface is making EP3 more accessible by handling tasks that traditionally require deep domain knowledge. HVAC system operation, for instance, often involves complex control sequences and interdependent parameters that can be challenging even for experienced modelers. The AI chatbot can interpret high-level operational requirements and translate them into the specific EnergyPlus objects and relationships needed. EP3 then interprets these objects and integrates them into the selected EEM—making these advanced modeling capabilities available to a broader range of users.

The Validation Challenge

The speed of these AI-assisted modifications is remarkable, but it highlights a crucial challenge: how do we maintain our industry’s necessary rigor while embracing AI’s efficiency gains?

Our approach at EP3 is to leverage AI’s ability to answer simple questions about EnergyPlus, and then integrate those insights into EP3 models. At the end of the day, EP3 is an EnergyPlus user interface, and generates EnergyPlus models. All the validation pathways that the industry has developed over decades remain fully available and can be applied without modification.

This represents a fundamentally different approach from AI systems that generate results through opaque processes. When AI tools modify lighting schedules or HVAC parameters within an established simulation framework, the output remains as verifiable as traditionally-created models. The simulation engine, reporting methods, and verification approaches are unchanged—efficiency gains come from improved model creation and modification interfaces.

The Broader AI Integration Landscape

We’re not the only ones exploring AI integration in building energy simulation. Other platforms are taking different approaches that are worth examining. Pollination is using AI to assist with the notoriously difficult task of translating Revit models to clean energy modeling geometry. BESOS has been trained on extensive EnergyPlus models and uses this learning to perform surrogate models.

Looking Forward: Questions for Our Community

As we continue developing AI integration, we’ll continue exploring these themes, with the end-goal of supporting the modeler without sacrificing quality or transparency.

• Where do you see AI in your energy simulation workflow?
• What AI tools are you currently using?
• Where does your existing energy simulation process bog you down?
• What tasks do you find yourself repeating?
• How comfortable are you with asking AI to suggest edits to your model?

Your insights will directly inform where we focus our development efforts as we continue to work with AI, so we’re genuinely curious about your experiences and priorities.