Add data center models by bayc · Pull Request #929 · NatLabRockies/H2Integrate

bayc · 2026-10-13T16:31:28Z

Add data center performance and cost models

This PR adds two data center model pairs in h2integrate/converters/data_center/. Both produce the compute_load commodity (MW) and calculate the electricity and water a data center needs to serve a compute load profile.

Model	Type	Description
`DataCenterPerformanceModel`	Performance	Electricity demand from a compute electrical efficiency and a cooling load ratio
`DataCenterCostModel`	Cost	CapEx, fixed OpEx, and variable OpEx per MWh of compute load
`DataCenterPUEWUEPerformanceModel`	Performance	Electricity and water demand from Power Usage Effectiveness (PUE) and Water Usage Effectiveness (WUE)
`DataCenterPUEWUECostModel`	Cost	CapEx, fixed OpEx, and electricity and water costs

The PUE/WUE model takes PUE and WUE in one of three ways:

Directly: set pue and wue.
By IECC climate zone: set climate_zone. PUE/WUE are then looked up from the climate- and technology-specific estimates of Lei and Masanet (2022) and Lei, Shehabi, and Masanet (2025). The table covers 12 cooling configurations (see below), each with an efficient (5th-percentile) and an inefficient (95th-percentile) value.
From the site location: the climate zone is inferred from the plant latitude/longitude using eeweather.

If cooling_configuration isn't set, the model picks the configuration with the lowest PUE, or the lowest WUE when optimize_for: wue. The choice can be limited by size category using size_sqft.

data_center:
  performance_model:
    model: DataCenterPUEWUEPerformanceModel
  cost_model:
    model: DataCenterPUEWUECostModel
  model_inputs:
    shared_parameters:
      system_capacity_mw: 80.
    performance_parameters:
      compute_it_workload_profile: 60.0
      cooling_configuration: 1
      efficiency_level: efficient
    cost_parameters:
      cost_year: 2026
      capex_per_mw: 4.0625e+6
      fixed_opex_per_mw_per_year: 1.5e+5
      electricity_rate: 0.0
      water_rate: 0.0

Example examples/38_data_center/ runs both model pairs: a grid-connected data center with an hourly AI training workload (compute_load_profile.yaml).

Table 1: Cooling configurations available in data center model

Case	Size category	Cooling configuration
1	Large-scale	Airside economizer + adiabatic cooling + (water-cooled chiller)
2	Large-scale	Waterside economizer + (water-cooled chiller)
3	Midsize	Airside economizer + (water-cooled chiller)
4	Midsize	Waterside economizer + (water-cooled chiller)
5	Midsize	Water-cooled chiller
6	Midsize	Airside economizer + (air-cooled chiller)
7	Midsize	Air-cooled chiller
8	Small	Water-cooled chiller
9	Small	Air-cooled chiller
10	Small	Direct expansion system
11	Large-scale or midsize	Direct-to-chip liquid loop + dry cooling + adiabatic cooling + (air-cooled chiller)
12	Large-scale or midsize	Direct-to-chip liquid loop + waterside economizer + (water-cooled chiller)

References:

Lei, N., & Masanet, E. (2022). Climate-and technology-specific PUE and WUE estimations for US data centers using a hybrid statistical and thermodynamics-based approach. Resources, Conservation and Recycling, 182, 106323.
Lei, N., Lu, J., Shehabi, A., & Masanet, E. (2025). The water use of data center workloads: A review and assessment of key determinants. Resources, Conservation and Recycling, 219, 108310.

Note: GitHub Copilot was used in the development of this PR.

Section 1: Type of Contribution

Section 2: Draft PR Checklist

Open draft PR
Describe the feature that will be added
Fill out TODO list steps
Describe requested feedback from reviewers on draft PR
Complete Section 8: New Model Checklist (if applicable)

TODO:

Replace the PR XYZ placeholder in CHANGELOG.md with this PR's number
Rebuild the docs and regenerate the class hierarchy diagram

Type of Reviewer Feedback Requested (on Draft PR)

Structural feedback: Should the PUE/WUE lookup data (pue_wue_data/pue_wue_data.csv) and the eeweather climate-zone lookup live in the converter folder, or somewhere shared?

Other feedback: The PUE/WUE cost model has electricity_rate and water_rate. When electricity and water are bought through separate techs (grid_buy, water_feedstock), those rates count the same costs a second time. The example sets them to 0. Should the rates stay?

Section 3: General PR Checklist

Section 4: Related Issues

N/A

Section 5: Impacted Areas of the Software

Section 5.1: New Files

h2integrate/converters/data_center/__init__.py
- Exports the four data center models.
h2integrate/converters/data_center/data_center.py
- DataCenterPerformanceModelConfig, DataCenterPerformanceModel: compute load capped at capacity; electricity demand from compute efficiency and cooling load ratio; water demand per MWh.
- DataCenterCostModelConfig, DataCenterCostModel: CapEx, fixed OpEx, and variable OpEx per MWh.
- DataCenterPUEWUEPerformanceModelConfig, DataCenterPUEWUEPerformanceModel: facility power and water from PUE and WUE.
  - determine_iecc_climate_zone: infers the IECC climate zone from the site lat/lon using eeweather.
  - determine_pue_wue_by_climate_zone, _classify_size: look up PUE/WUE by climate zone, cooling configuration, efficiency level, and size category.
- DataCenterPUEWUECostModelConfig, DataCenterPUEWUECostModel: CapEx, fixed OpEx, and electricity and water costs.
h2integrate/converters/data_center/pue_wue_data/pue_wue_data.csv
- PUE/WUE estimates by climate zone and cooling configuration (Lei and Masanet, 2022).
h2integrate/converters/data_center/test/__init__.py, h2integrate/converters/data_center/test/test_data_center_models.py
- Unit tests for both model pairs: performance, cost, climate-zone lookup, lat/lon lookup, and the error when site data is missing.
examples/38_data_center/ (run_data_center.py, driver_config.yaml, plant_config.yaml, plant_config_pue_wue.yaml, tech_config.yaml, tech_config_pue_wue.yaml, data_center.yaml, data_center_pue_wue.yaml, compute_load_profile.yaml)
- Grid-connected data center example that runs both model pairs with an hourly compute load profile.

Section 5.2: Modified Files

examples/test/test_all_examples.py
- test_data_center_example: runs both configurations in example 38 and checks key outputs.
pyproject.toml
- Adds the eeweather dependency, which the climate-zone lookup needs.

Section 6: Additional Supporting Information

New dependency: eeweather is only used when PUE/WUE and climate_zone are both omitted and the climate zone is inferred from the site lat/lon.
Double counting: in example 38, electricity_rate and water_rate are set to 0 because electricity and water are bought through grid_buy and water_feedstock. The docs page explains this.
Follow-on PR: a follow-up PR will add data center waste heat recovery, a heat pump model, and a district heating demand model.

Section 7: Test Results, if applicable

pytest h2integrate/converters/data_center/test examples/test/test_all_examples.py::test_data_center_example
10 passed

Example 38 LCOC: $190.79 (simple model) and $169.45 (PUE/WUE model).

Section 8 (Optional): New Model Checklist

- DataCenterPerformanceModel/DataCenterCostModel: compute load model using electrical efficiency and cooling load ratio - DataCenterPUEWUEPerformanceModel/DataCenterPUEWUECostModel: PUE/WUE-based model with IECC climate-zone lookup (Lei & Masanet 2022) via eeweather - Example 38_data_center, unit and example tests, docs page - Add eeweather dependency

johnjasa

Thanks for this, Chris! Love the docs and references to sources. Glad to get data center modeling in H2I! Submitting a partial review with some blocking changes now, will come back to this later on.

johnjasa · 2026-10-15T18:44:53Z

+load profile. An example that runs both model pairs with an hourly AI training workload
+is provided in `examples/38_data_center/`.


Could you please add where the training workload profile came from? Given all the questions we get!

johnjasa · 2026-10-15T19:21:00Z

+            "electricity_in",
+            val=0.0,
+            shape=n_timesteps,
+            units="MW/h",


The units are incorrect here; they should be commodity_rate_units; anything upstream that would be connected to this would fail with a units mismatch

johnjasa · 2026-10-15T19:23:19Z

+        outputs["unmet_electricity_demand_out"] = total_electricity_demand - electricity_used
+        outputs["water_consumed"] = water_consumed
+        outputs["compute_load_out"] = compute_load_demand
+        outputs["total_compute_load_produced"] = np.sum(compute_load_demand) * (self.dt / 3600)
+        outputs["capacity_factor"] = outputs["total_compute_load_produced"].sum() / max_production
+        outputs["annual_compute_load_produced"] = outputs["total_compute_load_produced"] * (
+            1 / self.fraction_of_year_simulated
+        )
+        outputs["rated_compute_load_production"] = inputs["system_capacity"]


These output names don't match the other model's; for interchangability, could you please use the same names for both models? Unless there's a meaningful difference between the outputs.

Simple model PUE/WUE model

compute_load_demand compute_it_workload

demand_profile compute_it_workload_profile

unmet_electricity_demand_out unmet_electricity_demand

none unmet_water_demand

Normalized!

johnjasa · 2026-10-15T19:24:30Z

+        if warnings:
+            self.logger.warning(
+                "Warnings encountered when selecting weather station for location (%s, %s): %s",
+                latitude,
+                longitude,
+                warnings,
+            )


There's no logger attr. If this is reached, it fails on an AttributeError. I'd suggest raising a UserWarning

Old habits die hard...

johnjasa · 2026-10-15T19:26:46Z

+        opex = fixed_om + variable_om
+
+        outputs["CapEx"] = capex
+        outputs["OpEx"] = opex


Could you please break out the variable OpEx into VarOpEx? The GenericConverterCost is a good example of this

johnjasa · 2026-10-15T19:28:25Z

+    climate_zone: str = field(default=None)
+    efficiency_level: str = field(default="efficient")
+    cooling_configuration: int = field(default=None)
+    optimize_for: str = field(default="pue")
+    size_sqft: float = field(default=None)


Could you please add a list of allowable inputs here? Each one of these could accept e.g. optimize_for: "babadook" and it'd pass initialization without a clear error

johnjasa · 2026-10-15T19:29:42Z

+        3600,
+        3600,
+    )  # (min, max) time step lengths (in seconds) compatible with this model
+


please add a _control_classifier for each performance model. I think "fixed" is the right option but it might be another one.

kbrunik

Thanks for adding this PR and all of the details on how to use it, this is going to be a great addition to H2I! The doc page looks great!

I have a bigger concern about making sure the model is correctly limited by the feedstocks available per timestep for both of the data center models that I think it would be good to take a closer look at.

To respond to your reviewer feedback questions:

I think the PUE/WUE file looks good where it's at, easy to see how it's related.
I think you should remove the rates from the cost model and allow the feedstocks to handle the costs. It will help add more clarity since in the current model they have to be lumped into a single VarOpex for the data center.

kbrunik · 2026-10-15T17:02:29Z

+\dot{V}_{\text{water}} = E_{\text{IT}} \times \text{WUE}
+```
+
+where PUE is the ratio of total facility power to IT equipment power, and WUE is the water


Can you add the acronym definitions for PUE and WUE?

kbrunik · 2026-10-15T17:06:07Z

+    """
+
+    system_capacity_mw: float = field(validator=validators.gt(0))
+    compute_electrical_efficiency: float = field(validator=validators.gt(0))


Suggested change

compute_electrical_efficiency: float = field(validator=validators.gt(0))

compute_electrical_efficiency: float = field(validator=(validators.gt(0), validators.le(1)))

Then you don't get folks trying to input percentages

kbrunik · 2026-10-15T17:09:52Z

+    compute_electrical_efficiency: float = field(validator=validators.gt(0))
+    cooling_load_ratio: float = field(validator=validators.ge(0))
+    water_use_gal_per_mwh: float = field(validator=validators.ge(0))
+    demand_profile: int | float | list = field()


I'd change this to demand_profile_mw since the units are preset in the initialize, or you could allow the user to specify the commodity_rate_units in the config and then leave it as demand_profile which is more common across the code base (see Demand Base Class).

And could you add this attribute to the docstring?

kbrunik · 2026-10-15T17:11:53Z

+            compute load (0 < efficiency <= 1).
+        cooling_load_ratio (float): Ratio of cooling load to compute load.
+        water_use_gal_per_mwh (float): Water usage per MWh of compute load.
+        electricity_in (float array): Electricity input profile in MW/h.


Suggested change

electricity_in (float array): Electricity input profile in MW/h.

electricity_in (float array): Electricity input profile in MW.

kbrunik · 2026-10-15T17:15:05Z

+            outputs: OpenMDAO outputs object for compute_load_out, water_consumed,
+                and unmet_electricity_demand.
+        """
+        # system_capacity = self.config.system_capacity_mw  # plant capacity in MW


delete commented out line

kbrunik · 2026-10-15T19:56:11Z

+        if not matching_rows:
+            raise ValueError(
+                f"No PUE/WUE data found for climate zone '{climate_zone}' and efficiency "
+                f"'{efficiency}'. Change the climate zone or efficiency level, or supply "
+                f"your PUE and WUE values directly."
+            )
+
+        if cooling_configuration is not None:
+            if cooling_configuration not in self.config.COOLING_CONFIGURATIONS:
+                raise ValueError(
+                    f"cooling_configuration must be an integer from 1 to 12, "
+                    f"got '{cooling_configuration}'."
+                )


You could remove these errors too if you use the validators above

kbrunik · 2026-10-15T19:57:08Z

+            valid_cases = self.config.SIZE_COOLING_CONFIGURATIONS[size_category]
+            matching_rows = [r for r in matching_rows if r[0] in valid_cases]
+            if not matching_rows:
+                raise ValueError(


This error could also be removed with use of the validators above

kbrunik · 2026-10-15T19:59:50Z

+            USD/(MW*year).
+    """
+
+    electricity_rate: float = field(validator=validators.ge(0))


I'd suggest using our feedstocks as is and remove the water and electricity rates from this model. I think you end up with a slightly obfuscated picture when you combine the water electricity into a joint varopex and you can see the impact of water/electricity rate changes more clearly if you use the feedstock models.

kbrunik · 2026-10-15T20:00:17Z

+            "system_capacity",
+            val=self.config.system_capacity_mw,
+            units="MW",
+            desc="IT equipment rated capacity in MW",


Suggested change

desc="IT equipment rated capacity in MW",

desc="IT equipment rated capacity",

kbrunik · 2026-10-15T20:01:06Z

+        self.add_input(
+            "capex_per_mw",
+            val=self.config.capex_per_mw,
+            units="USD/MW",
+            desc="Capital cost per MW of IT equipment capacity",
+        )


Suggested change

self.add_input(

"capex_per_mw",

val=self.config.capex_per_mw,

units="USD/MW",

desc="Capital cost per MW of IT equipment capacity",

)

self.add_input(

"capex",

val=self.config.capex_per_mw,

units="USD/MW",

desc="Capital cost of IT equipment capacity",

)

bayc added 5 commits October 7, 2026 08:04

Merge remote-tracking branch 'origin/develop' into feature/data_center

7004881

update class registration

1464fc0

fix isort issues

bea194e

update commodity rate units

f661e86

bayc self-assigned this Oct 13, 2026

bayc added the enhancement New feature or request label Oct 13, 2026

bayc added 4 commits October 13, 2026 09:32

update changelog with PR number

1e82cf3

updates to docstring

f87b31d

Merge remote-tracking branch 'origin/develop' into feature/data_center

43ad194

more updates to docstring

9ae8c27

bayc requested review from johnjasa and kbrunik October 13, 2026 19:26

Merge branch 'develop' into feature/data_center

675e702

johnjasa requested changes Oct 15, 2026

View reviewed changes

kbrunik requested changes Oct 15, 2026

View reviewed changes

updates based on PR feedback

d038508

		load profile. An example that runs both model pairs with an hourly AI training workload
		is provided in `examples/38_data_center/`.

Simple model	PUE/WUE model
`compute_load_demand`	`compute_it_workload`
`demand_profile`	`compute_it_workload_profile`
`unmet_electricity_demand_out`	`unmet_electricity_demand`
none	`unmet_water_demand`

	compute_electrical_efficiency: float = field(validator=validators.gt(0))
	compute_electrical_efficiency: float = field(validator=(validators.gt(0), validators.le(1)))

	electricity_in (float array): Electricity input profile in MW/h.
	electricity_in (float array): Electricity input profile in MW.

	desc="IT equipment rated capacity in MW",
	desc="IT equipment rated capacity",

Conversation

bayc commented Oct 13, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Add data center performance and cost models

Table 1: Cooling configurations available in data center model

References:

Section 1: Type of Contribution

Section 2: Draft PR Checklist

TODO:

Type of Reviewer Feedback Requested (on Draft PR)

Section 3: General PR Checklist

Section 4: Related Issues

Section 5: Impacted Areas of the Software

Section 5.1: New Files

Section 5.2: Modified Files

Section 6: Additional Supporting Information

Section 7: Test Results, if applicable

Section 8 (Optional): New Model Checklist

Uh oh!

johnjasa left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

kbrunik left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

This branch has not been deployed

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

bayc commented Oct 13, 2026 •

edited

Loading