Large Load Interconnection and Power System Modeling for Data Centers, AI Facilities and Electrolyzers

Large Loads, New Rules: Engineering Grid Reliability in the Age of Data Centers and AI

Introduction: A Fundamental Shift in Load Characteristics

The electricity grid has managed industrial loads for over a century. Steel mills, chemical plants, mines, and smelters have always demanded large blocks of power from transmission systems. But the large loads now connecting to the grid — AI training facilities, hyperscale data centers, cryptocurrency mining operations, advanced semiconductor fabrication plants, and green hydrogen electrolyzers — are categorically different from anything the industry has dealt with before.

The differences are not just in scale, although scale matters enormously. They lie in concentration, behavior, power electronics interfaces, load variability profiles, and the speed at which these facilities are being deployed. Existing interconnection frameworks, modeling tools, performance standards, and operational protocols were not designed for loads of this character. The result is a widening gap between what the grid needs to operate reliably and what current practices provide.

At Keentel Engineering, we track these developments closely. This brief draws on the most current technical research and industry practice to provide a comprehensive engineering perspective on what integrating large power-electronic loads actually requires — in terms of modeling, performance, and process.

What Makes Modern Large Loads Different

Grid engineers and planners are sometimes asked: we have had large industrial loads for decades — what is really new here? The answer is that several characteristics have converged simultaneously in a way that creates genuinely novel reliability challenges.

Size and Geographic Concentration

A major steel mill or aluminum smelter might draw two or three hundred megawatts. A modern hyperscale data center campus can exceed a gigawatt. More importantly, multiple such facilities are being sited in close proximity — sometimes in the same substation area or industrial park — creating load concentrations equivalent to entire cities in footprints measured in acres rather than square miles. The density of power demand in these areas has no historical precedent in transmission planning experience.

Highly Dynamic Load Profiles

Traditional large industrial loads — motors, furnaces, electrolytic processes — tend to draw relatively stable power with gradual ramp rates. Modern AI training facilities are fundamentally different. GPU compute clusters ramp up and down during training cycles, creating rapid swings in active power consumption that can occur on timescales of seconds. Cryptocurrency mining loads adjust aggressively based on profitability signals. These profiles introduce high-frequency variability that has the potential to excite oscillatory modes in power systems in ways that aggregated conventional load never did.

Engineering Insight Forced Oscillation Risk

A well-documented event involved a small generator oscillating at approximately four-second intervals. Despite modest individual magnitude, the oscillation excited a natural inter-area mode and produced hundreds of megawatts of power swings that propagated across the entire eastern interconnection from the southeastern United States to northern Manitoba. A large AI training facility exhibiting cyclic active power consumption at a similar frequency — but with far greater magnitude — represents a materially higher risk of the same phenomenon. Load-induced forced oscillation is an underappreciated reliability risk that must be addressed explicitly in interconnection studies.

Power Electronic Interfaces

Inverter-based generation has been the central modeling and stability challenge for the past decade. Large loads connected through power electronic interfaces — variable-frequency drives, rectifiers, switch-mode power supplies, uninterruptible power systems — present similar challenges from the load side. Protection systems designed to prevent damage to sensitive computing equipment can cause abrupt, coordinated disconnections of hundreds of megawatts. Phase-locked loops that lose synchronism can alter active power consumption in unpredictable ways. The control dynamics of these loads interact with the grid in ways that existing load models — developed for motors and heating elements — cannot represent.

Uncertainty in Load Growth Forecasts

Load growth across the grid had been essentially flat for two decades. Planners and economists had built careers around managing a largely static demand picture. The emergence of large power-electronic loads has broken this pattern abruptly. Forecasts for large load interconnection queues can vary by enormous margins — not due to poor forecasting methodology, but because the underlying development activity itself is rapid, uncertain, and difficult to observe from outside. Projects are announced, modified, relocated, or cancelled on short timescales. Planners are building transmission infrastructure against forecasts that carry unprecedented uncertainty.

The Modeling Challenge: Getting the Physics Right

Accurate power system modeling is the foundation of reliable interconnection. Before a large load facility connects to the transmission system, planners need models that accurately represent how that load will behave during normal operations and, critically, during disturbances. For modern large power-electronic loads, the modeling challenge is significant.

Why Legacy Load Models Fall Short

The dominant load modeling framework used across North American power systems is the composite load model — a sophisticated representation that captures motor loads, power electronics, discharge lighting, and other components typical of residential, commercial, and mixed industrial areas. The composite load model has been a major advance over earlier simplified approaches. But it was not designed to represent the specific characteristics of AI data centers, GPU clusters, or electrolyzers.

A composite load model cannot capture the cyclic active power variability of an AI training facility. It cannot represent the ride-through logic of a facility designed around protecting sensitive computing equipment rather than staying connected to the grid. It does not model the harmonic injection characteristics of large rectifier banks. It cannot represent the phase-jump tolerance thresholds of power-electronic front ends. New modeling frameworks are needed.

Positive Sequence vs. Electromagnetic Transient Modeling

Two fundamentally different simulation frameworks are used in power system studies, and understanding when each is appropriate is essential for correctly assessing large load impacts.

Positive sequence phasor-domain tools — the workhorses of transmission planning — operate in the sub-10 Hz frequency range and model the bulk power system at the level of buses, lines, and aggregated components. These tools are computationally efficient enough to model entire interconnections and are well suited for studying inter-area oscillations, transient stability, and voltage stability issues that affect large portions of the grid. When assessing the bulk system impact of a large load — its effect on voltage profiles across a region, its participation in frequency events — positive sequence modeling is the appropriate platform.

Electromagnetic transient tools model the system in three-phase detail, capturing fast dynamics associated with power electronic switching, phase-locked loop behavior, subsynchronous resonance, and converter-driven instabilities. EMT modeling is computationally intensive and is typically applied to a local portion of the system with an equivalenced representation of the broader network. For large loads with significant power electronics, EMT studies are often needed to identify local instabilities, harmonic interactions, and subsynchronous oscillation risks that positive sequence tools cannot capture.

Key Decision Principle Choosing the Right Tool

The choice between positive sequence and EMT modeling should be driven by the phenomena being studied, not by familiarity or computational preference. Positive sequence tools systematically miss fast-dynamics phenomena that can cause real stability problems near large power-electronic loads. The appropriate approach is to start with positive sequence analysis for bulk system impact assessment and to trigger EMT studies when screening criteria indicate elevated risk — based on facility size, system strength at the point of interconnection, proximity to other power-electronic equipment, and load profile characteristics.

Generic vs. OEM-Specific Models

For both positive sequence and EMT platforms, a distinction exists between generic library models and original equipment manufacturer-specific models. Generic models are white-box representations designed to capture the general behavior of a class of equipment — useful for planning studies where actual equipment vendors are not yet selected, and for training and exploratory analysis. OEM-specific models are black-box representations developed by equipment manufacturers to replicate the behavior of their specific products. They are generally more accurate for site-specific studies but are difficult to inspect, may have undisclosed limitations, and require coordination with the manufacturer for tuning and validation.

A critical engineering principle applies to both: more detailed modeling does not automatically mean better modeling. A complex EMT model with incorrect parameters or coding errors can produce misleading results that are worse than a well-calibrated positive sequence model. Model accuracy depends on the quality of development, parameterization, and validation — not on the sophistication of the platform.

Essential Model Characteristics for Large Load Facilities

Any dynamic model submitted for interconnection studies of a large power-electronic load facility should, at minimum, be capable of representing:

• Active power and frequency response — including any frequency-sensitive load reduction logic
• Reactive power capability and voltage control — including any on-site voltage regulation equipment
• Load profile variability — including cyclic ramp behavior during AI training or computing cycles
• Voltage ride-through logic — the thresholds and timing of load reduction or disconnection during voltage disturbances
• Ramp rate limitations — both normal operational ramps and post-disturbance recovery ramps
• Ride-through counter logic — including any multi-strike disconnection thresholds
• On-site generation behavior — if backup or co-located generation exists
• Mechanical load characteristics — for motor-driven processes within the facility
• Unbalanced voltage operation — for EMT models operating on per-phase voltages
• Pre-charge and startup behavior — if the model will be used for black-start or restoration studies

Model Validation: Closing the Loop Between Paper and Reality

A model that has not been validated against actual facility behavior is an educated guess. For large power-electronic loads — where behavior can be complex, non-linear, and sensitive to control settings unvalidated models represent a significant planning risk. Model validation requires:

• Parameter verification: Confirming that all model parameters match actual facility settings and equipment ratings, using OEM data sheets and commissioning records
• Dynamic event validation: Comparing model responses against recorded fault or disturbance data from digital fault recorders at the facility
• Continuous operating characteristic validation: Using phasor measurement unit data to compare steady-state and slowly-varying model behavior against observed performance

A structured model quality testing protocol should be applied before any model is accepted for use in planning or interconnection studies. Quality tests include: platform stability tests confirming the model initializes and runs without numerical artifacts; voltage and frequency ride-through tests confirming the model responds appropriately to disturbances; phase angle jump tests; and controlled load change tests confirming the model follows set-point changes without oscillatory or numerically unstable behavior.

Performance Requirements: What the Grid Needs from Large Loads

Model accuracy is necessary but not sufficient. Large load facilities must also be required to perform in specific ways during grid disturbances. Without clear performance requirements written into interconnection agreements, a facility may behave in ways that are rational from its own operational perspective but disruptive or dangerous from the grid's perspective.

Voltage Ride-Through: Two Perspectives in Conflict

A fundamental tension exists in how 'ride-through' is understood by grid operators versus by large load facility operators. From the grid perspective, voltage ride-through means that a facility remains electrically connected to the system during and after a disturbance — maintaining the post-contingency network configuration that planners designed for. From the facility perspective, ride-through means that internal processes continue without interruption regardless of grid conditions. These two definitions are not the same. An AI training facility can maintain internal computational continuity using on-site uninterruptible power systems while simultaneously disconnecting from the grid — providing 'ride-through' by its own definition but causing a large loss of load from the grid's perspective.

Grid performance requirements must be explicit that ride-through means remaining connected to the transmission system and restoring load according to specified timelines after disturbances clear — not merely maintaining internal operational continuity.

Reliability Alert: The bulk of computing equipment in modern data centers follows the ITIC (Information Technology Industry Council) curve for voltage tolerance — a standard developed to protect equipment from damage, not to maintain grid connectivity. ITIC does not constitute adequate grid ride-through requirements. Interconnection agreements must specify separate, grid-oriented ride-through curves.

Developing Site-Specific Ride-Through Curves

A single universal ride-through curve is not appropriate for all locations and system topologies. The appropriate curves for low-voltage and high-voltage ride-through at any given point of interconnection should be developed based on:

• Local fault clearing times — both normal clearing (protection operating correctly) and delayed clearing (backup protection scenarios)
• The most severe credible contingency events at or near the point of interconnection
• Normal voltage operating ranges at the facility's point of interconnection
• Whether automated post-contingency actions are used to restore voltages within acceptable ranges

A review of international practice reveals significant variability in ride-through requirements across jurisdictions — reflecting the different fault clearing times, protection philosophies, and system topologies in each system. This variability reinforces that jurisdiction-specific and site-specific curve development is the appropriate approach, informed by — but not copied from — requirements developed for other systems.

Multi-Disturbance Ride-Through: The Three-Strikes Problem

Several large load facilities have implemented 'three-strikes' or similar counter-based disconnection logic: after a specified number of voltage disturbances within a defined time window, the facility disconnects. This logic makes sense from an equipment protection perspective — repeated voltage disturbances can indicate an unstable grid condition that poses risk to sensitive equipment. But from a reliability perspective, the consequences can be severe.

A double-circuit transmission contingency — a tower failure taking two parallel lines simultaneously — followed by unsuccessful automatic reclosure attempts on both lines is a routine, designed-for event in transmission planning. Under many three-strikes implementations, this single normal clearing contingency would trigger facility disconnection. If multiple large load facilities sharing a transmission corridor all implement similar logic, a single design contingency could remove gigawatts of load simultaneously, triggering frequency excursions, voltage instability, and cascading outcomes that planners did not design for.

The engineering recommendation is clear: where technically feasible, counter-based disconnection logic should be disabled. Where it cannot be disabled, the parameters must be explicitly coordinated with the system operator's normal and emergency operating procedures, ensuring that designed-for contingency sequences — including unsuccessful reclosures — do not trigger disconnection.

Active Power Recovery After Faults

Ride-through requirements must address not only whether a facility remains connected during a disturbance but how it recovers load afterward. If a large load reduces its consumption during a fault to facilitate ride-through — a legitimate and often necessary approach — it must restore that load in a controlled, coordinated manner to prevent post-fault stability problems.

The appropriate active power recovery rate depends on the strength of the system at the point of interconnection. In electrically strong locations, restoration of 90% of pre-fault load within one second of voltage recovery to 90% of nominal is achievable without system instability. In weaker systems, more gradual recovery over several seconds may be necessary to avoid voltage collapse during the post-fault restoration period. Performance requirements should specify recovery rates appropriate to system strength, with interconnection studies confirming that the specified rates do not introduce secondary instabilities.

Ramp Rate Control and Area Control Error

The active power variability of large power-electronic loads directly stresses the area control error (ACE) management of balancing authorities. Automatic generation control was designed to manage the gradual, predictable ramp rates of conventional loads and the scheduled dispatch of generation. Second-to-second or minute-to-minute swings of tens or hundreds of megawatts from large loads operating outside coordinated frameworks can exhaust regulation capacity rapidly.

Ramp rate requirements for large loads should be established through analysis of the balancing authority's existing regulation capacity and the cumulative impact of planned large load additions. Where individual facility ramp rates cannot be technically limited without disrupting operations, additional regulation procurement may be necessary. The key principle is that the cumulative variability of large loads must be within the system's demonstrated regulation capability, or that capability must be expanded to accommodate the new loads.

High-Frequency Cycling and Torsional Interaction

The high-frequency component of large load active power variability — cycling at frequencies of one to tens of hertz rather than the one-to-several-second low-frequency cycles associated with training workloads — poses a specific risk to conventional generating units located electrically close to the load facility. Thermal generating units have shaft systems with natural torsional modes in the one to fifty hertz range. If load-driven power oscillations excite these modes, the resulting cyclic shaft torques can cause fatigue damage.

Generator manufacturers specify shaft damage criteria in terms of shaft torque magnitude versus cycle count. Operating in the 'continuous' region — where oscillations are small enough to accumulate indefinitely without damage — is acceptable. Operation at higher torque levels can cause measurable damage within thousands of cycles — which at frequencies of several hertz can occur within minutes. This risk must be evaluated explicitly when citing large power-electronic load facilities near generating stations, with site-specific torsional interaction studies conducted where risk screening indicates potential concern.

Monitoring Requirements



Performance requirements are only enforceable if facility behavior can be observed and measured at sufficient resolution. Monitoring equipment installed at large load facilities should be capable of both streaming real-time data to the system operator and recording high-fidelity event data for post-disturbance analysis. A minimum sampling rate of 100 hertz is recommended to capture the phenomena relevant to large load interconnection performance — including high-frequency active power cycling and fast voltage transients. Recordings should be maintained for a minimum of 20 days to ensure that data from any significant system event is available for analysis.

Interconnection: A Critical but Double-Edged Asset

Even the best technical requirements are ineffective without a well-designed interconnection process to implement them. The generation interconnection process in North America has been through a multi-decade evolution — from ad hoc utility-by-utility procedures to standardized FERC-regulated frameworks — driven by the reliability problems that emerged when that process could not keep pace with the scale and speed of new generation development. Large load interconnection is at an earlier stage of the same evolution.

What a Baseline Large Load Interconnection Process Should Include

A rigorous large load interconnection process should encompass the following phases and milestones:

• Pre-application readiness review: Verification of site control, financial commitment, and completeness of technical data packages before formal application is accepted. This is the viability filter that prevents speculative applications from consuming limited study resources.
• Application package with modeling and performance data: Submission of dynamic models (positive sequence and EMT as appropriate), facility specifications, load profile data, and preliminary performance capability documentation at the time of application.
• Tiered study process: Tier 1 RMS-based power flow and dynamic stability studies for all applicants; screening analysis to identify facilities requiring detailed EMT studies based on size, system strength, load profile characteristics, and proximity to sensitive equipment.
• EMT study execution where triggered: Subsynchronous oscillation studies, control stability analysis, harmonic and power quality assessment, and torsional interaction screening for facilities that meet EMT study trigger criteria.
• Interconnection agreement with enforceable performance requirements: Ride-through curves, ramp rate limits, monitoring requirements, and compliance testing obligations incorporated as binding interconnection agreement terms.
• Construction and commissioning oversight: Verification that as-built facilities match studied configurations, with commissioning testing to confirm performance against requirements.
• Post-interconnection monitoring and compliance: Ongoing monitoring against requirements, with mechanisms for review and remediation if behavior does not conform to studied assumptions.

Coordination Between Transmission Entities and ISO/RTOs

A specific complexity of modern large load interconnection is that large facilities — particularly those co-located with or supported by generation resources — simultaneously engage multiple regulatory entities. The transmission utility is responsible for local interconnection studies and facilities. The ISO or RTO is responsible for generation interconnection (if co-located generation is involved), transmission planning, and market participation. When a facility involves both load interconnection and co-located or nearby generation, these two processes must be coordinated to ensure that studies are consistent, requirements are compatible, and neither process creates gaps in reliability assessment.

The emerging policy landscape is creating new frameworks to manage this complexity — including new service categories for co-located generation and load, dedicated large load interconnection processes within ISO/RTO tariffs, and high-impact large load assessment procedures. These frameworks are still developing rapidly. The engineering principle that must be maintained throughout this evolution is that baseline technical requirements — modeling, performance, and monitoring — must be consistently applied regardless of which regulatory entity has primary jurisdiction over a given interconnection.

Lessons from Generation Interconnection Reform

The history of generation interconnection reform provides a useful template. Over roughly two decades, the North American generation interconnection process evolved from fragmented, utility-specific procedures to standardized frameworks with transparent queue management, defined study milestones, commercial readiness requirements, and enforceable interconnection agreements. This evolution was driven by the reliability problems and market inefficiencies that emerged when the process could not keep pace with the scale of renewable energy development. Large load interconnection is now in an analogous early period — and the engineering community has the benefit of knowing where the generation interconnection process eventually needed to go.

Key lessons applicable to large load interconnection include: the importance of commercial readiness filters at the application stage to weed out speculative projects; the value of standardized study processes that enable consistent treatment of similar applicants; the necessity of enforceable performance requirements incorporated into interconnection agreements rather than left as voluntary guidelines; and the need for ongoing monitoring and compliance infrastructure to ensure that as-built and operating facilities continue to meet the requirements they were studied to.

Demand-Side Flexibility: A Depth Reducer, Not a Duration Reducer

A specific sensitivity test fixed demand-side flexibility (DSF) capacity in one portfolio at 2030 levels and held it constant through 2040, rather than allowing it to grow. The results were nuanced:

• Overall LOLE change was modest — DSF growth did not dramatically change the number of hours of loss of load
• However, within stress events, expected unserved energy increased significantly when DSF was constrained

The practical interpretation: demand-side flexibility is most valuable not for preventing outages entirely, but for limiting the severity of outages that do occur. It reduces the depth of shortfall events — the total amount of energy not served during a crisis — rather than eliminating the events themselves.



For engineering teams designing demand response programs, this finding suggests that valuation frameworks focused purely on peak demand reduction may underestimate DSF's contribution to system resilience. Metrics that capture unserved energy depth, not just hours of outage, are needed.

Conclusions: Engineering the Grid for the Large Load Era

The integration of large power-electronic loads at the scale now being observed represents one of the most significant reliability engineering challenges the grid has faced. It combines the modeling complexity of inverter-based resources with the scale and speed of development that overwhelmed the generation interconnection process a decade ago — and it is occurring simultaneously with the integration of large amounts of inverter-based generation, making the overall system stability picture more complex than either challenge alone.

Keentel Engineering's perspective is that the tools, techniques, and frameworks needed to manage this challenge exist or are being developed. The gaps are primarily in implementation — in applying rigorous modeling and performance requirements consistently, in building interconnection processes capable of handling the volume and complexity of applications, and in investing in the monitoring infrastructure needed to close the feedback loop between studied and actual behavior.

The large load era is not a reliability crisis waiting to happen. It is an engineering challenge that can be met with the systematic application of good engineering practice: accurate models, clear requirements, structured processes, and continuous learning from operating experience.

CASE STUDIES — KEENTEL ENGINEERING

Large Load Integration in Practice: Three Keentel Engineering Engagements

The following case studies describe how Keentel Engineering has applied the technical principles discussed in this brief to real-world engineering challenges across modeling, performance assessment, and interconnection process design.

Dynamic Model Development and Validation for a 600 MW AI Data Center Campus

Creating and validating positive sequence and EMT models for a hyperscale AI training facility in a high-penetration IBR region

Background

A hyperscale technology company engaged Keentel Engineering to develop and validate dynamic power system models for a 600 MW AI training campus interconnecting at 345 kV. The facility comprised three phases of GPU compute clusters, each with a target installed GPU density of approximately 200 MW, with all three phases ultimately interconnected at a single high-voltage substation via dedicated step-up transformers.

The interconnection study region was characterized by high penetration of inverter-based wind generation and existing concerns about subsynchronous oscillation risk. The transmission utility required both positive sequence and EMT models as conditions of the interconnection application. The technology company had not previously developed grid-connected power system models and had limited experience engaging with transmission interconnection technical processes.

Technical Challenge

The facility's GPU compute clusters were procured from multiple OEMs, using different rectifier topologies and power electronics platforms. Each OEM provided black-box dynamic models, but the models were developed for different simulation platforms, had inconsistent documentation, and had not previously been validated against field data at any operating facility. Three specific technical challenges required resolution:

• Platform incompatibility: Two of the three OEM models were available only in platforms not used by the transmission utility for interconnection studies. Conversion or re-development was required.
• Active power cycling characteristics: Preliminary load data from a similar operating facility showed active power cycling with a dominant frequency of approximately 0.3 Hz during training cycles — directly coinciding with a known inter-area oscillatory mode in the regional interconnection.
• Multi-strike logic: The facility's internal protection systems included a five-strike counter that would disconnect the facility from the grid after five voltage dips within 30 seconds. Analysis showed this could be triggered by a normal delayed-clearing fault scenario in the interconnection area.

Keentel Engineering Approach

Phase 1: Model Development

Keentel Engineering worked with each OEM to obtain functional block diagrams and parameter definitions for their respective models. Where OEM models were available only in incompatible platforms, Keentel Engineering developed equivalent positive sequence models using the functional block diagrams as specifications, implementing them in the transmission utility's preferred simulation platform. EMT models were developed for the two GPU cluster designs with the most significant power electronic interaction risk, using a combination of OEM-provided switching-level models equivalenced to average models for the interconnection study application.

Phase 2: Model Quality Testing

All developed models underwent the full model quality test suite before submission for interconnection study use. Two models initially failed the ride-through response test — their simulated response to voltage dips did not replicate the protective relay settings documented in the facility's protection coordination study. After parameter correction in coordination with the OEMs, all models passed all quality tests.

Phase 3: Forced Oscillation Assessment

Keentel Engineering conducted a detailed forced oscillation assessment using the validated positive sequence model. The analysis confirmed that the facility's 0.3 Hz active power cycling would inject energy into the regional inter-area mode. Simulations showed oscillation amplitudes building to approximately 180 MW at the regional interchange points over a 60-second period — below thresholds for immediate system instability but representing a material and growing reliability concern.

Keentel Engineering worked with the technology company's operations team to evaluate four mitigation approaches: job scheduling smoothing, GPU power limits, rack-level battery buffering, and facility-level BESS  A combination of rack-level battery buffering and job scheduling smoothing was selected as the primary mitigation, achieving approximately 75% reduction in cycling amplitude at the dominant frequency in bench testing.

Phase 4: Multi-Strike Logic Coordination

The five-strike counter was reconfigured in coordination with the facility protection engineer and the transmission utility to require 12 voltage dips within 60 seconds before disconnecting — a threshold that could not be reached by any designed-for contingency sequence in the interconnection area, while still providing equipment protection against sustained multi-fault scenarios.

Outcomes

Post-energization monitoring data collected in the first six months of operation showed close agreement between modeled and observed facility behavior during two minor voltage disturbances in the interconnection area, providing validation of the model quality across two of the three OEM equipment types. The third OEM equipment type has not yet experienced a disturbance of sufficient magnitude for validation and remains on a schedule for parameter verification based on PMU continuous data.

Large Load Interconnection Process Design for a Regional Transmission Utility

Developing a standardized, technically rigorous interconnection framework for an emerging large load queue of over 8 GW

Background

A regional transmission-owning utility retained Keentel Engineering to design a comprehensive large load interconnection process framework. The utility had experienced a rapid influx of large load interconnection applications — primarily data centers and AI computing facilities — that had overwhelmed its existing interconnection processes. The existing process had been designed around conventional industrial load interconnections of tens of megawatts and did not include requirements for dynamic models, performance specifications, or EMT studies

Within 18 months, the utility had accumulated over 8 GW of large load interconnection applications, with individual project sizes ranging from 80 MW to 1.1 GW. Several early applications had been approved and were under construction using conventional process steps. Concerns were emerging about whether the reliability impacts of these facilities had been adequately studied.

Technical Challenge

The utility faced three distinct challenges simultaneously. First, retrospective assessment of approved projects: How to evaluate whether already-approved facilities posed reliability risks that had not been captured in their original interconnection studies. Second, prospective process design: How to design a new interconnection process framework that addressed the full range of technical risks posed by modern large power-electronic loads. Third, queue management: How to restructure its interconnection queue to prioritize projects with demonstrated commercial readiness while ensuring that speculative applications did not consume limited study resources.

Keentel Engineering Approach

Component 1: Retrospective Assessment Framework

Keentel Engineering developed a tiered retrospective assessment protocol for the 14 already-approved projects in construction or early operation. Projects were screened against risk criteria including: facility capacity above 200 MW; location in areas with existing subsynchronous oscillation concerns; load profiles with documented high-frequency cycling; and interconnection at 138 kV or below. Six projects were classified as requiring detailed dynamic assessment. Keentel Engineering conducted positive sequence dynamic studies for all six and EMT studies for three that met EMT trigger criteria. Two projects required mitigation measures — one required revised ride-through settings confirmed through model simulation; the other required installation of harmonic filters not included in the original interconnection design.

Component 2: New Process Framework Design

The new interconnection process framework established the following key elements:

• Commercial readiness filter: All new applications must demonstrate site control, letter of credit, and preliminary equipment procurement documentation before entering the study queue. Applications without commercial readiness evidence are placed in a pre-queue status.
• Technical data package requirements: Dynamic model submission (positive sequence and EMT where applicable), load profile documentation, protection settings documentation, and preliminary performance capability assessment required at application.
• Tiered study process: Tier 1 RMS studies for all applicants; EMT study trigger criteria applied based on six screening factors; Tier 2 detailed EMT studies for triggered projects.
• Standardized performance requirements: Ride-through curves developed for four system zones reflecting different fault clearing times and system strengths; ramp rate requirements established based on balancing authority regulation capacity analysis; monitoring requirements standardized across all new large load interconnections.
• Model quality testing: Mandatory quality test results required before models are accepted for Tier 1 or Tier 2 studies; OEM models must pass quality tests in the utility's simulation platform.

Component 3: Queue Restructuring

The existing interconnection queue was restructured using a cluster study approach, grouping applications by substation area and applying shared network upgrade cost allocation across clusters. Commercial readiness milestones were introduced at 90-day intervals; applications missing consecutive milestones were withdrawn from the queue, creating capacity for new applications with stronger development progress.

Outcomes

The retrospective assessment identified reliability risks in two projects that had been approved without dynamic modeling — risks that would have been identified and mitigated earlier under the new process framework. Both projects have since completed required mitigation measures and are operating within their interconnection performance requirements. The new framework has been shared with several neighboring transmission utilities as a reference model for their own large load process development.

Active Power Recovery and Ride-Through Compliance Assessment for a Green Hydrogen Electrolyzer Facility

Designing and verifying ride-through performance requirements for a 400 MW electrolysis facility in a weak grid interconnection area

Background

A green hydrogen developer engaged Keentel Engineering to perform a comprehensive ride-through performance assessment for a 400 MW electrolyzer facility interconnecting at 230 kV in a system area characterized by relatively low short-circuit ratios and significant penetration of inverter-based renewable generation. The project was the first large electrolyzer facility to seek interconnection in the utility's service territory, and no established interconnection requirements existed for this technology type.

The electrolyzer technology used proton exchange membrane (PEM) electrolysis — a power-electronic-interfaced process with operating characteristics significantly different from either conventional industrial loads or the alkaline electrolysis technology that had been studied in prior literature. The utility required Keentel Engineering to characterize the facility's behavior, develop appropriate performance requirements, verify that the proposed facility design could meet those requirements, and identify any mitigation measures needed.

Technical Challenge

The electrolyzer facility presented several specific technical challenges that were not addressed by existing interconnection standards or prior literature:

• Weak grid interconnection: Short-circuit ratio at the proposed point of interconnection was 2.8 — below the threshold at which many conventional ride-through requirements are designed to be achievable, and in a range where PLL stability under voltage disturbances is a documented concern for power-electronic devices.
• No OEM EMT model: The PEM electrolyzer OEM had not previously developed an EMT model for their equipment. A positive sequence model existed but had not been validated against field measurements from any operating facility.
• Hydrogen process continuity constraints: The developer required that the electrolyzer ride-through approach minimize disruption to the hydrogen production process. Abrupt full disconnection and reconnection cycles cause thermal and membrane stress that affects electrolyzer life. The ride-through approach needed to balance grid requirements with equipment protection considerations.
• Active power recovery in a weak system: The weak grid interconnection meant that rapid restoration of full electrolyzer load after a fault could itself trigger voltage instability — the load restoration acting as a second disturbance on an already recovering system.

Keentel Engineering Approach

Phase 1: Characterization Studies

Keentel Engineering conducted a detailed characterization of the PEM electrolyzer's electrical behavior through a combination of OEM technical documentation review, laboratory test data analysis, and engineering analysis of the power electronic front end. Key characteristics determined included: minimum voltage below which partial rather than full load reduction occurs; maximum PLL phase jump tolerance under laboratory conditions; natural ramp-down and ramp-up rates driven by the electrolysis process thermal dynamics; and harmonic current injection characteristics at the fundamental and key harmonic frequencies.

Phase 2: Site-Specific Ride-Through Curve Development

Keentel Engineering developed site-specific ride-through curves for the 230 kV interconnection point based on fault simulation studies for the interconnection area. Key parameters determined through simulation included: voltage at the facility terminals during a three-phase fault with normal clearing (shortest expected duration); voltage during a single-line-to-ground fault with delayed clearing (worst-case sustained voltage depression); and post-fault voltage recovery profile for the credible contingency cases. The resulting ride-through curves were specific to the system characteristics at this interconnection point, with a low-voltage ride-through requirement that was less stringent than standard IBR requirements in recognition of the weak grid context, combined with a requirement for partial load reduction during the fault to support voltage recovery.

Phase 3: Active Power Recovery Protocol

The active power recovery protocol was developed to address both grid stability and electrolyzer process constraints. The protocol specifies a four-stage recovery sequence following fault clearing: Stage 1 (0–2 seconds post-clearing): Maintain load at the partial level held during the fault; do not begin recovery until terminal voltage reaches 90% of nominal. Stage 2 (2–10 seconds): Ramp load recovery at a rate consistent with the system strength at the point of interconnection — slower than IBR standards recommend for strong system locations. Stage 3 (10–60 seconds): Continue controlled ramp toward pre-fault load level, subject to real-time voltage monitoring that can pause recovery if terminal voltage falls below a secondary threshold. Stage 4: Full load restoration confirmed when terminal voltage has been maintained above 95% of nominal for 30 consecutive seconds.

The recovery protocol was tested in simulation using the positive sequence model across all credible contingency scenarios for the interconnection area. In no scenario did the controlled recovery cause secondary voltage instability. The facility's modeled voltage profile remained within the normal operating range throughout all recovery simulations.

Phase 4: EMT Model Development

Working with the OEM, Keentel Engineering developed an EMT model of the PEM electrolyzer facility from functional specifications and laboratory data. The model was parameterized using the characterization data from Phase 1 and validated against available laboratory test data. Simulation studies using the EMT model confirmed acceptable PLL stability for the modeled fault scenarios at the facility's short-circuit ratio, with a specific finding that PLL performance degraded materially for faults causing terminal voltage below 0.4 per unit — informing a recommendation for enhanced PLL tuning as a condition of interconnection commissioning.

Outcomes

The engagement established a precedent for PEM electrolyzer interconnection performance requirements in the utility's service territory and has informed subsequent electrolyzer interconnection applications by other developers. The staged active power recovery protocol has been recognized by the utility's planning team as a model for active power recovery requirement design in weak grid interconnection areas — applicable to both large load and inverter-based generator interconnections in similar system strength contexts.

TECHNICAL FAQ — 20 QUESTIONS ANSWERED BY KEENTEL ENGINEERING

Integrating Large Loads: Technical FAQ for Grid Engineers and Planners

Keentel Engineering's transmission planning, interconnection, and power systems engineering teams address the twenty most frequently asked technical questions on large load integration.

About Keentel Engineering

Keentel Engineering provides transmission planning, interconnection engineering, power system modeling, and grid reliability services to utilities, developers, ISOs, and regulators navigating the challenges of large load integration, inverter-based resource growth, and grid decarbonization. Our engineers combine deep technical expertise with hands-on engagement in industry standards development and regulatory proceedings.