PS8. Modeling and predicting grid impacts from large load ramping

Large data centers can pose a significant challenge to transmission system balancing when their load drops off or ramps up suddenly. These step changes in demand can cause operational stress and reliability risks for both local substations and the broader transmission system. As PG&E begins to explore ways to accelerate interconnection timelines, they are beginning to request load flexibility, such as the ability to partially curtail or ramp down data center loads on demand, as a condition for siting. However, there is currently no robust, research-based scenario modeling capability to predict when and where these step changes become problematic, how they interact with high penetrations of inverter-based resources (IBRs) like large-scale solar and batteries, and what mix of technologies or operational practices is most effective for mitigation. 

Why is this important?

As the number and size of hyperscale data centers grow, their instantaneous load behaviors can create shock events that impact grid stability and require fast-responding backup capacity. If these facilities could behave more like “soft start” motors, ramping on and off gradually, their integration would become less disruptive to both transmission and distribution systems. Understanding the conditions under which these ramps create system-wide problems, especially when co-located with dense (inverter-based resources) IBR clusters, is critical for resilient grid operations and cost-effective interconnections.

What is the current state and its primary limitations?

While many data centers can modulate internally, they still often appear to the grid as abrupt, large net load changes, coming online or offline in bulk with little external coordination. Grid operators typically have limited visibility or control over these transitions, making it difficult to anticipate or mitigate their impact on system stability. PG&E currently lacks data and models that can capture the detailed interactions between large flexible loads and local clusters of IBRs, including potential “wave” effects that can cascade through the system and trigger voltage and frequency issues.

Primary limitations include:

• Rapid changes in data center load can occur due to controlled ramping for normal processes such as AI training and inference
• Sudden load changes of this magnitude can lead to frequency deviations, voltage instability, or require costly fast-ramping generation to maintain system balance 
• There is no standardized methodology to run scenario analyses that test how different ramp magnitudes and speeds interact with local grid conditions,  
• Lack of integrated scenario models that combine large flexible load and ramp behavior with high IBR penetration and protection system dynamics; specifically, under which conditions and at what magnitudes this becomes an issue 

What are the desired outcomes from R&D?

PG&E seeks to identify and adapt research-grade scenario modeling tools and supporting frameworks that can be directly incorporated into its system planning and operations, with desired outcomes as follows:

• Methods to simulate and predict load transition impacts of data center loads on the transmission system, and recommend mitigation strategies
• Advanced modeling that simulates large load ramps under realistic grid conditions with high IBR penetration, capturing voltage, frequency, protection, and system stability impacts
• Research and/or modeling that identifies thresholds and predictive triggers that indicate when and where large load ramps become problematic based on magnitude, ramp rate, and local grid context
• Grid system coordination between the data center controller and utility grid signals to enable synchronized ramping
• Design and validate advanced control algorithms that actively dampen system oscillations when data centers ramp load up or down rapidly
• Ability to identify relevant mitigation options that connect modeling insights to the most effective technologies and operational practices, whether those are customer-side (e.g., soft start controls) or utility-side (e.g., grid control signals, fast reserves)
• Technology solutions that stabilize the system by suppressing oscillations resulting from sudden data center load variations
• Guidance or standards to inform customer design and utility interconnection operation that support controllable ramp-in/ramp-out and system balancing

PS6. Enhancing visibility into DER and EV load flexibility potential

Visibility into Distributed Energy Resources (DER) and Electric Vehicle (EV) load data is crucial for PG&E for several key reasons. First, the inability to predict the demand expected from DERs when relying primarily on aggregated meter data creates challenges in accurately forecasting energy needs and ensuring reliable grid operations. Second, without detailed information, PG&E cannot assess the flexibility potential of DERs, including EVs, which limits our ability to leverage these assets for grid stability and efficiency. Lastly, the lack of visibility into how much of this flexibility potential can be captured under the right incentive structures challenges PG&E when seeking to design and implement highly effective load management programs. At the CAISO level, the unpredictable nature of DER behavior is starting to destabilize both day-ahead and real -time forecasting models, driving a need for last-minute procurement of expensive resources, with costs ultimately passed to ratepayers. Addressing these gaps is essential for enhancing grid reliability, optimizing resource use, and supporting the sustainable and efficient integration of renewable energy. Additionally, real-world DER performance data will provide essential inputs into orchestration strategies, creating a feedback loop between visibility and dispatch optimization.

While most DERs are connected to the internet and store location and usage data, these data are not consistently available to PG&E. Beyond site-level meter (Advanced Metering Infrastructure) data at the hourly or 15-minute interval level, PG&E lacks a comprehensive system to collect essential information about non-market participating resources, which influences accurate forecasting. While the existing AMI data provides site-level consumption data, it does not differentiate between behind-the-meter device-level performance (e.g., individual DERs or EV chargers). As a result, PG&E needs solutions that can both leverage existing AMI data and go beyond to identify DERs we don’t know about and monitor their real-time operational status, unlocking more precise forecasting and flexible load orchestration. Questions we must be able to answer with better data and better forecasts include: what programs are DERs participating in, who is operating them, what are the performance duration and fatigue limitations, and what forms of automation are they capable of using?

Primary limitations include:

• A lack of visibility into the location and asset type for both existing and new DERs connected to the system
• A lack of a standardized method to securely capture electricity usage data at the device level
• A lack of access to aggregated historical DER performance (global or national) data beyond nameplate values, limiting the ability to forecast with statistical confidence
• A lack of a feedback loop to update forecasts based on observed DER performance during dispatch events

Novel technologies, including AI/ML, to:

• Capture real-time DER-specific location and usage data comprehensively across all DERs in the network and securely store this information in a centralized repository
• Enable grid-edge computing and analytics to locally process DER and EV data for faster, more reliable visibility and decision-making
• Develop forecasting systems that can update dynamically based on actual DER performance outcomes, improving future dispatch accuracy and strategy.
• Forecast the ability of DERs to respond to direct scheduling, automation, price, or other firm and non-firm control signals at times when the grid needs it most – either to peak shave or absorb excess supply, both at the system level and the distribution level
• Evaluate data to maximize predictability with statistical confidence of various types or combinations of DER technologies and load management responses
• Support data-sharing processes with CAISO that enhance regional forecasting and operational planning while preserving customer privacy and regulatory compliance
• Create the foundational data infrastructure needed to support intelligent orchestration efforts