Source-linked AI summary
Long-Term Operational Planning Using Scenario-Based System Load Forecasting
Akash Debnath, Shiuli Subhra Ghosh, Jaime De La Ree, Alex Freeman, Kavin D Jones
TL;DR
Monthly peak-based outage studies may miss daily and weekly load variability, potentially restricting feasible short-duration outages. This paper compares coherent daily, weekly, and monthly load forecasts in long-term outage assessments. Granular forecasts improve outage accommodation, particularly for short-duration requests, while retaining existing reliability criteria.
Problem
Peak-based outage studies may overlook daily and weekly load variability, potentially restricting feasible outages, especially short-duration requests.
Source: §1.1 · p. 2Method
The study compares daily, weekly, and monthly load forecasts in long-term contingency-based outage assessments, using model selection and temporal reconciliation.
Source: §1.1 · p. 2, §5.1.3 · p. 9Results
Granular forecasts reduce unnecessary conservatism and improve outage-day approval and full-duration accommodation, particularly for short-duration outages, with daily loading generally providing the greatest improvement.
Source: §6 · p. 12Takeaways & Limitations
Weekly forecasts capture substantial benefit when loads are relatively stable, while daily forecasts help most when day-specific variation affects violation exposure and full-duration accommodation.
Source: §5.3.3 · p. 12
Abstract
from arXiv · showhide
The rapid expansion of hyperscale data centers is significantly increasing electricity demand in Northern Virginia. Dominion Energy, the region's primary electric utility, must reinforce its transmission network to support this growth. These projects require planned outages that must be evaluated months in advance to support construction planning and outage coordination while meeting NERC and PJM N-1 reliability requirements. Long-term outage studies are commonly performed day by day using monthly or seasonal peak-load assumptions. Although this approach simplifies analysis, it can be overly conservative because it does not capture granular load variability. Consequently, short-duration outages that may be feasible under realistic loading conditions are often postponed or denied, delaying critical transmission expansion and grid modernization projects. This paper evaluates the operational value of incorporating realistic multigranular load forecasts into long-term, contingency-based outage assessments and compares the results with conventional peak-based methods. Daily, weekly, and monthly forecasts are developed using statistical and machine-learning models, including SARIMA, Prophet, Gradient Boosting, and Random Forest, followed by bottom-up temporal reconciliation to maintain consistency across forecast horizons. Results show that granular load forecasts reduce unnecessary conservatism and improve outage accommodation, particularly for short-duration requests, without changing existing reliability criteria. The proposed approach can strengthen long-term outage coordination and support timely transmission reinforcement and grid modernization.
1.1 Introduction
Northern Virginia data-center growth is driving major transmission expansion, while planned outages must be coordinated and assessed against reliability requirements. This study tests whether coherent, finer-grained load forecasts can improve long-term outage assessments without changing those requirements.
1.1 Introduction
- Northern Virginia’s data-center growth has driven roughly 70 GW of large-load interconnection requests, requiring transmission reinforcement and planned outages.Roughly 25 GW of requests already have projected connection dates through 2031.Source: §1.1 · p. 2
- Planned outages require advance coordination and N-1 security assessment, with outages exceeding 30 days submitted to PJM about one year ahead.Future topology, outage schedules, generation availability, and load forecasts add uncertainty to these assessments.Source: §1.1 · p. 2
- Peak-based studies may unnecessarily restrict feasible outages, especially short-duration requests, because they miss daily and weekly load variability.The passage notes that reliability assessments are expected to consider projected loads across a broad range of system conditions.Source: §1.1 · p. 2
- The study compares coherent daily, weekly, and monthly forecasts against monthly peak loading to measure effects on outage feasibility and reliability outcomes.It selects forecasting models by out-of-sample accuracy and reconciles forecasts across granularities while retaining established reliability criteria.Source: §1.1 · pp. 2–3
2 PRELIMINARIES
The preliminaries define the transmission network and outage records, set out daily N-1 assessment, and describe Dominion’s staged outage review and ANODE/TARA workflow.
2 PRELIMINARIES
- The network is modeled as buses, branches, and generators, with each asset assigned to a voltage class.The defined voltage classes are 500, 230, and 115 kV.Source: §2 · p. 3
2.1 Outage Taxonomy
- Each planned outage record identifies an affected asset, its voltage class, and requested start and end dates.The requested duration is measured in days and bounded by the study window.Source: §2.1 · p. 3
2.2 Reliability Requirements
- For each study day, the system condition with active outages is evaluated under N-1 contingency conditions.Each contingency is applied on top of that day’s scheduled outage set.Source: §2.2 · p. 3
2.3 Outage Planning Process
- Dominion’s staged review begins with outage submission through iTOA, then evaluates requests in future reliability studies.Strategic Planning assesses major outages approximately one to five years ahead; later reviews update conditions closer to execution.Source: §2.3 · pp. 3–4
2.4 Outage Reliability Analysis using ANODE/TARA
- ANODE consolidates grid, forecast, and outage inputs; TARA builds daily study cases, and ORA assesses outage and N-1 reliability impacts.ORA classifies cases as approved, controlled, or declined based on violations and whether generation redispatch can mitigate them.Source: §2.4 · p. 4, Figure 1 · p. 4
3 PROBLEM FORMULATION
The formulation seeks to accommodate planned transmission outages while maintaining N-1 reliability under forecast future conditions. It combines load forecasting at multiple granularities with outage accommodation assessment.
3 PROBLEM FORMULATION
- The objective is to maximize accommodation of requested transmission outages while meeting N-1 reliability requirements under forecast future conditions.The formulation considers a set of requested outages within a study window.Source: §3 · pp. 4–5
- Forecasted load at each load bus is represented by day and forecasting granularity.Granularity indexes the multi-granular load forecasts used in the assessment.Source: §3 · p. 5
3.1 Problem Statement
The formulation represents future load at multiple temporal granularities and selects outage approvals that satisfy reliability constraints across the study window.
3.1.1 Multi-granular Load Forecast
- Forecasts are produced at the selected granularity, with weekly and monthly values replicated across the days in their periods.Daily forecasts remain unchanged.Source: §3.1.1 · p. 5
3.1.2 Maximizing Outage Approval
- The objective maximizes approved outages over the study window, subject to AC power-flow and post-contingency thermal and voltage constraints.For each active outage-day pair, x is binary; x=0 indicates that the outage does not contribute to a voltage or thermal violation in N-1 analysis.Source: §3.1.2 · p. 5
3.2 Evaluation Framework
The framework evaluates outage approvals under monthly, weekly, and daily load cases using measures of outage-day feasibility, full-duration accommodation, and the gains from finer granularity.
3.2 Evaluation Framework
- Approval metrics are calculated separately for monthly, weekly, and daily load cases.The cases are denoted month, week, and day, respectively.Source: §3.2 · p. 5
3.2.1–3.2.2 Outage-Day Approval Count (ODAC) and Outage-Day Approval Rate (ODAR)
- ODAC counts the days an outage does not create a violation, while ODAR measures the share of active outages meeting the reliability criterion on a given day.An outage-day with x=0 has no post-contingency thermal or voltage violations.Source: §3.2.1 · p. 5, §3.2.2 · pp. 5–6
3.2.3 Acceptance Gain (G)
- Acceptance gain is the difference in ODAR between two load cases, reported in percentage points.The defined comparisons are daily-over-monthly, weekly-over-monthly, and daily-over-weekly.Source: §3.2.3 · p. 6
3.2.4 Recovered Outage Count (ROC)
- ROC counts outages that are not fully accepted under monthly loading but become fully accepted under a finer-granularity case.The weekly and daily ROC variants count these recoveries under their respective load representations.Source: §3.2.4 · p. 6
3.2.5 Weekly Benefit Ratio (WBR)
- WBR compares average weekly-over-monthly acceptance gain with daily-over-monthly gain; values near one indicate weekly loading captures most of the daily benefit.A lower WBR indicates additional value from daily granularity.Source: §3.2.5 · p. 6
3.2.6 Outage Accommodation Rate (OAR)
- OAR is the fraction of requested outages approved for their full requested duration; the short-duration measure covers outages lasting 1–10 days.Unlike day-level measures, OAR requires approval over the entire outage duration.Source: §3.2.6 · p. 6
3.2.7 Acceptance Opportunity Index (AOI)
- AOI measures the proportion of studied outage-days that meet the reliability assessment, crediting accepted days even when an outage is not approved throughout its full duration.A higher AOI indicates lower violation exposure and greater flexibility to schedule, adjust, or coordinate outages within the study window.Source: §3.2.7 · pp. 6–7
4 PROPOSED METHODOLOGY
The methodology develops and selects load forecasts at multiple granularities, then compares outage reliability assessments while holding study assumptions constant.
4.1 Methods for Load Forecasting
- The forecasting framework generates base forecasts independently for each granularity.The additive forecasting representation includes trend, calendar and seasonal effects, autoregressive lags, weather sensitivity, and residual error.Source: §4.1 · p. 7, §4.1.1 · p. 7
4.1.1 Time-Series Models
- SARIMA, Prophet, and machine-learning regression models forecast load using historical observations and, where applicable, lag, calendar, and temperature features.The methods provide different modeling assumptions for persistence, trend, seasonal patterns, and nonlinear relationships.Source: §4.1.1 · p. 7
4.1.2 Model Selection and Error Metrics
- Models are ranked using rolling-origin cross-validation error metrics and then evaluated on an independent holdout test set.Selection considers forecast error across both stages, as well as stability and interpretability.Source: §4.1.2 · p. 7
4.1.3 Continuous Integration and Continuous Deployment (CI/CD)
- The production workflow validates new SCADA and weather observations, generates features, retrains and monitors models, and produces forecasts, with reevaluation and redeployment when performance degrades.Figure 3 is identified as the load forecasting workflow.Source: §4.1.3 · p. 8, Figure 3 · p. 8
4.1.4 Comparative Outage Reliability Assessment Framework
- The comparative assessment varies monthly, weekly, and daily load representations while keeping network topology, outage requests, contingencies, and reliability criteria consistent.Inactive outage-days and non-convergent outage-days are excluded consistently across the three cases.Source: §4.1.4 · p. 8, Figure 4 · p. 8
5 CASE STUDY
The case study models Dominion Energy’s transmission network while abstracting the exact number of line elements for confidentiality.
5 CASE STUDY
- The modeled network is represented by buses, transmission lines and transformers, and generators, with exact line-element counts withheld for confidentiality.Source: §5 · p. 8
5.1 Multi-granular Load Forecasting Results
The forecasting results use daily peak-load data to compare candidate models and construct temporally consistent daily, weekly, and monthly forecasts. LightGBM performs best on the holdout test, and bottom-up reconciliation derives coarser peak forecasts from daily forecasts.
5.1 Multi-granular Load Forecasting Results
- The forecasting results address the challenge of accurately predicting finer-granularity loads amid weather uncertainty and substantial data-center demand.Dominion serves approximately 5 GW of data-center load, representing 15%–20% of system demand.Source: §5.1 · p. 9
5.1.1 Data Collection and Sampling
- SCADA historian measurements recorded about every four seconds are aggregated into daily peak-load data spanning seven years, with a 365-day test window.The dataset is partitioned into training and test sets for forecasting evaluation.Source: §5.1.1 · p. 9
5.1.2 Model Selection and Model Evaluation
- LightGBM performs best on the holdout test, improving MAPE by 42.4% compared with the baseline model.Table 1 presents average forecasting accuracy across the independent test dataset.Source: §5.1.2 · p. 9, Table 1 · p. 9
5.1.3 Temporal Reconciliation for Multi-granular Forecasting
- Weekly and monthly peak forecasts are derived by taking the maximum of daily forecasts within each period, preserving the non-additive definition of peak demand.This bottom-up reconciliation maintains temporal consistency across forecast granularities.Source: §5.1.3 · p. 9, Figure 5 · p. 10
5.2 Study Configuration
The study compares monthly, weekly, and daily load cases over a six-month horizon while holding the network and reliability assumptions constant. Outage violations are evaluated by outage and date for each case.
5.2 Study Configuration
- The study evaluates 18 cases across monthly, weekly, and daily load representations over December 2026 through May 2027.The six-month horizon is divided into monthly study windows.Source: §5.2 · p. 10
- Network topology, outage requests, contingencies, and reliability criteria remain identical across cases; load forecast is the comparison variable.The cases use the ANODE-TARA outage reliability assessment workflow.Source: §5.2 · p. 10
- The study-period table reports ODAR and acceptance gain, while the multi-granular forecast figure identifies the load cases under comparison.Source: Table 2 · p. 10, Figure 6 · p. 10
5.3 Results
Granular loading generally improves outage-day approval and expands outage accommodation while maintaining the same reliability requirements. Daily forecasts provide the strongest overall benefit, while weekly forecasts capture much of it under relatively stable load conditions.
5.3.1 Day level approval performance
- Daily loading yields the highest average ODAR, with Gday|month gains of 20.90 in January and 21.05 in February.Weekly loading occasionally outperforms daily loading on individual dates, but daily loading provides the strongest and most consistent benefit overall.Source: §5.3.1 · pp. 10–11, Table 3 · p. 11
5.3.3 Acceptance Opportunity and Recovered Outages
- AOI and ROC results indicate that granular forecasts expand scheduling flexibility and turn part of that flexibility into actionable outage opportunities.Daily forecasts provide the greatest benefit when day-specific variations affect violation exposure and full-duration accommodation; weekly forecasts capture much of the value under stable load conditions.Source: Figure 8 · p. 11, §5.3.3 · p. 12, Figure 9 · p. 12
5.3.1 Day level approval performance
- Granular forecasts improve assessment fidelity while preserving the existing reliability requirements.The day-level results show that daily loading generally maintains the highest ODAR across the six-month study period.Source: §5.3.1 · p. 10, §5.3.3 · p. 12
6 CONCLUSION
The conclusion finds that coherent daily and weekly forecasts improve long-term outage assessment without changing reliability requirements. Daily forecasts generally yield the largest benefits, especially for short-duration outages, while weekly forecasts capture substantial value under stable load conditions.
6 CONCLUSION
- Granular load representations reduce unnecessary conservatism and improve outage-day approval and full-duration accommodation, particularly for short-duration outages.The comparison holds network topology, outage schedules, contingency set, and reliability criteria consistent.Source: §6 · p. 12
- Daily loading generally provides the greatest and most consistent improvement, while weekly loading captures substantial benefit during relatively stable load conditions.Source: §6 · p. 12
- The authors conclude that integrating coherent daily and weekly forecasts into outage-analysis workflows can support coordination, reduce avoidable rescheduling, and facilitate timely transmission reinforcement.Future work would assess transmission-transfer-capacity impacts and incorporate forecast uncertainty through probabilistic scenarios.Source: §6 · p. 12