Dr. Eshan SinghGrid Reliability Lab
Browser local · single-area · hourly Monte Carlo
Reliability report

California 2035 · Balanced reference

Preparing the hourly system model…

Waiting for run
LOLE
—days / yr

Expected shortfall-days

LOLH
—hours / yr

Expected shortfall-hours

EUE
—MWh / yr

Expected unserved energy

Annual risk
—%

Years with ≥1 shortfall

Study not run

The portfolio has not been stress-tested yet.

Run the model to compare expected shortfall-days with the common 0.1 day/year reference criterion.

Accredited proxy — planning margin
Hourly LOLP · timing of risk

Risk landscape

Hourly probability of firm-energy shortfall by month

0Run maximum
Contiguous shortfall episodes

Severe event explorer

Ranked by unserved energy across all simulated years

StartStress patternDurationPeak gapUnserved energy
Run a study to reveal the most severe simulated events.
Browser-local research model

How the reliability study works

This browser model estimates resource adequacy: whether available bulk-system supply can cover hourly firm demand. It is inspired by the transparent workflow of tools such as SERVM and PRAS, but it is not a production-cost model or a substitute for a regulatory study.

Learn grid reliability from first principlesOpen the student guide to adequacy, shortfall metrics, Monte Carlo studies, firm need, EFC, and ELCC.
Hourly trace→Correlated stress→Persistent outages→Storage dispatch→Shortfall metrics
01

Metrics with explicit event periods

LOLE · days/year

Average number of calendar days per simulated year containing at least one firm-energy shortfall hour.

LOLP · probability

Share of simulated case-years with a shortfall in a particular hour. The risk heatmap aggregates hourly LOLP by month and hour of day.

LOLH · hours/year

Average number of hourly intervals per year with positive unserved firm demand.

EUE · MWh/year

Average annual sum of hourly unserved demand. This captures both occurrence and depth.

LOLEV · events/year

Average number of contiguous shortfall episodes, regardless of how many days an episode crosses.

NEUE · ppm

Expected unserved energy divided by expected annual demand energy, multiplied by one million.

Annual risk · %

Share of simulated years containing one or more shortfall hours. It is not interchangeable with LOLE.

0.1 day/year does not mean 2.4 hours/year.The familiar “one day in ten years” reference counts shortfall-days, while LOLH counts hours and annual risk counts affected simulations. The app keeps these distinct.
02

Chronological simulation

Every simulated case walks through all 8,760 hours in order. When a multi-year ensemble is loaded, each supply-weather year receives the user-selected number of independent outage draws; the displayed case count keeps those dimensions explicit. For a system study, demand, solar, and wind retain their within-year alignment. For a facility study, constant or uploaded demand is repeated unchanged across the selected supply-weather years while solar, wind, hydro uncertainty, common-mode events, and the thermal/import stress signal still vary by weather year.

  • Weather stress changes several linked quantities together. In the single-year and explicit stress-overlay modes, heat raises demand while reducing renewable output and thermally derating firm supply and imports. Ensemble Reference mode does not add synthetic load or renewable variation.
  • Demand basis determines whether load comes from the weather set, stays constant at the reference peak, or follows an uploaded 8,760-hour facility shape. Flat and uploaded demand are authoritative; California system load is not imposed on them. Annual forecast uncertainty still scales the full facility shape, and an optional cooling uplift can link facility demand to the supply-weather heat signal.
  • Common-mode stress is a summer-only, low daily-onset 1–4 day regional event that applies one shared 6–18% derate to sensitive firm resources and a larger derate to imports. It is distinct from independent unit outages.
  • The 0.2%, 0.6%, and 1.2% inactive-day onset rates imply roughly a 22%, 52%, and 77% chance of at least one common-mode event in the 122-day summer window for Reference, Hot & dry, and Compound modes, respectively.
  • Dispatchable fleets use a two-state failure/repair process. Outages persist across adjacent hours from the entered forced-outage rate and mean repair time.
  • Hydro follows a seasonal capability shape with a trial-level drought draw.
  • Imports have a firm transfer cap and are derated during correlated heat and common-mode events.
  • Storage charges from hourly surplus and discharges against firm-energy deficits, subject to power, energy, efficiency, outages, and state of charge. Every independent trial starts at the entered Jan 1 SOC; there is no warm-up or cyclic end-state constraint.
  • Demand response has a daily energy budget equal to capacity × entered duration. The greedy dispatch uses this expiring daily budget before carry-over storage; it does not optimize flexibility with perfect foresight.
  • The shoulder-maintenance input is a transparent fleet-wide capability derate in February–May and October–November, not a unit-level maintenance schedule.
03

California reference case

The single-year option is normalized from this site’s 8,760-hour 2025 California dataset. The optional 25-year library is generated from official CPUC SERVM 2026-vintage weather-linked shapes for 2000–2024: CAISO consumption-load profiles for 2035 paired with the published corrected 2028 baseline utility-scale solar and wind mix. Each series is normalized and rescaled by this app’s demand and resource controls.

Outage draws do not create additional independent weather observations.A study with 25 weather years × 50 outage draws contains 1,250 simulated case-years but only 25 distinct historical-weather chronologies. Each draw is a new year-long equipment failure and repair sequence. The Weather years view therefore reports each year separately as well as the ensemble average.

The ensemble preserves the published load–solar–wind weather-year pairing, but it is modeled/weather-normalized input—not raw observation—and it is not an official 2035 portfolio or CPUC reliability result. Demand modifiers such as behind-the-meter PV are not modeled separately. Official CPUC weather-year ambient derates and hydro series are not yet bundled: Reference mode instead uses a transparent daily load-factor percentile as the thermal/import derate proxy and a trial-level 0.82–1.04 hydro multiplier. Tail-risk and climate-shift overlays remain separate from the historical-frequency result.

This is deliberately labeled a reference case, not a calibrated SERVM result. CPUC’s regulatory studies also include unit detail, neighboring regions, hydro constraints, ambient derates, emergency procedures, and transmission limits beyond this browser model.

04

Reading uncertainty and diagnostics

The LOLE interval conservatively envelopes the across-trial normal interval with a count-appropriate Poisson interval; if no shortfall-day is observed, its upper bound uses the Poisson zero-count limit rather than incorrectly reporting zero uncertainty. Rare events need many trials, and the interval does not cover input-weather or other model-form uncertainty. The stress signature sums conditions observed during shortfall hours. It is diagnostic—not causal attribution or ELCC.

Firm need searches for perfectly firm capacity to add or remove until the portfolio reaches the entered LOLE target. Capacity value uses matched, common-seed simulations with and without a selected resource. EFC is the firm-equivalent value used for the nameplate capacity-credit percentage. ELCC is reported separately as the increase in the reference-peak load parameter while preserving the modeled load shape; it can exceed nameplate when binding risk hours occur below that reference peak, without implying that the resource can deliver above its power rating. Both are research previews: sampling intervals describe Monte Carlo precision, not weather-data or model-form uncertainty.

The “accredited proxy” uses simple technology weights only as orientation. It is independent of the simulation-based firm-need and capacity-value calculations; the simulation metrics, not that proxy, determine the verdict.

05

Important limits

The model is a one-zone adequacy screen with reliability-directed dispatch. It omits zonal transmission, neighboring market dispatch, economic commitment, ramping, fuel supply networks, reserves, planned-maintenance optimization, hybrid interconnection limits, distribution reliability, dynamic stability, and black-start/restoration. Historical-pattern stress tests cannot establish the probability of unprecedented climate events.

06

What production tools add—and what the web can do

The interface labels capability by fidelity rather than implying commercial parity:

  • Browser local Chronological adequacy, loaded weather ensembles, per-weather diagnostics, and bounded firm-need or capacity-value searches run in Web Workers. Inputs remain in this tab and no backend account is required.
  • Research preview Firm need, EFC, and ELCC use transparent common-seed searches but remain single-area, reliability-directed approximations until benchmarked against production studies.
  • External solver required Multi-area market competition, unit commitment, detailed hydro and fuel constraints, emergency operating procedures, AC power flow, and dynamic stability need substantially deeper data and solver infrastructure.

Commercial and regulatory tools also add extensive validation, shared scenario governance, and very large parallel workflows. The current research features need no Render-style backend; a future compute service would extend scale and operational detail while the product remains a webapp.

Primary references

NERC · Probabilistic Adequacy and Measures NREL · Probabilistic Resource Adequacy Suite CPUC · System Reliability Modeling Datasets 2026 GE Vernova · PlanOS Resource Adequacy Astrapé · SERVM ELCC study methodology