Skip to content
Request a benchmark

AC load flow solver

Faster, more reliable load flow for the grids that can't afford to be wrong.

Velra is a fast, robust AC load flow solver for grid operators. It plugs into the tools you already use and is built for routines where load flows run by the thousand.

Request a benchmark See how it performs
  • 2x to 7x+ faster than a Newton-Raphson baseline, with larger gains on larger systems
  • Robust on hard cases: fewer iterations, and early detection of divergence
  • Plugs in to pandapower today, with GridCal and PowSyBl in progress
  • Deterministic and auditable: a mathematical method, not AI

How the speed-up was measured: Velra and a Newton-Raphson baseline, both implemented in C (not MATLAB or Python), on the same cases. About 2x on standard academic test networks (IEEE 14, IEEE 118, Polish 2383) and about 3x on a large synthetic US model.

The problem

Grids are harder to operate, and the load flows behind every decision keep multiplying.

More renewables and more flexibility move the operating point more often, and closer to its limits. Operators respond with more computation.

Load flows now run by the hundreds, and in some routines by the millions. Every run has to be fast, and failures have to be rare and predictable.

Some routines fall back to DC load flow today, not because AC is wrong for the job, but because there are too many cases to run.

When thousands of load flows are computed, nobody has time to rescue a divergent case by hand.

Where the load flows run

  • Contingency analysis

    Every outage case (N-1, N-k) is its own load flow.

  • Remedial action optimisation

    Candidate actions are checked with repeated load flows.

  • Capacity calculation

    Repeated across scenarios and time steps.

  • Merged models

    Large models solved for many time steps, often overnight.

Technology

How Velra performs

Velra is a proprietary AC load flow (power flow) solver for deterministic studies. Here is how it behaves against the Newton-Raphson baseline most tools use. How it works inside stays with us.

01 Speed

2x to 7x+ faster, growing with system size

Measured C-to-C, so the comparison is like for like rather than against MATLAB or Python tools. Gains grow with system size: about 2x on standard academic test networks and about 3x on a large synthetic US model covering Texas and surrounding states. Velra runs on standard CPUs today, with GPU acceleration as further headroom.

Speed-up over the Newton-Raphson baseline (1x = same speed)

  • Range: 2x to 7x+, depending on the system
  • Academic test networks (IEEE 14, IEEE 118, Polish 2383): about 2x
  • Large synthetic US model (Texas and surrounding states): about 3x
  • Very large systems: values to follow
  • Gains grow with system size
Measured, approximate Points are approximate. The band is the stated range; the trend line is qualitative.

02 Robustness

Predictable on hard cases, quick to flag divergence

Velra converges on the same cases as Newton-Raphson. The difference shows on hard cases near the edge of solvability, where predictable behaviour matters as much as speed.

  • Fewer iterations on hard cases
  • Softer behaviour when a case starts to diverge: it may stray briefly, then recover
  • Early detection when a case diverges
Example operating point
Shade map by iterations for

Stress parameter

No solution exists Same convergence region for both

Loading level

  • fewer iterations
  • ≤4 iterations
  • 5 to 6 iterations
  • 7 to 9 iterations
  • 10 to 13 iterations
  • 14+ iterations
  • more

Error per iteration, log scale

Newton-Raphson
Converged in 17 iterations
Velra
Converged in 8 iterations

Illustrative Iteration counts and traces are illustrative, not measured data.

Use cases

Built around how each operator works

Pick your role to see where faster, more predictable load flow changes the work.

TSOs

Contingency analysis and remedial actions, in AC

Contingency analysis and remedial action optimisation run hundreds to millions of load flows.

  • Run more contingencies (N-1, N-k) inside operational time windows.
  • Stay in AC where routines fall back to DC today only because of the number of cases.
  • Fewer divergent runs to chase by hand, and an early signal when a case diverges.

RCCs

Large merged models, many time steps, overnight

Pan-European models merged and solved for many time steps, with no time for manual intervention.

  • Robust on very large merged models, where speed gains grow with network size.
  • Early detection of divergent cases, so overnight runs do not stall.
  • Built to fit open-source stacks such as PowSyBl and OpenRAO.

DSOs

More studies on the hardware you already have

Connection and hosting-capacity questions multiply as distributed generation grows.

  • Faster hosting-capacity and connection studies.
  • Full AC load flow on larger meshed networks.
  • Works inside pandapower workflows today.

Integration and security

Plugs into your workflow. Stays inside your perimeter.

A proprietary solver core with an open integration surface. Velra plugs into open-source power system tools as a solver component, with no rip-and-replace.

  • Closed core, open integration surface

    Adapters for open-source stacks, so Velra works inside the tools you already run: pandapower today, with GridCal and PowSyBl in progress.

  • Runs in your environment

    Velra modules run inside your infrastructure. Your grid data does not leave it.

  • Standard data formats

    Support for CIM/CGMES and PSS/E RAW is planned.

  • Deterministic and auditable

    A mathematical method, not AI, so its behaviour can be audited and explained.

  • Verification and security

    Benchmark Velra on your own cases. We are open to verification arrangements that fit your security processes.

  • Integration status

    • pandapower Toolchain Available
    • GridCal Toolchain In progress
    • PowSyBl via Open Load Flow In progress
    • CIM/CGMES Data format Planned
    • PSS/E RAW Data format Planned

    Only items marked Available are released. The others are in development.

run_loadflow.py Illustrative API 

                
                    
                    import pandapower as pp
                  
                    
                    import pandapower.networks as pn
                  
                    
                     
                  
                    
                    net = pn.case118()
                  
                    
                     
                  
                    
                    pp.runpp(net)  # default: Newton-Raphson
                  
                    
                    pp.runpp(net, algorithm="velra")  # illustrative

The solver argument is a placeholder, not the published interface.

Expected solver time, IEEE 118

Newton-Raphson (default) 1.00 · converged
Velra ~0.50 · converged

From the C-to-C benchmarks (about 2x).

A one-line solver swap in a familiar workflow.

Benchmark

Benchmark Velra on your own cases.

A like-for-like comparison of Velra and your current solver, on cases you know well.

  1. 1

    Scope

    Agree the networks, cases and KPIs, and how data is handled. NDA and cybersecurity processes are welcome.

  2. 2

    Run in your environment

    Velra and your current solver run on the same models, inside your infrastructure.

  3. 3

    Review the KPIs

    We go through solver time, iterations and convergence together, and agree what comes next.

Request a benchmark

A few lines is enough. We reply by email.

Prefer email? Write to contact@velra.se

We use your details only to reply.

Team

The people behind Velra

Mark leads the technology. Liam, Zoé and Daniela, all MSc students in Entrepreneurship and Business Design at Chalmers, run business development: customer discovery, benchmarks with operators, partnerships and bringing Velra to market.

The Velra team, from left: Zoé Opdendries, Mark Jeeninga, Liam Aljundi and Daniela Padilla
Mark Jeeninga, PhD

Mark Jeeninga, PhD

Technical lead

Inventor of the Velra solver. Postdoctoral researcher at Lund University, Department of Automatic Control, and expert in power grids and networked dynamical systems.

Liam Aljundi

Liam Aljundi

Business development

Former lead design researcher at Arduino Education, turning user research into product roadmaps. Background in interaction design, with further study in innovation strategy and IP at UC Berkeley Haas.

Zoé Opdendries

Zoé Opdendries

Business development

Software engineer by training (BSc, Chalmers) and former sole frontend developer and UI/UX designer at LexEnergy, an EV charging startup. Works across the technical and commercial sides, from IP to sales.

Daniela Padilla

Daniela Padilla

Business development

Co-founded Alkimia, an award-winning chocolate brand, and led its product and go-to-market work. Product designer for clients in six countries; Swedish Institute Scholar.

Backed by

Affiliated with

FAQ

Questions operators ask

How does Velra compare with Newton-Raphson?

Newton-Raphson is our baseline because most tools use it. Measured C-to-C, Velra is 2x to 7x+ faster depending on the system, and it converges on the same cases. On hard cases it typically needs fewer iterations and flags non-convergence earlier. It is our own proprietary solver, not a modified Newton-Raphson.

Is Velra a replacement for PSS/E or PowerFactory?

No. It sits alongside your existing tools as a solver component, so your models, workflows and validation stay as they are.

Can we run it in our own environment?

Yes. Velra modules run inside your infrastructure, and your grid data does not leave it.

It is closed source: how do we trust it?

Start by comparing it with your current solver on your own cases. Beyond that, we are open to verification arrangements that fit your security processes, and NDA and cybersecurity reviews are welcome.

Is Velra open source?

No. The solver core is proprietary. Adapters let it plug into open-source tools: pandapower today, with GridCal and PowSyBl in progress.

Will it speed up my whole workflow?

Not necessarily by the same factor. Load flow is often a large part of the work, but not always the only bottleneck. We measure solver time like for like and look at your workflow with you, rather than promise end-to-end speed-ups.

What accuracy and convergence guarantees are there?

Velra solves the same AC power flow equations to the tolerances you set, so results can be compared case by case. We make no unqualified guarantees: validation on operator data is in progress, and a benchmark on your own cases is the best test.

Is the method published?

Publication and IP strategy are being finalised. Ask us what we can share today.

What hardware does it need?

Standard CPUs. GPU acceleration is further headroom we are exploring.