Why Accuracy Matters - AC vs DC Power Flow
For decades, DC power flow has been the standard in power system analysis. It is simple, fast, and for many applications, considered good enough. But as power systems evolve, becoming more congested, dynamic, and constrained, the assumptions that once justified linear approximations no longer hold.
At eRoots, we benchmarked DC power flow against the full nonlinear AC formulation using our own modeling and analysis tools. The goal was to quantify how far linear simplifications can deviate from reality across a variety of networks, from academic test cases to operational models.
Benchmarking Approach
We applied both DC and AC power flow methods on a range of systems. These included well-known Matpower test cases as well as real-world grids used in planning and operations. Inputs, topology, and conditions were kept consistent. Our tools ensured stable initialization and reliable convergence for AC simulations, allowing for direct comparison.
Results
The difference in active power flow between the two methods was substantial. In Matpower cases alone, the relative error ranged from 0.1 percent to more than 500 percent. These are textbook examples, widely used in teaching and research. In operational grids, similar patterns emerged. DC power flow frequently produced inaccurate results when voltages dropped, losses increased, or lines operated near their limits. These errors are often unpredictable and difficult to correct without switching to the full AC formulation.
| name | n_buses | n_branches | time (s) | linear time (s) | DC flow error average (%) |
| case_SyntheticUSA.m | 82000 | 104121 | 2.968 | 0.08632 | 96.58 |
| case_ACTIVSg70k.m | 70000 | 88207 | 4.685 | 0.08175 | 30.52 |
| case_ACTIVSg25k.m | 25000 | 32230 | 0.855 | 0.03105 | 62.21 |
| case13659pegase.m | 13659 | 20467 | 0.282 | 0.01492 | 1580.84 |
| case_ACTIVSg10k.m | 10000 | 12706 | 0.174 | 0.01467 | 4.18 |
| case9241pegase.m | 9241 | 16049 | 0.314 | 0.01125 | 11.35 |
| case8387pegase.m | 8387 | 14561 | 0.373 | 0.00974 | 0.71 |
| case6515rte.m | 6515 | 9037 | 0.182 | 0.01182 | 0.45 |
| case6495rte.m | 6495 | 9019 | 0.202 | 0.00742 | 6.43 |
| case6470rte.m | 6470 | 9005 | 0.192 | 0.00722 | 4.86 |
| case6468rte.m | 6468 | 9000 | 0.112 | 0.01084 | 46.63 |
Implications
Linear power flow models remain popular, especially in planning and optimization. Ironically, these are the very settings where their assumptions fail most often, since they involve stressing the system. Nonlinear models like AC power flow are more accurate, but they demand stricter data quality and make errors harder to detect and debug.
While we cannot eliminate the modelling effort required for nonlinear methods, many of the traditional barriers have been lowered. Improvements in solver performance, initialization techniques, and computing power have made AC power flow both accurate and practical, even for large-scale and operational use.
High-Performance AC Power Flow with GSLV
To support this shift, we developed GSLV, a high-performance power flow engine designed for speed, scalability, and interoperability. GSLV offers assumption-free topology processing, parallel and fast AC power flow computation, native Python and C++ support, and seamless integration with GridCal. It enables efficient benchmarking, scalable planning, and accurate simulation workflows without compromising model complexity or execution time.
Our Perspective
At eRoots, we believe it is time to move beyond oversimplified models. If your work involves analysing stressed conditions, evaluating investments, or preparing for the future of the grid, AC power flow should be your default. With GSLV and our broader modeling ecosystem, we aim to make high-fidelity nonlinear analysis both accessible and practical.
If you are interested in running similar benchmarks on your networks or want to learn more about our tools, feel free to get in touch.
