Future GB transmission grids will be dominated by converter-interfaced renewables whose dynamics are invisible to the steady-state test cases still common in planning studies. We present a Python framework that converts any multi-bus, multi-generator power-flow file into a positive-sequence, grid-code-compliant dynamic model. Demonstration on a 100-bus reduced GB system shows the full end-to-end workflow.

A script first ingests a PowerFactory network, labels every generator (synchronous, grid-forming, grid-following, interconnector) and fuel, and attaches the matching open-source or user-supplied controller template. The largest units are islanded on an infinite bus, tuned to National Electricity System Operator (NESO) frequency and voltage-envelope limits, then re-inserted, eigenvalue scans damp residual modes.

Dynamic tuning is driven by a live-portfolio “dual-hybrid” optimiser. Particle Swarm (rapid exploration) and CMA-ES (covariance-driven exploitation) run concurrently in a shared simulation queue; after each evaluation wave, a sliding-window credit allocator shifts population budget toward the better performer while leaving ≥20 % evaluations for the other to preserve diversity. Elites are cross-seeded every few waves to escape cycles. Fitness is mean-absolute error against NESO envelopes for nadir, RoCoF, voltage recovery and settling time measured in nonlinear RMS fault and step tests. When progress stalls near tolerance, the process switches to a local surrogate phase: a Gaussian process trained on the best global samples, fast L-BFGS-B descent and Bayesian optimisation to polish uncertain pockets. Coherence analysis then zooms into inverter-dense or weakly damped zones and reruns the same portfolio on the reduced parameter set. System operational scenarios are run where each instability triggers a brief retune until no further secure-load rise is possible.

Running in parallel is a graph-neural-network meta-initialiser that learns from every completed optimisation. It maps “network snapshot → controller gains” and, after each overnight fine-tune, seeds future runs with smart defaults that aim to shorten global search time.

The output is a tuned .pfd file, and an auto-generated PDF report that flags weak areas, giving system planners and researchers a repeatable, overnight route to up-to-date dynamic data as portfolios or grid-codes evolve. Replacing weeks of manual trial-and-error, the framework demonstrates how machine-learning automation can supply the dynamic models essential for a secure, net-zero GB grid. The work is purely simulation-based and underpins a Master’s thesis collaboration between Blake Clough Consulting and The University of Manchester.