Time-Series & Sensors

Adaptive LSTM-RNN Dynamic Equivalents for Power Systems

An Adaptive Dynamic Model for Reducing the Size of A Power System for Large Variations in the Operating Point using LSTM RNNs

A smaller, faster stand-in for a large power grid that keeps up as operating conditions change.

The problem

Stability studies of large grids with lots of renewables are slow, and reduced models break when the operating point shifts.

The approach

A many-to-one LSTM predicts tie-line power flows from boundary measurements, and Bayesian hyperparameter optimisation retunes it when the operating point moves.

The outcome

Maintained accurate predictions across multiple load-flow conditions on the simplified Australian 14-generator system.

In detail

Co-developed a black-box dynamic equivalent model for large interconnected power systems using LSTM recurrent neural networks. Boundary measurements are used to predict tie-line power flows, while a shift index detects significant operating-point changes and triggers Bayesian hyperparameter optimisation when required. The approach was evaluated on the simplified Australian 14-generator system and maintained strong prediction performance across multiple load-flow conditions.

Built with

• Power Systems
• LSTM
• Bayesian Optimisation

Source

• An Adaptive Dynamic Model for Reducing the Size of a Power System for Large Variations in the Operating Point Using LSTM RNNs, 2023 IEEE International Conference on Energy Technologies for Future Grids (ETFG), 2023

More in Time-Series & Sensors

• Probabilistic Deep Ensembles for FRDA Severity Assessment
• Cross-Device Domain Adaptation for FRDA Assessment
• AIM-M: Multiview IMU Assessment of Friedreich Ataxia

Next project: Robot Vision for Search-and-Rescue Facial Recognition

By Kanishka Ranaweera, PhD, Machine Learning Engineer in Melbourne, Victoria, Australia. hello@kanishkabuilt.me