Physics-Informed Neural Networks for Power Systems (1911.03737v3)

Abstract: This paper introduces for the first time, to our knowledge, a framework for physics-informed neural networks in power system applications. Exploiting the underlying physical laws governing power systems, and inspired by recent developments in the field of machine learning, this paper proposes a neural network training procedure that can make use of the wide range of mathematical models describing power system behavior, both in steady-state and in dynamics. Physics-informed neural networks require substantially less training data and can result in simpler neural network structures, while achieving high accuracy. This work unlocks a range of opportunities in power systems, being able to determine dynamic states, such as rotor angles and frequency, and uncertain parameters such as inertia and damping at a fraction of the computational time required by conventional methods. This paper focuses on introducing the framework and showcases its potential using a single-machine infinite bus system as a guiding example. Physics-informed neural networks are shown to accurately determine rotor angle and frequency up to 87 times faster than conventional methods.

• The paper introduces a novel framework that embeds physical constraints into neural network loss functions to accurately capture rotor angle and frequency dynamics.
• It shows that the PINN approach achieves computational speeds 87 times faster than conventional numerical methods while preserving high fidelity.
• The study demonstrates effective system identification by accurately estimating uncertain parameters such as inertia and damping with limited data.

Physics-Informed Neural Networks for Power System Applications

Physics-Informed Neural Networks (PINNs) represent a compelling methodology for enhancing the analysis and simulation of power systems. This paper introduces a novel approach leveraging PINNs to address the frequency dynamics within power systems. Exploiting physical laws inherent in power system models, this framework provides an efficient computational strategy, reducing the reliance on large training datasets while achieving high fidelity in capturing system dynamics.

Framework and Methodology

The paper presents a training procedure that incorporates the differential and algebraic equations, which underpin power system models, directly into the neural network architecture. This approach reduces the computational burden typically associated with traditional machine learning methods that require extensive training data and complex neural network architectures. PINNs work by embedding the physical constraints directly into the loss function of the neural network, enabling it to learn not just from data but also from the physics describing the system. In particular, this paper focuses on simple yet illustrative Single Machine Infinite Bus (SMIB) systems to validate the approach.

Experimental Validation and Results

The research highlights several significant findings through simulations. By employing a neural network training strategy informed by the swing equation, the framework effectively predicts rotor angle and frequency dynamics far more efficiently than conventional numerical methods. Specifically, PINNs accomplished this computation 87 times faster than traditional methods, maintaining accuracy while significantly reducing computational time. This performance attribute is crucial for real-time applications where rapid decision-making and response are required.

Furthermore, the paper demonstrates that PINNs can perform system identification tasks to determine uncertain parameters such as inertia and damping with high accuracy, using limited data. This ability has substantial implications for dynamic state estimation and monitoring power system operations under varying conditions, providing system operators with a robust tool for assessing and ensuring system stability.

Implications and Future Directions

The implications of deploying PINNs in power systems are considerable. Fast and accurate dynamic predictions can greatly enhance operational efficiency and system security. Moving forward, this framework offers a pathway to developing new numerical solvers and optimizations within power system applications. With further development, PINNs could potentially handle more elaborate systems featuring complex, higher-order dynamics and multiple variables.

The challenges that remain involve scaling this approach to larger, more intricate power system models beyond the SMIB case. Expanding the capability of PINNs to model wide-ranging phenomena in larger grids, involving diverse dynamic behaviors and interactions, represents a promising avenue for future research. Additionally, integrating PINNs with existing neural network verification methods could provide much-needed guarantees of reliability and performance, promoting broader acceptance and integration within the power systems community.

In conclusion, this paper heralds a significant methodological enhancement in how dynamic power system behaviors are modeled and computed, paving the way for advanced AI-assisted decision-making in managing modern power systems.