EFFECTIVE OPTIMAL POWER FLOW STRATEGIES

USING GRAPH ARTIFICIAL INTELLIGENCE

(PAPER ID- 1091)

Presented By:
Name – Manish Nandy
Affiliation- Kalinga University, Raipur, India.
 Contents
1. Introduction

2. Literature Review

3. Motivation for the Proposed Work

4. Proposed Model/Work

5. Result Analysis

6. Conclusion

7. References
 1. Introduction
• One of the most important problems posed in the
energy sphere is apex optimization for workable
energy transfer or optimal power flow. It is done
for system design, price formation of electricity
on the next day markets and also for efficient day
by day generation scheduling.
• This issue was raised over half a century ago and
it is quite easy to state that no effective resolution
has been found which would have enabled the
world’s economy to cease wastes exceeding a
hundred billion dollars.
• Quite simply, electrical supply and demand
management strategy is implemented through Figure 1: Workflow depicting
power flow control where a grid’s generation the methodology for utilizing
units are instructed on what quantity of power Graph Neural Networks in
they are to generate in line with a defined the OPF problem
demand that has been set.

12/25/2024
 2. Literature Review

Figure 2: Diagram depicting the approach for solving the OPF problem through Graph Neural Networks

•As far as we are concerned, the OOPF it is well and good to state the core idea of the
OOPF is able to be considered LP linear formulation if in the case that passive and
active P and Q parameters are included considering the DCOPF model. In an OOPF
context, It can be assumed that P and Q parameters are exercised in a logical order in
the OOPF model whereby Pe is deployed in active DCOPF However, the power flow
problem is known to be handled by five solvers CONOPT, IPOPT, KNITRO, MINOS
and SNOPT.
•Nevertheless, one can assume that these solvers are slow in convergence regarding
large networks. More specifically, we focus on the use of IPOPT since it is reported to
be the one most advantageous. Also, focus of this paper is using imitation learning to
accomplish a specific problem in a distributed and efficient manner. In information so
provided, let us consider p* as the solution which IPOPT is said to be able to provide.

12/25/2024
 3. Motivation for the Proposed Work
This paper particularizes to the case where the
adjacency matrix W is used. However, one can use
any matrix, which honour the sparsity of the
interconnection of the graph M. Matrices like these
are commonly referred to as graph shift operators
and other popular ones in the GSP literature include
the Laplacian, random walk Laplacian, and several
of their normalized versions. Graph convolution
makes use of the shift operator as its most essential
component. So the discrete version, graph
convolution is viewed as an operation whereby an

colored disks as a linear combination. With Hk ∈

information rotates around as illustrated by the

Every output Y ∈ RN×G is considered an additional

RF×G the coefficient matrix for k = 0,…,K − 1.
Figure 3: Conceptual framework
graph signal that encodes G features over every
outlining how Graph Neural
node of the graph. The operation X = WX, as we
Networks integrate with existing
have seen in this work, constitutes the evaluation of
the function as a linear combination of the
power grid models
arguments that are local to it.
12/25/2024
 4. Proposed Model/Work

Figure 4: Strategy for implementing Graph Neural Network-based OPF solutions, focusing on system
integration

It is possible for a trained model to be employed for real-time OPF optimization

provided that the configuration and generation data of a power system is made
available. A GNN based model is used and its output is aimed at the intelligent control
of the systems operation, which encourages the automatic g generation and flow
predictions for the network. Security of the model in relation to grid structure and
operation mode alterations is also exercised in order to ascertain its adequateness over
different systems.
12/25/2024
 Method
4. Proposed Model/Work Description Key Applications Advantages & Challenges
Graph Neural Networks (GNN) Utilizes graph-based models to learn the Grid optimization, power system analysis, Effective in capturing complex
relationships between various nodes real-time power flow control. interdependencies but requires large
(power generation and consumption datasets and computational resources.
points) for optimal power distribution.

Message Passing Network (MPN) A type of GNN where nodes exchange Distributed energy systems, microgrid Can handle dynamic systems with multiple
information (messages) across edges to management. variables but computationally intensive for
update their state and optimize power large grids.
flow decisions.

Deep Graph Convolutional Networks A deep learning approach that aggregates Smart grid management, voltage High accuracy and can capture non-linear
(DGCN) information from neighboring nodes to regulation, fault detection. relationships, but training can be slow with
improve prediction accuracy for power large datasets.
flow.

Graph Convolutional Networks (GCN) Uses convolutional layers to process graph- Load forecasting, power generation Simple architecture, but performance can
structured data, helping model power grid scheduling, transmission line optimization. degrade in highly complex grids with noise
behavior and solve optimal power flow or missing data.
problems.

Graph Reinforcement Learning (GRL) Combines reinforcement learning with Real-time power management, adaptive Enables continuous learning and
GNNs to continuously improve power flow grid control. adaptation, but requires significant
decisions based on real-time feedback. computational resources.

Node Embedding Techniques Maps nodes to continuous vector spaces Grid analysis, fault detection, performance Reduces dimensionality and improves
to better understand the structural and monitoring. scalability, but may lose detailed local
functional relationships within the grid for information in large grids.
optimized flow.

End-to-End Optimization A framework where GNN models are Grid optimization in smart cities, End-to-end learning for improved accuracy,
directly trained to optimize the power flow renewable energy integration. but requires substantial training data and
decision-making process from input to high computational costs.
output.

Table 1: Optimal Power Flow Using Graph Neural Networks

12/25/2024
 Method Solution Time
5. Result Analysis
Convergence Objective Computation
(s) Rate (%) Function Value Complexity
(MW) (Order) Epochs Training Training Validation Accuracy
Time (s) Loss Loss (%)

50 120 0.02 0.03 95

Newton- 12 98 135 O(n^3)

Raphson
Method
100 240 0.015 0.025 96

Graph Neural 30 95 136 O(n^2)

Network
(GNN)
200 480 0.01 0.02 97

Genetic 45 93 134 O(n^2)

Algorithm
(GA) 300 720 0.005 0.01 98

Particle Swarm 40 94 133 O(n^3)

Optimization
(PSO)

Table 3: GNN Model Training

Performance

Table 2: Performance Comparison of

GNN and Traditional OPF Methods
12/25/2024
 6. Conclusion
• The core activity of any professional in the power grid is the solving OPF problem.
OPF is a global optimisation problem. But the method is costly and efficient for
small networks only. In this paper, we proposed the use of GNNs in an attempt to
reach the optimal OPF solution.
• GNNs are an advantage and process information with fewer calculations since they
are spatially distributed over the network. We train GNNs for a given network state
and approximate the output of a good IPOPT solution’s output during imitation
learning. We employ the conventional IEEE-30 and IEEE118 benchmark problems
and conduct the experiments showing that local methods which take into full
consideration the grid structure are the better ones in comparison to other methods.
• The use of Graph Neural Networks together with Optimal Power Flow can be
considered as a new and unique approach to power systems. Despite their
numerous advantages, the application of traditional OPF techniques is limited by
factors of scalability, efficiency, and even versatility, concerns that will likely
deepen due to the addition of renewable energy sources, distributed generation, or
increasing variability in demand

12/25/2024
 7. References
1. M. B. Cain, R. P. O’Neill, and A. Castillo, “History of optimal power flow and formulations,”
Federal Energy Regulatory Commission, Increasing Efficiency through Improved Software, pp.
1–31, Dec. 2012.
2. B. C. Lesieutre and I. A. Hiskens, “Convexity of the set of feasible injections and revenue
adequacy in FTR markets,” IEEE Trans. Power Syst., vol. 20, no. 4, pp. 1790–1798, Nov. 2005.
3. D. Bienstock and A. Verma, “Strong NP-hardness of AC power flows feasibility,” Operations
Research Letters, vol. 47, no. 6, pp. 494–501, Nov. 2019.
4. S. Chatzivasileiadis, “Lecture notes on optimal power flow OPF,” arXiv:1811.00943v1 , 2 Nov.
2018.
5. D. K. Molzahn and I. A. Hiskens, “Convex relaxations of optimal power flow problems: An
illustrative example,” IEEE Trans. Circuits Syst. I, vol. 63, no. 5, pp. 650–660, May 2016.
6. M. R. AlRashidi and M. E. El-Hawary, “Applications of computational intelligence techniques
for solving the revived optimal power flow problem,” Electric Power Syst. Research, vol. 79,
no. 4, pp. 694–702, Apr. 2009.
7. A. G. Bakirtzis, P. N. Biskas, C. E. Zoumas, and V. Petridis, “Optimal power flow by enhanced
genetic algorithm,” IEEE Trans. Power Syst., vol. 17, no. 2, pp. 229–236, May 2002.

12/25/2024