Electricity storage technology, smart grids and GNNs

introduction

Nuclear fusion technology, as described in ‘Nuclear fusion and AI technology’, is a field where vigorous research is being conducted as a next-generation power generation technology, while technologies for storing and utilising electric power, such as those described in ‘Development of the world’s first “electric carrier”: bringing surplus renewable energy to cities’ and ‘CO2-free hydrogen production using electricity from nuclear power’, also occupy an important position in energy technology. Technology for storing and utilising electricity also occupies an important position in energy technology. Here, we will discuss this power storage technology, smart grids and the application of AI technology to them.

power storage technology

Power storage technology is a generic term for technologies that temporarily store electricity and release it when needed, mainly when supply and demand for electricity do not match or to regulate the variable generation of renewable energy sources. Key electricity storage technologies include

1. Battery technology: storage batteries are technologies that store electrical energy through chemical reactions and release it when needed. Common storage battery technologies include lead-acid batteries, lithium-ion batteries and sodium-sulphur batteries, which are used on a wide scale in household and industrial applications.
2. Pumped hydro storage: a technology that stores electricity by converting the power into a difference in the height of water, lifting the water higher when demand is higher and lowering it when demand is lower.
3. Superconducting magnet storage: superconducting magnets operate at very low temperatures, resulting in near-zero electrical resistance and efficient energy storage and release. Research is also being conducted into the use of superconducting magnets to store electricity.
4. Compressed air energy storage: a technology that uses electricity to compress and store air. When demand increases, the compressed air is released to turn turbines and generate electricity.
5. Thermal energy storage: technology that stores thermal energy and releases it when needed to generate electricity. Includes thermal solar storage and thermal storage technology using molten salts.
6. Power-to-gas conversion: a method of using excess electricity to split water to produce hydrogen. This hydrogen may later be used to generate electricity via fuel cells.

These power storage technologies have become an important factor contributing to the security of energy supply and the effective use of renewable energy, and the evolution and diffusion of the technologies is expected to improve the efficiency and sustainability of electricity networks.

Smart grids and electricity storage technologies

Smart grid is a generic term for systems that utilise information technology to efficiently control electricity networks and maintain a balance between supply and demand, and will have the following main characteristics and roles

1. Real-time monitoring: smart grids have the ability to collect and analyse data on electricity supply, demand and energy usage in real time. This enables efficient control and operation.
2. Demand response: smart grids have mechanisms in place to limit electricity usage during times of peak demand. This helps to reduce over-consumption of electricity and maintain a balance between electricity supply and demand.
3. Distributed energy: distributed generation facilities are increasing, as they utilise more renewable energy sources (e.g. solar, wind). Smart grids can properly integrate and control these power generation sources.
4. Communication and control: communication technologies can be used to co-ordinate and control different elements of the electricity network in real time, thereby improving the stability and reliability of the electricity supply.

In contrast, the combination of the aforementioned power storage technologies makes it possible to co-ordinate energy fluctuations within the smart grid, as described below.

1. Regulating fluctuations: renewable energy sources generate fluctuating amounts of electricity depending on weather conditions, and electricity storage technology can help to regulate these fluctuations. Excess power can be stored and released when demand increases, thereby maintaining a stable electricity supply.
2. Peak shaving: by utilising power storage systems to support power supply during times of peak demand, power supply peaks can be cut and over-consumption of electricity reduced.
3. Back-up power: the use of power storage technology to provide backup power in emergencies can increase the stability and reliability of the power supply.

Thus, smart grids and power storage technologies both contribute to the evolution of energy systems and play an important role in promoting the diffusion of renewable energy and the sustainability of energy supply.

GNN and the smart grid

Graph Neural Networks (GNN), described in Graph Neural Networks, are machine learning models for processing graph-structured data. It is designed to handle unstructured data consisting of nodes (points) and edges (lines) (e.g. social networks, molecular structures, knowledge graphs, etc.), whereas traditional neural networks mainly deal with regular data structures such as matrices and images, and is a model with strengths in learning complex network The model has strengths in learning structures.

Smart grids are complex network systems with multiple interrelated elements such as energy supply, demand, renewable energy and electricity consumer behaviour, and the GNN is an ideal model for these due to the following advantages.

1. modelling network structures: a smart grid is physically a network of many interconnected power plants, substations, consumers, energy storage systems, etc. A GNN can naturally model such a graph structure and learn the states and interactions of each node.

2. integration of local and global information: GNNs can simultaneously learn the local information of each node and the structure and behaviour patterns of the entire network, making them very effective in optimising energy flows and fault detection.

3. adapting to dynamic environments: smart grids need to respond to fluctuations in energy supply due to fluctuations in demand and supply, weather, etc. GNNs can learn and adapt to these dynamic changes in real time, maximising the efficiency of the overall system.

Possible applications of smart grids using GNNs include.

1. electricity demand forecasting and load management: electricity demand forecasting is important in smart grids, and traditional statistical models have difficulty in accurately reflecting seasonal variations and supply-side uncertainties due to the introduction of renewable energy sources GNNs treat the electricity network as a graph and take into account relationships among consumers and demand patterns can be taken into account to forecast demand, e.g. by representing the energy consumption patterns of different regions as nodes in a graph and using GNNs to learn the interdependencies and impacts, future electricity demand can be forecasted with greater accuracy.

2. optimising power distribution: in smart grids, optimising power distribution is essential to balance supply and demand. As electricity networks are complex and prone to unstable generation, especially when renewable energy sources (e.g. wind and solar) are heavily used, real-time optimisation is important, but GNNs can be used to understand the overall structure of the electricity network and reflect the state of each node (power plants, consumers, energy storage systems etc.), it is possible to build algorithms that reflect the state of each node (power plant, consumer, energy storage system, etc.) to achieve optimal power distribution. This prevents overloads and blackouts and ensures efficient energy distribution.

3. fault detection and system recovery: fault detection and rapid recovery are important in power grids, and GNNs can be used to monitor the state of the power network and identify faulty nodes and edges (transmission lines), so that, in the event of a fault, measures can be taken quickly to minimise the impact on the entire system. GNNs not only detect local anomalies, but also learn the impact on the overall network, enabling them to predict how a fault will spread to other parts of the network and guide the appropriate recovery process.

4. integration of renewable energy: renewable energy sources such as wind and solar power are unstable in terms of energy supply, as their generation depends on weather and time of day, and to stabilise this, they need to be coordinated with energy storage systems and other sources of generation Using GNNs, power plants, renewable 6. the complex relationships between power plants, renewable energy sources, energy storage systems and consumers can be modelled to manage fluctuations in energy supply, thereby maximising the use of renewable energy sources while maintaining the stability of the electricity supply.

5. optimising energy trading: peer-to-peer energy markets are becoming increasingly common in smart grids, where consumers can trade energy with each other. In such a market, where consumers buy and sell energy generated by themselves, energy trading needs to be optimised: using GNNs, energy trading networks can be modelled with each consumer as a node to improve the efficiency of transactions. This enables the balance between supply and demand to be adjusted in real time, and optimal pricing and trading to be achieved.

6. energy flow monitoring and anomaly detection: energy flows in the entire power network can be monitored and anomalous flows can be detected, and GNNs can be used to learn patterns of power flows, enabling early detection and prevention of anomalies when unusual flows are detected This will enable early detection and prevention of abnormalities when unusual flows are detected. This technology is expected to contribute to fault prevention and efficient energy management, especially in large-scale power systems.

GNN-based approaches to smart grids are a promising method for efficiently modelling and optimising the complex interconnections of electricity networks. They provide powerful solutions to the diverse challenges of smart grids, such as energy supply efficiency, renewable energy integration, demand forecasting and fault detection, enabling the creation of sustainable and efficient energy systems.

implementation example

The following are examples of GNN-based implementations of smart grid optimisation.

Electricity supply path optimisation: Optimise the paths that efficiently supply power between nodes of the electricity grid (e.g. power plants, substations, consumers, etc.) to minimise losses.

Definition of a graph: a representation of the smart grid as a graph.

• Nodes: power plants, substations, homes, businesses, etc. Features (e.g. power consumption, generation, storage, etc.)
• Edges: connection paths in the electricity network. Characteristics (e.g. transmission capacity, resistance, loss rate, etc.).

import networkx as nx
import torch
from torch_geometric.data import Data

# Sample graph construction.
graph = nx.Graph()
graph.add_nodes_from([
    (0, {"type": "plant", "generation": 50}),
    (1, {"type": "substation", "capacity": 30}),
    (2, {"type": "consumer", "demand": 20}),
    (3, {"type": "consumer", "demand": 15}),
])
graph.add_edges_from([
    (0, 1, {"capacity": 100, "loss": 0.1}),
    (1, 2, {"capacity": 50, "loss": 0.2}),
    (1, 3, {"capacity": 50, "loss": 0.15}),
])

# Converted to data format for GNN.
edge_index = torch.tensor(list(graph.edges)).t().contiguous()
edge_attr = torch.tensor([[d["capacity"], d["loss"]] for _, _, d in graph.edges(data=True)])
node_features = torch.tensor([
    [n.get("generation", 0), n.get("demand", 0)]
    for _, n in graph.nodes(data=True)
])
data = Data(x=node_features, edge_index=edge_index, edge_attr=edge_attr)

2. model definition: learning relationships between nodes using GCN (Graph Convolutional Network).

from torch_geometric.nn import GCNConv

class SmartGridGNN(torch.nn.Module):
    def __init__(self):
        super(SmartGridGNN, self).__init__()
        self.conv1 = GCNConv(2, 16)  # Input features: 2 (power generation/consumption)
        self.conv2 = GCNConv(16, 8)  # Intermediate dimension: 16
        self.fc = torch.nn.Linear(8, 1)  # Output dimension: 1 (e.g. route efficiency score)

    def forward(self, data):
        x, edge_index = data.x, data.edge_index
        x = self.conv1(x, edge_index)
        x = torch.relu(x)
        x = self.conv2(x, edge_index)
        x = torch.relu(x)
        return self.fc(x)

# model initialization
model = SmartGridGNN()

3. training process:

• Loss function: minimisation of power losses (e.g. MSE).
• Optimisation target: efficient power distribution from supply nodes to demand nodes.

import torch.optim as optim

# Loss functions and optimisation methods
criterion = torch.nn.MSELoss()
optimizer = optim.Adam(model.parameters(), lr=0.01)

# Training data (dummy example)
labels = torch.tensor([[1.0], [0.8], [0.6], [0.7]])  # Ideal efficiency score for each node

# learning loop
for epoch in range(50):
    model.train()
    optimizer.zero_grad()
    out = model(data)
    loss = criterion(out, labels)
    loss.backward()
    optimizer.step()
    print(f"Epoch {epoch+1}, Loss: {loss.item()}")

4. application: path optimisation:
After training, the trained GNN model can be used to

• Optimise electricity demand and supply at each node.
• Identification of inefficient edges (high-loss pathways) and suggestion of diversions.
• Efficient operation of the entire smart grid.

5. expansion directions

• Demand forecasting: incorporate time-series data (e.g. weather and time of day) and add LSTM and time-series GNN.
• Fault detection: integrate anomaly detection models to detect anomalies at nodes and edges.
• Real-time optimisation: combine reinforcement learning (e.g. DQN and PPO) with GNNs for real-time control.

reference book

Reference books on graph neural networks (GNNs) and smart grid optimisation are summarised below.

Theory and implementation of GNNs.
1. “Graph Representation Learning” by William L. Hamilton
– Systematically describes the basic concepts, algorithms (GCN, GAT, GraphSAGE, etc.) and applications of GNNs.

2. “Deep Learning on Graphs” by Yao Ma and Jiliang Tang
– Provides an overview of deep learning methods that utilise graph structures.
– Includes a mathematical description of GNNs and Python code examples.

3. “Machine Learning with Graphs” by Johan Bollen et al.
– Fundamentals and practical methods for graph mining and network analysis, including GNNs.
– Examples from smart grids and social networks will also be discussed.

4. “Graph Convolutional Neural Networks and Beyond: Exploring Advanced Techniques in Applied Machine Learning”
– Brief description of the basic structure of GNNs and their applications, with a focus on GCNs.

Smart Grid related.
1. “Smart Grids: Advanced Technologies and Solutions” by Stuart Borlase
– Comprehensive description of the overall picture, components and advanced technologies of smart grids.
– Few specific examples of GNN implementations, but useful for understanding the background to network design and optimisation.

2. “AI-Powered Smart Grid: Revolutionizing Electricity Distribution and Generation”
– Details AI applications in the smart grid and energy sectors.
– In particular, examples of the use of AI technologies, including GNNs, are described.

3. “Green Machine Learning and Big Data for Smart Grids: Practices and Applications”
– This book focuses on AI in energy systems, particularly optimisation and forecasting techniques.
– Although there are few specific applications of GNNs, the book teaches about the integration of AI technologies.

Papers and online resources
– “Semi-Supervised Classification with Graph Convolutional Networks” (GCN paper).
Thomas Kipf, Max Welling

– “Graph Attention Networks” (GAT paper)
Petar Veličković et al.