Research on a Unified Data Model for Power Grids and Communication Networks Based on Graph Databases

Abstract

With the continuous development of the power grid, its structure is becoming increasingly complex. The occurrence of faults in transmission lines may lead to cascading failures in the power grid, ultimately resulting in widespread power outages. The transmission of equipment information and the sending of fault reports in the power grid rely on the power communication network. This network is crucial for ensuring the safe, stable, and economical operation of the power grid. As the number of devices in the power grid increases and sensor technology becomes more widespread, the volume of data generated by both the power grid and the power communication network has increased sharply. However, relational databases have limited scalability and struggle to meet the growing volume of data and user demands. This paper proposes a graph mapping method based on the power grid and communication network, utilizing data from both networks to construct a unified data plane in a graph database. Taking power transfer operations as an example, a unified standard data model and monitoring indicator system are established for both networks, enabling faster response and power restoration to blackout areas in the event of power grid faults. Simulation results demonstrate that compared to traditional relational databases, graph databases exhibit significantly improved efficiency in handling large-scale, highly connected data, making them more suitable for future power grids.

1. Introduction

With the continuous development of the power grid, ensuring a stable power supply has become a matter of great concern. The power communication network plays a crucial role in ensuring the safety, stability, and economic operation of the power grid. It is an integral part of the power grid, facilitating the transmission of information from primary equipment and reporting faults. The diagnosis, handling of faults, and power transfer operations in the power grid also rely on the power communication network [1]. The deployment of power grid equipment and the power communication network are closely interlinked, with power grid equipment and associated communication devices typically being deployed at the same location. The power communication network swiftly transmits business data, such as power consumption information and distribution automation data, from power grid equipment to the control center. Based on the real-time data transmission and continuous monitoring of power grid operations facilitated by the power communication network, network operation personnel can efficiently conduct power grid topology analysis, understand the structure and operational status of the power grid, and ensure its normal operation [2].

In recent years, with the explosive growth of data, advances in computing capabilities, and progress in AI and machine learning algorithms, big data technology has experienced unprecedented development. Big data technology, the Internet of Things, and cloud computing have jointly contributed to power grid regulation. Big data development has made it more diverse and complex, making it very difficult for relational databases to handle data [3]. Currently, relational databases are mainly used to represent power grid topologies, whereas relational databases use tabular structures to store data, with each table having a fixed schema. Relational databases have very low query efficiency when dealing with large amounts of data and numerous connections, prompting people to turn to non-relational databases [4]. Graph databases use graph structures to store data, and they perform well in handling large volumes of complex relational data, making it easy to represent and query complex relationships. Additionally, since graph databases do not need to adhere to the strict schema, they are more flexible in scenarios where data models need frequent modifications [5].

Based on the operational data of the power grid and communication network, a graph mapping method based on the power communication network and power grid can be constructed to map the data of the two networks to a unified data plane, establishing a unified standard data model and monitoring index system for both networks. Taking power transfer as an example, network operators can grasp real-time information about the operation of the power grid through the power communication network. In the event of equipment or line failures, the communication network operation analysis module can quickly analyze the propagation range of the fault. Through remote monitoring and control of power grid devices via the power communication network, fault resolution solutions can be promptly generated, and power transfer operations can be completed.

In recent years, many experts and scholars have conducted numerous research schemes for power grid planning combined with graph databases to improve the efficiency of power grid dispatch and ensure stable power supply. Reference [6] constructed some power grid data models based on domain modeling theory and proposed methods to transform these models, greatly enhancing the efficiency of data retrieval. However, the paper did not elaborate on combining different data models or address conflicts and contradictions between various models. Reference [7] combined power topology with graph databases and proposed a method for power outage area analysis based on graph databases. Reference [8] designed a common information model-oriented graph database (CIMGDB), which can enhance the efficiency of the power system. However, The complex model structure results in a large amount of data. Reference [9] used the CIM/E model to reduce data volume and speed up the response time of the data storage system, where the network topology processing algorithm was implemented using the messaging method. It was based on the TigerGraph graph database and was applied to process the topology information of substations. Reference [10] built a graph database-based power grid topology analysis application framework, addressing issues in data modeling, storage, and processing in power grid asset data management systems represented by power grid topology analysis systems. Reference [11] combined a breadth-first search algorithm with the graph database algorithm to study a method for calculating fault range, improving the efficiency of analyzing power grid faults. Reference [12] proposed a topology analysis tool, TAnalyzer, based on the graph database Neo4j, which could better perform tasks such as power grid topology analysis and fault range detection. The drawbacks were the insufficient accuracy of topology analysis and the need for ample storage space. Reference [13] proposed a fault propagation algorithm based on graph databases and predicate calculus, effectively improving the efficiency of the power grid. Reference [14] completed the solution of power outage range and the design of power transfer schemes based on graph data modeling. However, the paper lacked corresponding experimental data to verify the feasibility of the proposed solutions.

This paper combines graph theory algorithms such as Dijkstra, depth-first search, and breadth-first search, proposing a large-scale power grid fault propagation range calculation and optimal power transfer path planning scheme based on the Neo4j graph database, with distance as the weighting factor. Test results indicate that, compared to traditional schemes based on relational databases, the proposed scheme in this paper achieves faster planning of power transfer paths, thereby reducing losses caused by faults.

2. Basic Framework of Power Transfer Scheme

As shown in Figure 1, formulating a power transfer scheme in the graph database generally involves six main steps: collecting data, creating the graph model, determining the fault propagation range, identifying the optimal power transfer path, performing switch operations based on the selected path, and validating the generated power transfer scheme. Specifically, it begins with collecting data on the equipment and lines in the distribution network. Subsequently, we can create the graph database model based on the collected data, representing nodes and connections in the graphical format. The fault propagation range can be determined by combining the relevant graph algorithms with the fault location. Then, the optimal power transfer scheme is generated using shortest path algorithms and specifying the areas requiring power supply. Finally, the generated power transfer path performs corresponding switch operations, and the scheme’s feasibility is validated.

3. Relational Database and Neo4j Graph Database

In the development of database technology, relational and graph models are commonly used data models. The relational model is established on a rigorous mathematical foundation, exhibiting high data independence and security with simple usage. The relational database is a structured database that employs tables to store data, and its data query language is SQL. After decades of development, relational database technology has become relatively mature, but its strict schema constraints make it challenging to extend existing databases [15].

Relational databases organize data through tables. When representing entity relationships, it is necessary to repetitively store information in different columns of the tables, leading to extremely low efficiency in recursive queries and multi-table join queries. In complex data models, relational databases require multiple tables and table connections to represent relationships between entities. Multiple join operations result in decreased query performance and increased query complexity [16].

Graph databases use the graph structure to store data, represented by nodes, edges, and attributes. In contrast to relational databases, graph databases can flexibly store relationships between entities directly [17]. In the simple graph model shown in Figure 2, nodes represent entities, such as power sources or electrical equipment. Edges represent relationships between entities, such as power supply or electricity usage. Attributes denote information about nodes and edges, such as voltage, power, etc.

The power circuit of a simple dual-machine system is shown in Figure 3, with a busbar, a circuit breaker, switches, and loads in the lines. The relational model of this dual-machine system is illustrated in

Table 1. Table of busbars.

ID	Label	Name
001	M	M1

,

Table 2. Table of circuit breakers.

ID	Label	Name
001	CB	CB1

,

Table 3. Table of switches.

ID	Label	Name
001	S	S1
002	S	S2

,

Table 4. Table of loads.

ID	Label	Name
001	R	R1
002	R	R2

and

Table 5. Table of connection relationships.

ID	Name1	Name2
001	M1	CB1
002	CB1	M1
003	CB1	S1
004	S1	CB1
005	CB1	S2
006	S2	CB1
007	S1	S2
008	S2	S1
009	S1	R1
010	R1	S1
011	S2	R2
012	R2	S2

. Different types of equipment are stored in their respective component tables (see Table 1, Table 2, Table 3 and Table 4), and the connection relationships between devices are stored in the connection table (see Table 5). Information retrieval involves operations between several tables, thus limiting speed.

The graph model of the dual-machine system is depicted in Figure 4. The graph database stores the system’s topology information in the form of nodes and edges, facilitating a visual representation of the connection relationships between components.

When querying in graph databases, utilizing the relationships represented by edges allows for direct retrieval of entity records, eliminating the performance overhead caused by multiple join operations in relational databases and significantly enhancing query efficiency. Compared to relational databases, graph databases have several advantages:

(1)

The natural expression of the relational model. In graph databases, it is possible to define nodes and edges of any type to represent entities and relationships. This model can intuitively and naturally reflect the relational model in the real world, making data modeling more flexible and intuitive.

(2)

The flexible data model. The model of graph databases is more flexible than relational databases, allowing for the dynamic addition and removal of nodes, edges, and attributes. This flexibility enables better adaptation to changes in the data model.

(3)

Efficient relationship queries. By directly navigating the graph structure, graph databases can efficiently execute relationship and path queries without loading irrelevant, redundant data.

(4)

Node parallel computation. Unlike relational databases that only support simple statistical analysis and calculation functions, graph databases support node parallel computation, enabling the processing of more complex computing tasks. Some graph databases also support user-defined computation function libraries.

(5)

Ease of understanding and visualization. The data model and query language of graph databases make data more understandable. Through intuitive graphical representations, users can better understand relationships between data.

Neo4j is a widely used high-performance graph database known for its excellent performance in scenarios with complex entity relationships and strong data correlations. Cypher is the query language of Neo4j, providing a flexible and intuitive syntax that allows users to perform queries between nodes and edges efficiently. Additionally, Neo4j integrates various algorithms such as Dijkstra, LPA, etc., which assist users in complex graph analysis and processing [18]. Neo4j has found extensive applications in various fields, including social network analysis, network security, knowledge graphs, and power enterprise information systems.

In summary, Neo4j is a robust graph database management system that excels at tasks like fault range analysis and rapid generation of the power transfer scheme.

4. Construction of Power Grid Model

As shown in

Table 6. Table of nodes.

Number	Attribute	Note
1	ID	Identification of the node
2	Label	Type of the node
3	Name	Name of the node
4	Color	Color of the node

, each node has four attributes in the established graph model: ID, Label, Name, and Color. The label is used to distinguish different types of nodes. For instance, P and D can represent power sources and electrical equipment, respectively. The color allows us to visually identify the type of each node within the graph database.

Table 7. Table of relationships.

Number	Attribute	Note
1	ID	Identification of the relationship
2	Label A	Type of the start node
3	Name A	Name of the start node
4	Name a	Relationship between the start node and the end node
5	Label B	Type of the end node
6	Name B	Name of the end node
7	Distance	Distance between the start node and the end node

is a relational table containing seven attributes: ID, Label A, Name A, Name a, Label B, Name B, and Distance. For example, in Figure 2, Node_A and Node_B are the start and end nodes of the relationship Edge_a, respectively. Both nodes have label and name attributes, while the relationship Edge_a has ID, Name, and Distance attributes.

When modeling in the relational database, data are ultimately stored in tabular form, making it challenging to represent relationships between nodes intuitively. Retrieving connection relationships in such databases often requires a significant amount of time. In contrast, data are transformed into a graphical format when modeling in a graph database, allowing node relationships to be intuitively visualized. The graph database can be viewed as a collection of nodes and edges, and the inherent network structure of the power grid makes it well-suited for storing data in a graph format [19].

5. Power Transfer Scheme Based on Graph Database

CIM/E language is named after CIM and the prefixes ‘easy’, ‘efficiency’, and ‘electric-power’. It is simple, efficient, and well-suited for describing and exchanging large-scale power grid models  [20]. The data model of the power system adopts the CIM/E standard and is described in the form of relational tables, which represent various objects within the system. Equipment and connection relationships are stored in separate tables, namely the equipment and relationship tables. In the traditional relational model, equipment and relationships can be equivalently represented as nodes and edges in a graph model [21].

5.1. Graph Database Modeling

Figure 5 depicts the partial representation of the lines by modeling a regional line diagram in the graph database. Each node has four attributes: ID, label, name, and color. Different types of equipment have distinct labels and colors. For instance, the label for power nodes in Figure 5 is P, the color is green, and the name is P01. In the diagram models in Figure 5 and the following figures, we use the lowercase letters along with relationship arrow to represent different connections. Specifically, the connection relation for circuit breakers is uniformly designated as ‘y’; the connection nodes for equipment areas are assigned ‘a’; ‘z’ signifies the relation for disconnecting switches; while other lowercase letters, such as ‘b’ and ‘d’, denote the relationship names for line nodes.

5.2. Scheme to Determine the Fault Propagation Range

The data volume in the distribution network is increasing, and the connection relationships have become intricate. To reduce the loss caused by equipment or line failure, it is essential to determine the fault propagation range [14]. As shown in Figure 6, the specific steps are as follows:

(1)

When a fault occurs in the distribution network, it is necessary to identify the specific equipment or line where the fault has occurred and locate the corresponding node in the established distribution network model.

(2)

Take this node as the starting node and determine the connected relationships associated with this starting node.

(3)

Using either the depth-first traversal or breadth-first traversal method, and building on the connected relationships determined in step (2), traverse the graph database to obtain a maximal connected subgraph related to the starting node. This subgraph represents the range of fault propagation for that node.

5.3. Planning Scheme for Optimal Power Transfer Path

When equipment or line faults result in power outages, it is necessary to rapidly generate a power transfer scheme to supply power to electrical equipment. As shown in Figure 7, the specific steps are as follows:

(1)

Traverse the graph database to identify all available power sources and select one as the start node. Simultaneously, find the node corresponding to the electrical equipment in the model and designate it as the target node.

(2)

Determine the factors considered in finding the shortest path, such as distance, path length, or optimal power flow.

(3)

Use an appropriate shortest path algorithm, such as Dijkstra’s algorithm, to calculate the shortest path from the start node to the target node.

(4)

Analyze the obtained power transfer path, verifying its feasibility and safety.

(5)

Replace the start node with other available power sources and repeat steps (2)–(4). Finally, compare the power transfer paths from different power sources to obtain the optimal power transfer path.

The steps of Dijkstra’s shortest path algorithm [22] are illustrated in Algorithm 1.

Algorithm 1 The Dijkstra shortest path algorithm

1: Initialization. Let set S contain only the start node s, and set U contain all other nodes except s. If node v in U is adjacent to s, the distance of v is represented as an arc $(s,v)$; if they are not adjacent, the distance of v is ∞. 2: Selection. Select node w with the shortest distance from U, add w to S, and simultaneously remove w from U. 3: Update. Because w was added to S, it can be $(s,w)+(w,v)<$$(s,v)$, then replace $(s,v)$ with the path $(s,w,v)$. 4: Repeat. Repeat steps 2 and 3 until U becomes an empty set.

The optimal power transfer path planning algorithm with distance as the weight is shown in Algorithm 2.

Algorithm 2 The power transfer path planning algorithm

1: Initialization. The graph G; the available power sources S; the target node T; $\mathrm{closest}\_\mathrm{source}=\mathrm{None}$; $\mathrm{shortest}\_\mathrm{distance}=\infty$; $\mathrm{final}\_\mathrm{path}=\left[\right]$ 2: for each s in S do 3:        $(\mathrm{current}\_\mathrm{dist},\mathrm{current}\_\mathrm{path})=\mathrm{Dijkstra}(G,s,T)$ 4:        $\mathrm{if}\mathrm{current}\_\mathrm{dist}<\mathrm{shortest}\_\mathrm{distance}$ 5:              $\mathrm{shortest}\_\mathrm{distance}=\mathrm{current}\_\mathrm{dist}$ 6:              $\mathrm{closest}\_\mathrm{source}=\mathit{s}$ 7:              $\mathrm{final}\_\mathrm{path}=\mathrm{current}\_\mathrm{path}$ 8: end for 9: return shortest_distance, final_path, closest_source

6. Application of Power Transfer Scheme

This section validates the feasibility of two schemes: fault propagation range calculation and optimal power transfer path planning, based on the topological model of a specific area’s distribution network using the Neo4j graph database [23]. The model comprises 265 nodes and 286 relationships, including 3 power sources, 7 circuit breakers, 56 types of electrical equipment, etc. During the initial operation, the equipment in the network is powered by node P03. Figure 8 illustrates a partial connectivity graph for power source P03, where each electrical equipment has only one connection relationship.

6.1. Determination of Fault Propagation Range

During the power grid operation, if an anomaly is detected in equipment D30, it is first necessary to disconnect all circuit breakers and switches in Figure 8, identify the nodes connected to D30, and then proceed with the fault diagnosis.

Starting from D30, the following Cypher query can obtain the maximal connected subgraph of D30, as shown in Figure 9. The graph represents the propagation range of the current fault, and the query executed in 50 ms.

•  MATCH (n:D {name:’D30’})-[:a*0..]-(m)-[:d*0..]-(k)-[:a*0..]-(w)
   RETURN DISTINCT w;

6.2. Power Transfer Path Planning

Due to the malfunction of equipment D30, power source P03 can no longer supply the entire power grid. In this case, if equipment D49 requires an urgent power supply, it is necessary to calculate the shortest paths from the other two power sources to D49 separately. Subsequently, the optimal power transfer path is selected from the two paths. Finally, the switches along the chosen path are closed to complete the power transfer.

Using power source P01 as the start node and equipment D49 as the target node, the shortest path between them can be obtained through the Dijkstra algorithm. The Cypher statement is shown below, and the query time for this operation is 132 ms.

•  MATCH (startNode:P {name:’P01’}),(endNode:d {name:’D49’})
   CALL apoc.algo.dijkstra (startNode, endNode, ’’, ’distance’) YIELD path,weight
   RETURN path,weight,length(path);

Figure 10 illustrates the shortest path from P01 to D49, with a path length of 17 and a total distance of 8377.

Changing the name of the start node in the above Cypher code to P02 yields the shortest path from P02 to D49, with a query time of 121 ms.

Figure 11 depicts the shortest path from P02 to D49, with a path length of 12 and the sum of distances equal to 6596.

Upon comparison, it is evident that power to equipment D49 should be performed by power source P02, whether from the path length or distance perspective.

In this experiment, for the sake of simplicity, we assigned different distances to each edge, and each edge’s distance was randomly generated. In practical applications, specific weights such as the optimal power flow can be set to edges between nodes based on the actual conditions of the power grid. These weights can then be considered during the optimal power transfer path planning.

6.3. Comparison between the Graph Database and Relational Database

In order to obtain more accurate conclusions, we conducted experiments separately on two different computers. The configurations and operating environments of the computers are shown in

Table 8. The configurations and operating environments.

Hardware/Software	Computer 1	Computer 2
CPU	Intel(R) Core(TM) i7-9750H CPU @ 2.60 GHz	Intel(R) Core(TM) i7-6700H CPU @ 2.60 GHz
RAM	16.0 GB	16.0 GB
The version of the relational database	MySQL Server version: 8.0.34
The version of the graph database	Neo4j Server version: 5.9.0

. Compared with graph databases, the results obtained from power transfer path planning in relational databases are the same, but there is a significant difference in query speed. The query times between two databases on different computers are shown in

Table 9. Comparison of query times between two databases on different computers.

Query Operation	Computer 1		Computer 2
	Relational Database	Graph Database	Relational Database	Graph Database
Maximal connected subgraph of D30	15 ms	50 ms	15 ms	61 ms
Power transfer path from P01 to D49	531 ms	132 ms	7406 ms	175 ms
Power transfer path from P02 to D49	453 ms	121 ms	6532 ms	172 ms

.

When querying the maximal connected subgraph of D30, due to the small data volume, queries are relatively simple and involve fewer nodes and relationships, resulting in better performance of relational databases. However, when querying the power transfer paths, which involve a significantly larger number of nodes and relationships, the performance of graph databases is much better than that of relational databases. Moreover, a decrease in computer performance will rapidly increase query time for relational databases, but the impact on graph databases is relatively minor.

Additionally, defining the sum of nodes and relationships as complexity, we conducted experiments on networks of different scales. The query time curves for the two computers are shown in Figure 12 and Figure 13. Obviously, when the complexity is low, both relational databases and graph databases perform well. As the complexity increases, the query time of relational databases rapidly increases, while graph databases exhibit better robustness, showing no significant performance degradation. At a complexity of 1000, the performance gap between graph databases and relational databases becomes very pronounced. The experiments demonstrate that graph databases can perform efficiently with increasing data volume and complexity, making them more suitable for large-scale and highly connected data storage and query requirements.

7. Conclusions

This paper uses the graph database to explore the calculation of the fault propagation range and the planning of power transfer paths in distribution networks. This paper proposes corresponding schemes and conducts experimental verification in the graph database. The proposed schemes successfully address the problems and visually represent the experimental results. Additionally, the experimental results in the graph database are compared with those in the relational database. The comparison reveals that the graph database performs less efficiently than the relational database when calculating the fault propagation range with a relatively small dataset. However, in planning power transfer paths, where many nodes and edges are involved, the relational database experiences a significant increase in processing time. The model established in this paper only consists of a few hundred nodes and relationships. In practical applications, the number of devices will grow exponentially, and the efficiency of graph databases will far exceed that of relational databases.

The experimental results indicate that the power transfer scheme based on the graph database can rapidly calculate the fault propagation range and promptly complete the planning of power transfer paths in large-scale and complexly interconnected distribution networks.