Data-driven smart grid carbon emission control methods using graph-based power flow computing

Abstract

In order to solve the problem of large-scale power grid, complex connection relationship, and large demand for electricity carbon factor calculation resources, this paper proposes an algorithm that is easy to model the power grid topology and has high computational efficiency to implement the application of “One Electricity Carbon Chart”. In this paper, the graph computing component is used to realize the graph modelling, distributed storage, and high-performance calculation of the electric carbon factor for the data representing the topology of the power grid, such as the power grid table, AC line segment table, and AC line end table. This paper realizes the rapid grid topology graph modelling of EMS dispatching system data, so that the grid structure and active power flow can be intuitively displayed and dynamically studied. This paper also realizes the distributed parallel calculation of electric carbon factor in the case of a ring network, which solves the problem of difficult analysis and calculation of large-scale power grid topology. We conduct experiments on real datasets, and the experimental results demonstrate the effectiveness of the proposed method. In addition, the research on the “One Electricity Carbon Chart” makes the carbon emission of power plants depend on the power flow of the power grid and transfers the accounting method, realizing the calculation of the carbon emission factor of the power grid at the plant level, and provides a new perspective for the sharing of carbon emission responsibilities in the power system.

Keywords

Electric carbon emission control data-driven method grid topology

1. Introduction

With the change of the domestic ecological environment [1, 2], China will adopt more robust policies and measures to peak CO2 emissions [3] by 2030 and strive to achieve carbon neutrality [4, 5] by 2060. It has been more than ten years since the last release of the carbon emission factor for power grids [6]. The factor is a statistical type factor with low resolution for unified national accounting. Due to the untimely release and low resolution of this factor, it is not conducive to the formulation of carbon peaking and carbon-neutral planning by provinces, nor is it conducive to the compliance of emission control enterprises and their participation in carbon market transactions. In the process of realizing “One Electricity Carbon Chart”, it is necessary to model the topology of China’s entire power grid [7], and the power generation data [8] and active tide data of power plants need to be used to calculate the carbon emission factors [9] of the power grid at high frequency.

In the context of “double carbon” [10] and a new power system [11], a large number of new energy access need an indicator to measure the impact of new energy access. There are a series of problems with the electric carbon factor. First of all, the release is not timely. For example, the average carbon emission factor of the grid has not been updated since 2012 and is no longer current. This is not conducive to the formulation of carbon peak and carbon neutral planning by provinces (municipalities), nor is it conducive to carbon quota compliance and participation in carbon market trading by emission control enterprises. Secondly, the resolution is lower and released after using the year as the time unit. It is impossible to distinguish the electricity consumption characteristics of different industries and enterprises, such as seasonal characteristics, time-sharing characteristics, regional aggregation characteristics, etc. In fact, these distinctions will have a significant impact on the carbon emission measurement of enterprises in each industry. The calculation cycle is annual, using statistical data after the fact, there is a severe lag. In addition, the regional span is too large, taking the large region as the calculation unit, which is not conducive to playing the central role of the administrative regions of provinces, prefectures (cities) and counties (districts) for the development of new energy. The characteristics of new energy generation in different regions, seasons, and time periods differ greatly, and their spatial and temporal distribution characteristics directly affect the carbon emission factor of the power grid. Finally, the value is not sufficiently played. The value of the average carbon emission factor of the traditional power grid is relatively fixed, which cannot influence the power consumption behavior of enterprises and their trading behavior in the power market and carbon market, so it is difficult to promote the “electricity-carbon” linkage effectively, and it is also difficult to promote the organic integration and coordinated development of carbon trading market [12, 13] and power trading market (including green power trading market) [14, 15].

At present, the accounting methods of electric carbon factors in China’s power system on the enterprise side and regional side mostly adopt the material balance method and emission factor method, among which emission factors mainly contain carbon emission factors of power generation enterprises, carbon emission factors of power grid enterprises and average emission factors of power consumption in three categories. The average emission factors of regional power grids [16] released by the National Climate Center from 2010 to 2012 are more widely used. The essence of electricity carbon factor calculation is to connect the generation, transmission, transformation, distribution, and consumption links to realize the integration of electricity data. Bridging the carbon emission source (direct emission), flow path (trajectory), and end-user carbon emission (indirect emission) is based on graph computation and knowledge mapping technology, which is easy to maintain, supports node parallel computation, fast computation speed, and good real-time performance. In the traditional carbon flow transfer calculation method, the matrix is constructed for the grid carbon flow transfer, and the matrix inversion is involved in the carbon intensity calculation. Considering the large scale of the grid, the complex connection relationship, and the large demand for the electric carbon factor calculation resources, the matrix dimensionality constructed by the traditional method is high and involves the inversion of the high-dimensional matrix, which requires high computational arithmetic power, so there is an urgent need for an easy grid topology modeling and high computational efficiency. Therefore, there is an urgent need for an algorithm that is easy to model grid topology and computationally efficient.

For the existing problems, this paper proposes a fast grid topology modeling of EMS dispatch system data, and solves the problem that large-scale grid topology is difficult to analyze and compute. In addition, “One Electricity Carbon Chart” is a panoramic electricity-carbon coupling model based on the power supply side, grid side and customer side electricity carbon emission and related data, with high-performance graph computing support for parallel computing and iterative solution, and built by graph computing and knowledge mapping technology, which is a cross-provincial, cross-professional and cross-information system. The model uses power tide analysis as a means to achieve carbon flow tracking, realize accurate and dynamic quantification of power carbon emission factors by time and partition, support the government to establish a power carbon emission accounting system, help enterprises to double control the transformation of energy consumption, and promote the development of trading market. Finally, we have also conducted experiments and applications in real scenarios.

In summary, our main contributions can be summarized as follows:

•
This paper proposes a fast grid topology graph modeling method for EMS dispatching system data, which solves the problems of topology analysis and calculation difficulties in large-scale power grids.
•
This paper completes the development of the data analysis process of “One Electricity Carbon Chart”, realizes the service process of automatic execution of analysis and calculation, regularly executes task programs, automatically downloads and analyzes business data, automatically reads the script of analysis and calculation data to the database, and automatically executes and generates reports data.
•
We conduct extensive experiments on the real-world dataset and the result strongly demonstrates the superior performance of our proposed approach.

In the rest of this paper, we introduce the related work in Section 2. In Section 3, we introduce our method in detail from grid modeling based on high-performance graph computing and electricity carbon trajectory analysis, and carbon emission calculation model based on high-performance graph computing. We report the experimental results in Section 4, and we finally conclude the paper in Section 5.
2. Related work

Low-carbon has become the focus of academic and practical circles. At present, there are many macro-thinking about national low-carbon development policies and systems at home and abroad. This section mainly introduces the carbon emission calculation method.

2.1 Current status of domestic and international research

With the development of technology, people have realized the importance of reducing carbon emissions, and many countries have taken measures to limit and reduce carbon emissions. The most obvious measures include using renewable energy and improving energy efficiency. In addition, some countries have implemented market mechanisms such as carbon trading systems. Currently, research on carbon emissions is mainly focused on three areas: Firstly, controlling carbon emissions at the source. Policies, technologies and other means are used to control carbon emissions in fields like energy, transportation, and industry, for example, promoting clean energy, restricting high-polluting vehicles from entering urban areas, and implementing carbon trading. Secondly, monitoring and compiling carbon emission inventories. A comprehensive carbon emission monitoring system is established, and carbon emission inventories are compiled for different industries and regions to better identify, track, and manage carbon emissions. Thirdly, carbon capture and storage technologies. Carbon capture technologies can capture and store carbon from the atmosphere, such as underground gas storage and rock sequestration. Developing these technologies can effectively reduce carbon emissions and achieve carbon reduction goals. In summary, the research directions on carbon emissions are diverse, but global cooperation is still needed in policy, technology, and market-based measures to achieve global carbon reduction goals. In addition, “IPCC Fourth Assessment Report: Climate Change 2007 (AR4)” written by IPCC (United Nations Intergovernmental Panel on Climate Change) is the most valuable climate change report after the three climate change assessment reports in 1990, 1995 and 2001 [17]. The report has become a standard reference work and is widely cited by policy makers, scientists, other experts and students. “2006 IPCC Guidelines for National Greenhouse Gas Inventories” is an update and improvement of “1996 IPCC Guidelines for National Greenhouse Gas Inventories” [18]. The guidelines provide general guidance on the calculation methods of greenhouse gas emissions, including data collection, uncertainty analysis, method selection, and calculation quality assurance. The “Specification for Assessment of the Life Cycle Greenhouse Gas Emission of Goods and Services” (PAS2050 for short) issued by the British Standards Institution is a specification for the evaluation of greenhouse gas emissions in the life cycle of goods and services [19]. The concept definition of carbon emissions from processing, transportation to consumption provides a theoretical basis for the calculation of product carbon emissions. The “GHG Protocol” was established by experts and scholars from the World Resources Institute (WRI) and the World Business Council for Sustainable Development (WBCSD) [20].

Graph computing [35, 36] is a data processing technology mainly used for analyzing and processing large-scale and complex data graph structures. In terms of carbon emissions, graph computing can be applied in the following areas: Firstly, controlling carbon emission sources. By using relevant data from industries such as energy, transportation, and industry to establish a graph structure, graph computing can be utilized to analyze and calculate the main sources of carbon emissions, and implement targeted measures to reduce emissions. Secondly, monitoring carbon emissions. Carbon emission data collected by sensors, meteorological stations, satellites, etc. are used to establish corresponding graph structures, which are then analyzed and calculated using graph computing to better monitor and manage carbon emissions in different regions, times, and industries. Thirdly, evaluating carbon emissions. Based on graph computing, carbon emission evaluation models can be established to simulate and evaluate the effects of different policies and strategies using real-time data, in order to determine the best emission reduction solutions. In summary, the application of graph computing in the field of carbon emissions has extensive value, helping us better understand and solve carbon emission problems, and promoting the achievement of global emission reduction targets. With the continuous increase and diversification of data volume, graph computing also needs to be continuously combined with other technologies to better play its role.

2.2 Current status of industry-specific applications

Xuena Wang summarized the current research status of carbon emission calculation methods for energy-based carbon sources, analyzed the system dynamics and its advancement based on the concept of system dynamics, and pointed out the necessity and feasibility of its application in the problem of carbon emission measurement of energy-based carbon sources [21]. Yungong Guo [22] and Wenying Chen [23] calculated energy and carbon emissions macroscopically by establishing models. There are currently three main models: ERM-AIM/China Energy Emissions Model, Logistic Model, and MARKAL Model. Druckman [24] analyzed the characteristics of using the input-output method to calculate carbon emissions from the perspective of consumption and production, and calculated the reduction of carbon emissions in the United States. Gang Wang [25] gave the calculation method of CO2 emission reduction caused by energy saving, and took the energy integration project of a terephthalic acid plant of a petrochemical company as an engineering example to calculate the CO2 emission reduction caused by energy integration. Zhihui Zhang [26] analyzed the calculation process of carbon emissions in the whole life cycle of buildings. The calculation results showed that energy integration has a significant effect on CO2 emission reduction. Based on the basic equation of carbon emissions, Guoquan Xu et al. [27] used the log-average weight Disvisia decomposition method to establish a factor decomposition model of per capita carbon emissions in China, and quantitatively analyzed the effects of changes in energy structure, energy efficiency and economic development on per capita carbon emissions in China during 1995–2004. An Zhang [28] used the continuous approximation method and the traffic turnover method to construct a carbon emission evaluation model under the influence of urban micro-land use, and based on this model, he evaluated the carbon emission caused by the land use pattern of a unit in Nanjing. Zhang et al. [29] established a bottom-up approach to calculate and map carbon emissions from urban building stock, using inventory data to derive a comprehensive reference model for calculating urban building carbon emissions and sinks, and using building life Periodic assessment models and architectural prototyping methods. Wang et al. [30] provided a new method for studying the impact of the epidemic on carbon emissions. Based on the autoregressive integrated moving average method and the backpropagation neural network method, they developed two combined ARIMA-BPNN and BPNN-ARIMA simulation methods to simulate Carbon emissions in China, India, the US and the EU. From the perspective of carbon emission efficiency, Sun et al. [31] applied stochastic frontier analysis to screen the factors affecting carbon intensity, and constructed a carbon emission intensity prediction model based on factor analysis and extreme learning machine. The results show a high correlation between carbon emission efficiency and carbon emission intensity.

3. Proposed method

The overall framework of the method in this paper is shown in Fig. 1, which includes two parts: power grid modeling based on high-performance graph computing and electricity carbon emission trajectory analysis, and carbon emission calculation model based on high-performance graph computing.

Figure 1.

The overall framework of the method.

3.1 Power grid modelling and electricity carbon trajectory analysis based on high-performance graph computing

Through grid analysis of grid tide data: grid analysis and measurement analysis, base data and initial data of carbon intensity iteration are generated, and carbon intensity iteration calculation method based on graph calculation is designed.

3.1.1 Power grid structure analysis

Grid analysis refers to the analysis of plant stations and lines, which needs to go through the steps of file analysis-data verification governance-file extraction-reading into the database, etc. Because the data of the network frame is fixed, it is usually necessary to re-execute it only when the network frame has been changed significantly. The script parses the network frame data file and extracts the required power network frame data according to the tags and fields in the matching file, which includes: A01 Data, A02 Data, A03 Data, A04 Data and etc. The specific matching and field information is shown in Table 1.

Table 1
Parsing matching data tables

Parsing data tables	Matching tags	Data table fields
A01	A01_Label	ID
		Name
		AliasName
		Type
		Record
		Voltage
		Area
A02	A02_Label	ID
		Name
		SubLine
		Voltage
		StartID
		EndID
		Type
A03	A03_Label	ID
		Name
A04	A04_Label	ID
		Name
		Number
		Record

Among them, the processing of the parsed fields in A01 and A02 is as follows. The A01 parsed data is mapped to the field station type according to the content in the CIM file, and the mapping logic of the Type is shown in Table 2.

Table 2

Station type mapping logic

Field station category	CIM file mapping content substation
D01	schemaSubstationType.trans
D02	schemaSubstationType.hydro
D03	schemaSubstationType.thermal
D04	schemaSubstationType.virtual
D05	schemaSubstationType.tjoint
D06	Empty value

The maximum basic voltage mapping logic is the actual voltage level obtained by associating the voltmeter with the ID. The mapping logic of the area to which it belongs is the actual control area obtained by associating the area table with AreaID. When analyzing data in A02, the mapping logic of the basic voltage level is the actual voltage level obtained by associating the voltmeter with the ID.

The resolution of the measurement refers to the resolution of the active power on the line, which needs to go through the steps of file resolution-data verification governance-file extraction-reading into the database-related analysis to generate reports, and other actions. Because the active power is generated every 5 minutes, the measurement analysis needs to be updated in real-time, and the update frequency is determined according to the specific situation. Taking a place P10 as an example, the measurement data will be updated every hour.

To extract the active power values from both ends of a line stored in a file, we parse the crime file. As the AC line segment tag does not cover the plant station’s lines, we use the measured active power value of the synchronous generator under these stations as the line measurement. As a result, we divide the line table into two sections: the AC line table and the synchronous generator table.

(1) For AC line measurement, we examine the data in the AC line table of the file. After reading it, we use this table as the primary table and perform a left join with an analysis table using the line name and starting plant name as the associated fields. Finally, we output the matched outcome to the line meter. However, there are some crucial points that require particular attention:

Firstly, since the AC line includes a line corresponding to the virtual plant station, we need to use the original plant station name (i.e., T wiring) for the end of the line and the virtual plant station name for the beginning of the line in the AC line table. After a successful match in the line measurement table, the virtual plant station corresponds to the beginning of the line, whereas the end of the line needs to use the previously specified plant station name, which is a certain line’s virtual station. Secondly, when analyzing the measurement data table of the same line, we consider both the starting and ending stations’ measurement values. To simplify the process, we assume that the line measurement value equals the input station’s measurement value, regardless of line loss. Thirdly, since the virtual station may exist as the line’s start station, and the parsed line measurement’s virtual station side is typically 0, we need to check whether the start station contains the word “virtual.” If it does, we associate the line name with the end station name. Fourthly, if some lines are not matched to active measurement or have a matched active value of 0, we assign a fixed value of 0.0001 to these lines to ensure accurate data when calculating carbon intensity.

(2) Synchronous generator measurement

We parse the data of the synchronous generator table in the file and read the table. Using the synchronous generator table as the primary table, we associate it with the parsing table using the starting station name. As the PV station typically has multiple generators supplying power, we need to sum up all the generators matched by the station’s active measurement name as the line’s measurement data. Finally, we output the matched results to the line measurement table.

3.1.2 Basic data verification governance

Power grids have a lot of T wiring in use, T wiring connected to various plant stations. To statistical this part of the line of electric carbon flow, we need to add virtual plant stations. Firstly, we associate the beginning and ending plant stations of each line with the plant station ID from the initial plant station table. We ensure that the end of the line plant station and the beginning of the plant station are for T wiring plant stations. Next, we group the lines by their main line name, and for each main line, we determine a virtual plant station. As shown in Fig. 2, if we take the L01 line as an example, and the analysis of the line table has three lines: L02, L02, and L03, whose main line is also L01. We can associate the three plant stations with a virtual station, abstracting them into a specific plant station named L04. Finally, we insert all virtual stations into the initial substation table. The coordinates of the virtual plant station are determined based on the two plant stations connected by the main line. The longitude is $A_{x}*3/5+A_{y}*2/5$ and latitude is $B_{x}*3/5+B_{y}*2/5$ , where $A_{x}$ is the longitude of the starting plant station, $A_{y}$ is the longitude of the ending plant station, $B_{x}$ is the latitude of the starting plant station, and $B_{y}$ is the latitude of the ending plant station.

Figure 2.

P10 line.

3.1.3 CIME file extraction

The EMS system creates a DT file with active power data every 5 minutes and saves it to a set directory. We can use SFTP get command to download the file to a local directory. For real-time large-screen data, we need to run the process every hour, extracting the 12 DT files created in the previous hour.

(1) Automatically extract CIME files

To extract data using SFTP, connection details such as password and directory are required. To automate the process, we encapsulate the execution command in a Java jar package and execute it with a command line. This extracts the 12 DT files generated in the previous hour to a fixed directory. Finally, we write the command to execute the jar package into an executable BAT file.

(2) Automatic timing tasks

Press Windows $+$ R to open the command line window, enter the command to open the task scheduler. Create a scheduled task with a new trigger set to execute daily, every hour with “Yes” selected for executing the BAT file. Don’t forget to check the repeat task interval.

(3) Automatic deletion of extracted data

For data security, delete the DT file from the directory after analysis and calculation. Use a bat file to execute the command for deleting the directory and write it into an executable BAT document.

3.1.4 Network frame and report correlation analysis

We read grid data, initial line table, analysis data and calculation data into the database through scripts, and generate large screen report. Because many large-screen reports need to display carbon flow data between regions, we combine line measurement and plant carbon intensity with grid data to build a regional carbon flow intermediate table to facilitate report generation. The main area report calculation logic is shown in Table 3.

Table 3
Regional report

Table name	Field name	Field description
B01	Id	Region ID
	Area	region name
	Line	Line name
	Type	Line type
	SId	Start plant station ID
	SSub	Start station name
	EId	Ending station ID
	ESub	End station name
	Val	Line direction
B02	Time	Current date
	Code	Area code
	Name	Region name
	Tqd	Regional carbon intensity
	Unit	Carbon intensity unit
B03	Time	Current date
	Code	Region code
	Name	Region name
	P	Electricity consumption
	GF	Photovoltaic power generation
	SD	Hydroelectric power generation
	Pout	Electricity transmission volume
	Unit	Electricity unit

(1) B01 table

We have divided the factories and stations into various regions before, and now we need to divide the lines into regions. In order to display the carbon flow data between regions, we need to identify the regional flow direction of the lines. The specific inflow and outflow of the lines are composed of two parts. One is to analyze the direction of the line, from the start station to the end station. The second is the positive and negative values of the active power measurement. A positive value indicates that the line flow direction is from the start station to the end station, and a negative value indicates that the line flow direction is from the end station to the start station. The Val field in the regional plant station table identifies the regional flow direction of the line, and a piece of data in the table indicates whether a line in a certain area flows in from other areas or flows out to other areas.

(2) B02 table

The regional carbon intensity needs to be calculated based on the interrelationship between the calculated carbon intensity table, the analytical measurement table, and the regional plant and station table. Regional carbon emissions are also divided into three parts. The first is the carbon emissions generated by thermal power plants, that is the power generation of thermal power plants * 0.87. The second is the carbon emissions of P10’s exported electricity, that is the exported electricity * 0.601, and the third is the carbon emissions generated by local new energy electricity.

(3) B03 table

According to the measurement data of the analytical line and the table of regional plants and stations, we can get the power consumption of each region through the association of the line name. The regional power consumption is divided into three parts. One is the power of the region itself, including power generation by thermal power plants power generation P0 with new energy power plants, the second is the power P1 transmitted from other regions or outside P10, and the third is the power P2 transmitted from this region to other regions, that is regional power $=$ P0 $+$ P1 $-$ P2. What we need to pay special attention to is that the storage in the analytical measurement table is the active power, which needs to be divided by twelve when converting it into electricity consumption. In addition, because the active power measured analytically has its own positive and negative signs, that is, the flow direction of electric power may be opposite to the direction of the line. Therefore, we need to do the product with the val value in the regional plant and station table to ensure that the power flow is correct. The calculation company of power consumption is U $=$ V * val/12, where U is the power consumption and V is the power consumption.

3.2 Carbon emissions calculation model based on high-performance graph computing

3.2.1 Graph computing basic data analysis

Determine the grid structure, and assign power to the line segment according to the real-time power collection data. The specific steps are: analyze the line segment corresponding to the grid table File One and File Two in the DT file, match according to different fields and add power respectively field, where File one matches according to the station name. Since there is a virtual station in File Two, the analysis is divided into two types. If the start station is a virtual station, the line segment name and the start station are used for matching. If the starting station is a non-virtual station, match the line segment name and the ending station. Since the DT file is generated every five minutes, 12 DT files need to be aggregated as the corresponding active power value per hour.

3.2.2 Research on the flow direction power supply of the power plant

Since the carbon intensity calculation data includes the power plant flowing to the power source, the electricity in some substations will flow to photovoltaics and power plants to offset the initial carbon flow of the power plant. The strength calculation node table and relationship table additionally add station nodes and station relationship line segments. The specific addition method is: if there is a line segment flowing from the substation to the power plant, a new node is added to the node table. The initial carbon intensity of this node is the initial carbon intensity of the corresponding power plant, and add a new node pointing to the relationship edge of the original power plant. The active power on the edge is the active power flowing from the original power plant to other stations. If there are multiple outflow edges, add them one by one.

3.2.3 Research on design of graph computing algorithms

The graph is an important data structure, which is composed of nodes V and edges E. We generally represent the graph as G (V, E). Typical examples of graph data include web page link relationships, social networks, product recommendations, etc. The most basic data structure in a graph computing system consists of three factors: vertex V, edge E, and weight D, that is G $=$ (V, E, D), where V is the vertex, E is the edge, and D is the weight. Now the popular graph neural network based on deep learning is a network used to process graph-type data, and the goal of this network is to learn the representation $h_{v}$ of each node v, and each node’s representation is calculated from the features of the node $x_{v}$ , the features of the edges connected to the node $x_{co[v]}$ , the neighbor representation of the node $h_{ne[v]}$ and the features of its neighbor nodes $x_{ne[v]}$ :

$\displaystyle h_{v}=f(x_{v},x_{co[v]},h_{ne[v]},x_{ne[v]})$ (1)

For tasks that focus on nodes, we can directly use the representation of $h_{v}$ to complete a specific task, and for tasks that focus on the entire graph, we can obtain a global representation information by pooling or other methods for all node representations and then do the corresponding Task.

In this paper, the graph calculation method is the same as the power flow matrix calculation method input. It is necessary to construct a plant-station node table and a carbon flow relationship table between plants and stations. Among them, the node table includes node ID, node name, and initial carbon intensity field, and the relationship table includes carbon flow outflow node ID (from_key), carbon flow inflow node ID (to_key), inter-node active power transfer value, read-in nodes, relational table, iterative calculation of carbon flow transfer based on spark graphx. The carbon intensity algorithm process is shown in Fig. 3, and the specific calculation logic is as follows:

Figure 3.

Carbon intensity algorithm flowchart.

Step 1: Read the node table and relationship table, which contain information about nodes and their relationships, and use this information to construct vertex and edge objects. The node table contains fields for node ID, node name, and initial carbon intensity. The relationship table contains fields such as carbon flow outflow node ID, carbon flow inflow node ID, and active power transfer value between nodes. Additionally, initialize the number of iterations n.

Step 2: Construct a graph using the vertex and edge objects created in Step 1. Vertex is the node of the graph, and edge is the edge of the graph.

Step 3: Check if the number of iterations is less than n. If it is, proceed to Step 4. If it’s not, go to Step 5.

Step 4: The process of updating the carbon flow transfer in the graph is described. This step involves applying an “aggregate messages” function to the messages received from all the from_key nodes pointing to the to_key node. The update formula is:

$\displaystyle CD_{th}=\frac{\mathop{\sum}\nolimits_{i=1}^{n}CD_{fk\_i\ast P_{i% }}}{\mathop{\sum}\nolimits_{i=1}^{n}P_{i}}$ (2)

Step 5: Store the results of carbon flow transfer that were calculated in Step 4.

where $CD_{tk}$ is the carbon intensity calculated by the to_key node, $CD_{fk\_i}$ is the carbon intensity of the ith from_key node, $n$ is the number of from_key nodes pointing to the to_key node, $P_{i}$ is the active power of the i-th node. Overall, this process can help model carbon flow transfer within a system represented by a graph, taking into account the different carbon intensities of various nodes and their active power levels.

4. Experiment

4.1 Calculation description

4.1.1 Grid situation in P10

The types of “One Electricity Carbon Chart” experimental sites include C01, C02, C03, C04 and C05. Plant statistics include 124 plants, classified by plant type, including 53 C01, 40 C05, 7 C03, 4 C04, 5 C02, and 15 virtual stations. Classified by voltage level, among the C01, there are 22 35 kV C01, 20 110 kV C01, and 11 220 kV C01. Divided by region, there are 114 P10 substations and 10 P10 outer Substations. Statistics on the initial carbon intensity of the plants and stations: 0.87 for thermal power plants, 0.601 for P10 Foreign Transmission Power Plant, and 0 for other plants and stations. The specific data are shown in Table 4.

Table 4
P10 grid situation

Area/plant type	C01			power plant			C02	Virtual station
	35 kV	110 kV	220 kV	C05	C03	C04
T01	0	1	0	1	0	0	0	4
T02	0	1	0	0	0	0	0	0
T03	7	4	1	5	2	0	0	3
T04	4	1	1	11	1	0	1	2
T05	1	1	0	3	1	0	0	0
T06	3	1	1	1	0	2	1	2
T07	0	2	0	2	1	0	1	0
T08	1	0	0	0	0	0	0	0
T09	3	2	0	0	0	1	1	0
T10	1	1	0	2	0	1	0	0
T11	0	1	0	1	1	0	0	1
T12	2	2	1	12	1	0	1	1
T13	0	1	0	2	0	0	0	1
T14	0	2	7	0	0	0	0	1
Summary	22	20	11	40	7	4	5	15

4.1.2 P10 line situation

There are 146 statistical lines, including 52 substation lines, 49 power plant lines, and 45 wiring lines. The line measurement is obtained from the line where the photovoltaic plant is located in the power plant line from the measurement values of each unit on the side of the synchronous generator, and the line measurement is the summary value of the unit measurement of the photovoltaic plant. The wiring line measurement is the measurement value of the real plant side of the line. The measurements of other lines are unified as the measurement values of the power generation side. According to the positive and negative of the active power measurement and the direction of the line, the power generation side and the discharge side of the line are judged.

4.2 Carbon emission control efficiency experiment

In this article, we compare the proposed method with three methods: GraphX [32], Neo4j [33], and JanusGraph [34]. GraphX is a distributed graph processing framework of Apache Spark, which allows writing graph algorithms in Scala, Java, and Python and running them efficiently on Apache Spark. Neo4j is a graph-based NoSQL database specifically designed for storing, managing, and querying large connected data. It uses Cypher language as its query language, supports ACID transactions and high scalability. JanusGraph is a distributed graph storage and computing system that uses an Apache TinkerPop compatible graph model to support multiple backends. It also provides API that supports the Gremlin query language, which can be used to execute complex graph operations and algorithms.

The logical association of the generated intermediate result table is correct, which meets the calculation requirements of the algorithm. The iterative calculation result of the graph calculation carbon intensity is consistent with the calculation result of the power flow matrix. It has been verified that the carbon intensity has reached convergence in about 30 generations, it takes about 20 ms to iterate once. We compare the computation iteration time of different methods and the experimental results are shown in Table 5. It can be seen from the table that our method is better than GraphX, Neo4j, and JanusGraph under different iteration numbers. In addition, when it tends to converge, the error of the carbon intensity results of the nodes located in the loop is less than 1e-10 compared with the calculation results of the power flow matrix. The result of the iterative calculation of carbon intensity in graph calculation is accurate, and the iterative efficiency meets the business requirements.

Table 5
Experimental result table of iterative time cost

Iterations	Our method	GraphX	Neo4j	JanusGraph
1	0.4 ms	0.5 ms	0.5 ms	0.6 ms
10	6.4 ms	7.1 ms	7.9 ms	8.3 ms
20	12.4 ms	13.6 ms	16.6 ms	17.2 ms
30	20.0 ms	23.5 ms	25.2 ms	27.1 ms
40	28.5 ms	32.7 ms	35.8 ms	36.1 ms
50	35.3 ms	41.4 ms	43.2 ms	44.2 ms
60	42.7 ms	49.3 ms	50.5 ms	53.6 ms

Figure 4.

The iterative coverage of carbon emission control estimation.

4.3 Carbon emission control coverage experiment

Then, to evaluate the coverage effectiveness of our proposed method, we conduct the experiment on the coverage grade in different iteration numbers and report the result in Fig. 4. We set the iteration stop threshold to 0.00001 and all results have experimented with the same setting. As we can see, our method constantly outperforms the compared baselines from 10 to 60 iterative times. In all baselines, GraphX is the most comparative method which achieves the best performance compared with Neo4j and JanusGraph, demonstrating the effectiveness of employing distributed data-driven techniques in carbon emission control estimation. Our method is considerably better than GraphX, the best baselines method, in all experimental scenarios. The result strongly proves the effectiveness of our proposed method in data-driven carbon emission control estimation.

5. Conclusions

When the traditional method is used to calculate the carbon flow transfer, due to the large scale of the power grid and the complex connection relationship, the resource demand for the calculation of the electrical carbon factor is large. To solve this problem, this paper studies the source of carbon intensity data, data analysis steps, and calculation method of carbon intensity, realizes the pilot application of “One Electricity Carbon Chart” to meet the business requirements, data requirements, components, resource requirements of the data center platform, and designs a carbon-based on spark graphx intensity iterative calculation scheme. Combining graph computing and knowledge graph technology, this paper builds a panoramic electricity-carbon coupling model across provinces, regions, disciplines, and information systems. The model uses power flow analysis as a means to track carbon flows, and realizes accurate and dynamic time-sharing and partitioning of electricity. The quantification of carbon emission factors effectively supports the government in establishing an electricity carbon emission accounting system, assists enterprises in the transformation of dual control of energy consumption, and promotes the development of the green electricity trading market.

Footnotes

Funding

This work is supported by the State Grid Corporation of China’s Science and Technology Project as well as the Research on Key Technologies for Carbon Emission Monitoring and Diagnosis of Urban Energy and Power Based on Multi-source Heterogeneous Data Fusion and Sharing (No.5700-202290184A-1-1-ZN).

References

Spaargaren

Van Vliet

. Lifestyles, consumption and the environment: The ecological modernization of domestic consumption. Environmental politics. 2000; 9(1): 50-76.

Andersen

Massa

. Ecological modernization-origins, dilemmas and future directions. Journal of Environmental Policy and Planning. 2000; 2(4): 337-345.

Dong

Wang

, et al. The process of peak CO2 emissions in developed economies: A perspective of industrialization and urbanization. Resources, Conservation and Recycling. 2019; 141: 61-75.

Wang

Harindintwali

Yuan

Wang

Yin

Huang

, et al. Technologies and perspectives for achieving carbon neutrality. The Innovation. 2021; 100180.

Iqbal

Abbas

Shinwari

Guangcai

Ahmad

Tang

. Does exports diversi-fication and environmental innovation achieve carbon neutrality target of OECD economies? Journal of Environmental Management. 2021; 112648.

Hawkes

. Long-run marginal CO2 emissions factors in national electricity systems. Applied Energy. 2014; 197-205.

Mwasilu

Justo

Kim

, et al. Electric vehicles and smart grid interaction: A review on vehicle to grid and renewable energy sources integration. Renewable and Sustainable Energy Reviews. 2014; 34: 501-516.

Das

Tey

Seyedmahmoudian

Mekhilef

Idris

MYI

, Deventer

Horan

Stojcevski

. Forecasting of photovoltaic power generation and model optimization: A review. Renewable and Sustainable Energy Reviews. 2018; 912-928.

Scheer

Del Grosso

Parton

Rowlings

Grace

. Modeling nitrous oxide emissions from irrigated agriculture: testing DayCent with high-frequency measurements. Ecological Applications. 2014; 528-538.

10.

Chen

Shen

van Ake

Maier

. Dual-functionalized double carbon shells coated silicon nanoparticles for high performance lithium-ion batteries. Advanced Materials. 2017; 1605650.

11.

AlRashidi

El-Hawary

. A survey of particle swarm optimization applications in electric power systems. IEEE Transactions on Evolutionary Computation. 2008; 13(4): 913-918.

12.

Zhao

Jiang

Nie

Chen

. How to improve the market efficiency of carbon trading: A perspective of China. Renewable and Sustainable Energy Reviews. 2016; 1229-1245.

13.

Zhao

. Research on the efficiency of carbon trading market in China. Renewable and Sustainable Energy Reviews. 2017; 1-8.

14.

Bond

Syropoulos

. The size of trading blocs market power and world welfare effects. Journal of International Economics. 1996; 411-437.

15.

Botterud

Zhou

Wang

Bessa

Keko

Sumaili

Miranda

. Wind power trading under uncertainty in LMP markets. IEEE Transactions on Power Systems. 2011; 894-903.

16.

Cai

Wang

Jin

Chen

. Quantifying baseline emission factors of air pollutants in China’s regional power grids. Environmental Science & Technology. 2013; 3590-3597.

17.

Gehrels

. Sea-level changes since the Last Glacial Maximum: An appraisal of the IPCC Fourth Assessment Report. Journal of Quaternary Science. 2010; 25(1): 26-38.

18.

Hergoualc’h

Mueller

Bernoux

, et al. Improved accuracy and reduced uncertainty in greenhouse gas inventories by refining the IPCC emission factor for direct N2O emissions from nitrogen inputs to managed soils. Global Change Biology. 2021; 27(24): 6536-6550.

19.

Sharma

Saxena

Sethi

, et al. Life cycle assessment of buildings: A review. Renewable and Sustainable Energy Reviews. 2011; 15(1): 871-875.

20.

Hickmann

. Voluntary global business initiatives and the international climate negotiations: A case study of the Greenhouse Gas Protocol. Journal of Cleaner Production. 2017; 169: 94-104.

21.

Andrew

. A comparison of estimates of global carbon dioxide emissions from fossil carbon sources. Earth System Science Data. 2020; 12(2): 1437-1465.

22.

Jha

Sharma

Singh

, et al. Greenhouse gas emissions from municipal solid waste management in Indian mega-cities: A case study of Chennai landfill sites. Chemosphere. 2008; 71(4): 750-758.

23.

De Laquil

Wenying

Larson

. Modeling China’s energy future. Energy for Sustainable Development. 2003; 7(4): 40-56.

24.

Druckman

Bradley

Papathanasopoulou

Jackson

. Measuring progress towards carbon reduction in the UK. Ecological Economics. 2008; 594-604.

25.

Gang

Xiao

. Determination of CO2 emission reduction based on energy integration. Chemical Progress. 2006; 1467-1470.

26.

Zhihui

Chunjing

Kun

. Building Life Cycle Carbon Emission Evaluation. Building Economy. 2010; 44-46.

27.

Guoquan

Zeyuan

Zhaohua

. Factor Decomposition Model and Empirical Analysis of Carbon Emissions in China: 1995–2004. China Population

\cdot

Resources and Environment. 2006; 158-161.

28.

Liu

Tian

Wang

, et al. The impact of farmland circulation on the carbon footprint of agricultural cultivation in China. Economic Analysis and Policy. 2023; 78: 792-801.

29.

Zhang

Luo

Liu

, et al., Towards low-carbon cities through building-stock-level carbon emission analysis: A calculating and mapping method. Sustainable Cities and Society. 2022; 78: 103-133.

30.

Gupta

Bherwani

Gautam

, et al. Air pollution aggravating COVID-19 lethality? Exploration in Asian cities using statistical models. Environment, Development and Sustainability. 2021; 23: 6408-6417.

31.

Sun

Wang

Zhang

. Factor analysis and forecasting of CO2 emissions in Hebei, using extreme learning machine based on particle swarm optimization. Journal of Cleaner Production. 2017; 162: 1095-1101.

32.

Kumar

Huang

. Graphone: A data store for real-time analytics on evolving graphs. ACM Transactions on Storage (TOS). 2020; 15(4): 1-40.

33.

Sahu

Mhedhbi

Salihoglu

, et al. The ubiquity of large graphs and surprising challenges of graph processing: extended survey. The VLDB Journal. 2020; 29: 595-618.

34.

Cheng

Chen

Wang

, et al. Efficient top-k vulnerable nodes detection in uncertain graphs. IEEE Transactions on Knowledge and Data Engineering. 2021; 1: 1-14.

35.

Sahu

Mhedhbi

Salihoglu

, et al. The ubiquity of large graphs and surprising challenges of graph processing: extended survey. The VLDB Journal. 2020; 29: 595-618.

36.

Heidari

Simmhan

Calheiros

, et al. Scalable graph processing frameworks: A taxonomy and open challenges. ACM Computing Surveys (CSUR). 2018; 51(3): 1-53.

Data-driven smart grid carbon emission control methods using graph-based power flow computing

Abstract

Keywords

1. Introduction

2.1 Current status of domestic and international research

2.2 Current status of industry-specific applications

3. Proposed method

3.1.1 Power grid structure analysis

Table 1 Parsing matching data tables

(1) Automatically extract CIME files

(2) Automatic timing tasks

(3) Automatic deletion of extracted data

3.1.4 Network frame and report correlation analysis

Table 3 Regional report

(1) B01 table

(2) B02 table

(3) B03 table

3.2.1 Graph computing basic data analysis

3.2.2 Research on the flow direction power supply of the power plant

3.2.3 Research on design of graph computing algorithms

4.1 Calculation description

4.1.1 Grid situation in P10

Table 4 P10 grid situation

4.2 Carbon emission control efficiency experiment

Table 5 Experimental result table of iterative time cost

5. Conclusions

Footnotes

Funding

References

Table 1
Parsing matching data tables

Table 3
Regional report

Table 4
P10 grid situation

Table 5
Experimental result table of iterative time cost