Knowledge Mapping in Electricity Demand Forecasting: A Scientometric Insight

Forecasting electricity demand plays a fundamental role in the operation and planning procedures of power systems and the publications about electricity demand forecasting increasing year by year. In this paper, we use Scientometric analysis to analyze the current state and the emerging trends in the field of electricity demand forecasting form 831 publications of web of science core collection during 20 years: 1999-2018. Employing statistical description, cooperative network analysis, keyword co-occurrence analysis, co-citation analysis, cluster analysis, and emerging trend analysis techniques, this article gives the most critical countries, institutions, journals, authors and publications in this field, cooperative networks relationships, research hotspots and emerging trends. The results of this article can provide meaningful guidance and some insights for researchers to find out crucial research, emerging trends and new developments in this area.


Introduction
Nowadays, electricity has become one of the most important energy and plays an indispensable role in many fields. In recent years, a large number of researches have proved that the accuracy of electricity demand prediction plays a fundamental role in the operation and planning procedures of power systems [1,2]. Accurate electricity demand forecasting can not only ensure the reliable operation of power systems but also have a great cost-saving potential for power corporations [3]. On a temporal scale, electricity demand forecasting can be classified into short-term, medium-term and long-term. With the increase of electricity demand and the rapid development of artificial intelligence technology, electricity demand forecasting has attracted more and more scholars' attention and new research methods, emerging trends and new developments have been generated at the same time [4]. A large number of forecasting researches and methods have been proposed and applied in the field of electricity load forecasting in recent years [5,6].
Mohan et al. [7] proposed a data-driven method for short-term load forecasting using dynamic mode decomposition. Musaylh et al. [8] proposed a hybrid model that including the multivariate adaptive regression, artificial neural network and multiple linear regression models to forecast short-term electricity demand in Australia. Shao et al. [9] conducted decomposition methods for electricity demand forecasting and presented that Empirical mode decomposition and wavelet decomposition are the most popular technique. Kuster et al. [1] presented a review which revealed that artificial neural network, multivariate regression, time series analysis and multiple linear regression are popular and effective methods for electricity and electricity forecasting. Hong and Fan [10] offered a tutorial review of probabilistic electric load forecasting and introduced the techniques, methodologies, applications, evaluation methods and future research needs. However, very little research has analyzed the collaborative relationship, new developments and emerging trends of electricity demand forecasting and visualized the knowledge map in this field.
Scientometrics is an important method to find out the rules of scientific activities, identify research trends, and evaluate the development of the field [11,12]. Yu and Xu analyzed the current status and explore future research trends of the carbon emission trading domain by the scientometric method [13]. Olawumi and Chand evaluated the research development status of institutions, countries, and journals in the research field [12]. Niazi and Hussain evaluated all sub-domains of agent-based computing and found agent-based computing extensive in other dominos [14].
With the rapid growth of attentions and publications for electricity demand forecasting, it is necessary and urgent to summarize the current situation and analyze the collaborative relationship, new developments and emerging trends of electricity demand forecasting. In this paper, scientometrics analysis is performed in the electricity load forecasting domain and utilizing software named CiteSpace to analyze and visualization the emerging trends. CiteSpace, invented by Dr. Chen Chaomei, is a particularly popular method of scientometrics that uses citation analysis in a visual form and can be used to identify knowledge areas and emerging trends in researches [15,16]. In recent years, CiteSpace has attracted the interest of many scholars and has been applied to many fields. Chen used published literature to investigate Emerging trends and new developments in regenerative medicine [17]. Yang et al. comprehensively analyzed the status of PM2.5 research and found the frontiers of research in this field [18]. Fang et al. analyze the interaction between climate change and tourism and describe the research characteristics of the field in the past 25 years [19].
The structure of this article is as follows: Section 2 gives the source and search strategy of publications. Section 3 introduces the basic summary of electricity demand forecasting research. In Section 4, we visualize the cooperation network of authors, institutions, and countries. Section 5 analyzes the active topics and emerging trends in electricity load forecasting, including keyword analysis and co-citation analysis. Section 6 gives comprehensive conclusions and discussions.

Methodology
This section provides the search strategy of data. For the searched phrase in WOS, some articles perform an exact search on a certain phrase, such as Yu and Xu, and some articles perform an exact search on multiple phrases and merge the results, such as Chen [13,20]. Few articles obtain articles through searching inexact themes, because searching inexact themes requires that the query words do not have to appear consecutively, which gets a large number of publications that are not related to the search subject. It is worth noting that this article searches precise themes and nonprecise titles. This article focuses on a more subdivided field, and the number of related articles is little. Searching precise themes will ignore indispensable publications in the field of electricity load forecasting and affect the conclusion of this article seriously. To improve the recall rate and avoid a large number of irrelevant publications being retrieved, this article adopts the strategies of searching precise themes and inexact titles which ensures the accuracy of publications being retrieved through means of manual screening.
The data used for analysis in our research is downloaded from Web of Science (WoS), and the search strategy we followed is below: (1) (TS=("electricity demand forecasting" OR "electricity demand prediction" OR "electrical demand forecasting" OR "electrical demand prediction" OR "electric demand forecasting" OR "electric demand prediction" OR "power demand forecasting" OR "power demand prediction" OR "electricity consumption forecasting" OR "electricity consumption prediction" OR "electrical consumption forecasting" OR "electrical consumption prediction" OR "electric consumption forecasting" OR "electric consumption prediction" OR "power consumption forecasting" OR "power consumption prediction" OR "electricity load forecasting" OR "electricity load prediction" OR "electrical load forecasting" OR "electrical load prediction" OR "electric load forecasting" OR "electric load prediction" OR "Electricity load forecasting" OR "Power load prediction" OR "grids load forecasting" OR "grids load  (4) Document types = "article" or "review";

Basic summary of electricity demand forecasting research.
This section provides the statistical analysis form five parts including distribution of time, subject categories, high-yield journals, high-yield institutions, high-yield authors and highly cited publications in electricity demand forecasting.  publications in electricity demand forecasting, the US for 9.99%(83), Iran for 6.74%(56) and the UK for 6.14%(51).   Table 1 lists the top 10 journals, it can be seen that energy and power are areas of greatest concern to the top 10 journals. "Energy" is the highest yield journal with 81 publications, followed by "Energies", "International Journal of Electrical Power Energy Systems", "Applied Energy", "Energy Conversion and Management","

Subject categories
Electric Power Systems Research"," Energy and Buildings"," International Journal of Forecasting"," IEEE Transactions on Power Systems" and " Lecture Notes in Computer Science". In the top 10 journals, the impact factor of "Energy"," Applied Energy"," Energy Conversion and Management" and " IEEE Transactions on Power Systems" are all more than 5.       Table 5 shows the top 10 papers with the highest average citations per year.

High-yield Authors
Obviously, only the articles published by Li

Co-occurrence network
Keywords are a clear sign of understanding the key content of research papers.
Co-occurrence analysis is the number of times the term "statistics" appears in an article to measure the relationship between different articles and to further understand the research status in this field. The burst detection of keywords is often used for the emergence of hotspots and active topics in the field. Figure 9 shows a keyword cooccurrence network for power demand forecasting. For ease of observation, Figure 9 only retains points where the co-occurrence frequency is greater than 10. The key words co-occurrence network is intricate and complex, and the nodes are closely   9. Keyword co-occurrence network in electricity demand forecasting. Table 6 shows the 17 keywords with the highest burst detection. An entity with a frequency burst means that it has a sudden frequency change within a certain period of time. We found that the neural network is the most powerful keyword of burst and its duration is as long as 9 years (1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007), which indicates that the neural network is one of the most important basic methods in the field. It should be noted that the length of the whole line in the last columns of Table 8 represents the total research period (1991-2015) and the red line means the certain time period of the citation burst. At the same time, we also found that the larger keywords detected by burst are mostly method categories. This aspect is because the field of interest in this research is too small and the problems are concentrated. On the other hand, the new hotspots in this field are mostly methodological changes. The timeline in CiteSpace visualizes the cluster along the horizontal timeline

Co-citation network analysis
Co-citation network analysis is an analysis tool, usually used to examine a large number of documents and reveal the knowledge map of a scientific discipline. It analyzes and examines the related literature (such as documents, journals or authors) and is cited by other literature. Table 7 Top 10 references with the strongest citation bursts during 1999-2018. Table 7  in these papers. So it becomes a key node in the network [6]. Taylor  networks, support vector regression, ant colony and particle swarm optimization [46]. Fig.11. Reference co-citation network in electricity demand forecasting. Fig.12 is a clustering network co-citation in the literature. It is obvious that the network has six clusters of a combined model, price forecasting, electricity consumption forecasting, peak load forecasting, support vector regression, neural networks, probabilistic forecasting, and turkey. Figure 13 is a line graph of the cocitation of the literature, similar to the keyword timeline. It shows the time evolution process of the six clusters. From the clustering results, we can find that the earliest clustering is neural networks, which is also the same as the keyword clustering results, which together illustrate the importance of neural networks in this field. The combined model is the largest cluster, and the time is close to 2018, indicating that the research frontier is a hybrid model. Price forecasting, electricity consumption forecasting, peak load forecasting and probabilistic forecasting reflect the main content of the research field. Price forecasting and peak load forecasting are the contents of earlier attention. Probabilistic forecasting and electricity consumption forecasting are more concerned in the near future. Support vector regression shows that it is one of the important methods in the field. Turkey was the main cluster between 1998 and 2008, indicating that during this period turkey's power forecasting was an area of concern. From the clustering results, we can find changes in the method of the field and changes in the content of the research. In particular, it is pointed out that the recent research method hotspot is the combined model.
In Table 8, Size indicates the numbers of the publications in the cluster. For example, the largest cluster (#0) has 140 members. Silhouette is an index to measure the homogeneity of a cluster, the greater value of this index, the better of the homogeneity. Mean (Year) represents the average year of the published documents of the regarding cluster. It is a very useful index since it can be used to judge the cluster whether is new or old.

Conclusions
In this paper, we made a scientometric review and visualization analysis on electricity load forecasting studies based on 831 publications retrieved from Web of Sciences. An integrated knowledge map of the electricity load forecasting field and hot topics with emerging trends are presented by collaboration network, keywords cooccurrence analysis and co-citation analysis. Some interesting and useful conclusions are as follows.
First, electricity load forecasting has received more and more attention, the numbers of citations and publications are increasing rapidly, especially in the last decade. "Energy fuels", which accounts for 36.82%, is the largest subject category in the electricity load forecasting research area. "Energy" is the highest yield journal with 81 publications, followed by "Energies", "International Journal of Electrical Power Energy Systems" and "Applied Energy". "Renewable & Sustainable Energy The limitations in our research are that, due to the limits of co-citation analysis in citespace, the literature in our paper are only retrieved from the core database of WoS, Document types are limit in "article" or "review" and Literature type are limit in "English", which may make some significant literature have been overlooked.