Inﬂuence Mechanism of Knowledge Network on Regional Innovation Capability - Based on Hydrogen Energy Industry

. As an important factor of innovation, knowledge plays a vital role in promoting the growth of innovation ability. In recent years, many scholars have explored the inﬂuence of knowledge network structure on innovation. The hydrogen energy industry is highly technical and covers a wide range, which is in urgent need of regional cooperation and easy to form a knowledge network. The function of the knowledge network depends on inter-organizational communication, so tra ﬃ c development may be a boundary condition. This paper takes the hydrogen energy industry as an example to study the inﬂuence mechanism of the knowledge network on regional innovation ability and explore the moderate e ﬀ ect of high-speed rail opening. This paper takes hydrogen patents as samples, uses backward citation data to construct the knowledge network, and conducts regression tests. The results show that the density of the knowledge network has a negative e ﬀ ect on regional innovation ability, while the structural hole and centrality have a positive e ﬀ ect. The opening of high-speed rail will strengthen the inﬂuence of density and structural holes, and weaken the inﬂuence of centrality. Finally, it is proposed that the development of the hydrogen energy industry should reduce the dependence on the network, and the government should prevent the loss of large enterprises.


Introduction
As the innovation-driven development strategy has been proposed and consensus has been reached, strategic emerging industries such as new energy have become the fundamental way for the country to accelerate industrial upgrading, adjust industrial structure, improve the economic structure and promote economic growth in the process of accelerating national knowledge transformation and upgrading and structural reform. At the 75th session of the United Nations General Assembly, the Chinese government proposed that "carbon dioxide emissions should peak before 2030 and achieve carbon neutrality by 2060". Since hydrogen energy comes from a wide range of sources, is clean and carbon-free, and has a variety of application scenarios, it can be regarded as the best choice to promote the realization of carbon neutrality [1]. The hydrogen energy industry has strong innovation, a long industrial chain, and strong support from the state. Therefore, the healthy and rapid development of the hydrogen energy industry is expected to play a good driving and active role in regional innovation and promote the transformation and upgrading of the regional energy economic structure [2]. The top-level design of hydrogen energy requires cross-department, cross-industry and crossdisciplinary collaboration. In the past, the mode of a single discipline, single industry, and single department is not competent for the top-level design of hydrogen energy, so the hydrogen energy industry needs to break down regional cooperation and exchange, which plays a crucial role in the construction of knowledge network and the improvement of regional innovation ability. In conclusion, the hydrogen energy industry has an extremely important impact on regional innovation ability, which is mainly realized through knowledge network.
In recent years, there are more and more research on innovation synergy and regional innovation [3,4], but few scholars refer to innovation synergy in the knowledge network. Current research indicates that knowledge networks may play a dual role in regional innovation development. The rational use of the knowledge network can increase the efficiency of knowledge search, obtain more non-redundant knowledge, promote the circulation of knowledge and improve the level of innovation. Therefore, how to balance the position of the knowledge network and the role of knowledge network characteristics becomes a proposition worth further study. The research on the characteristics of knowledge networks is helpful to provide suggestions for regional innovation and development to break through the bottleneck, help the government to formulate relevant policies to promote innovation and development, and help enterprises to take measures to improve the level of technological breakthrough. Based on the characteristics of the knowledge network and regional innovation development, this paper comprehensively analyzes its connection and influence mechanism and makes an empirical study on the lack of knowledge networks in the hydrogen industry on regional innovation. Therefore, based on hydrogen patents and citation data, this paper constructs China's interregional knowledge network, analyzes the influence of network density, structural holes, and centrality on regional innovation capability, and explores the boundary condition of the influence mechanism (i.e., the opening ratio of high-speed rail).
The main contributions of this paper are as follows: First, it provides a method for the construction of an interregional knowledge network of the hydrogen energy industry, studies the influence of the knowledge network on regional innovation ability, and takes specific industries as the perspective. Second, the empirical analysis of the influence of the knowledge network is carried out. The analysis model and corresponding countermeasures and suggestions can provide the scientific basis for future research. Finally, the scope of this research framework is not limited to specific industries and regions but also applies to other industries or regions where knowledge networks exist.
Section 2 reviews the literature on the knowledge network and regional innovation capability and then presents the study's hypotheses on the characteristics of the knowledge network and the moderating effect of high-speed rail opening. This is followed by Section 3, a section on sample establishment, variable measurement, and methodology. Section 4 reviews the empirical analysis to test the hypotheses and robustness test. Section 5 reviews the results and discusses their practical implications and limitations and avenues for future research.

Literature Review
(1) Knowledge network The study of the knowledge network in management began in 1995, and the cognition at that time could be summarized into two models. One was the economic model, which believed that the knowledge network was limited to the academic scope and could not be applied, but it could generate and transfer knowledge [5]. The other is the market structure model, which holds that an enterprise is an individual of the knowledge network and expands the knowledge stock through research and development activities. This model mainly studies how knowledge penetration affects the market structure [6]. Knowledge network embedding can be divided into relational embedding and structural embedding. The former mainly refers to the strength and direction of the relationship, while the latter is mainly discussed based on the resources that the enterprise can control and the strength of its control in the position of the social network [7]. The network characteristics related to this study are network density, structural holes, and centrality. The network density is the relational dimension and the network structure hole and centrality are the structural dimension.
For the overall social network, the higher the network density, the closer the relationship between network members. The network density reflects the density of connections in the network. The level of network density has a significant impact on the effect of knowledge sharing. The higher the network density, the more connections between the nodes in the network, and the more fully and effectively the communication between the members of the organization [8]. There is a discontinuity between organizations or individuals in the social network (that is, there is no direct connection between the middle nodes of the social network). Therefore, from the perspective of the overall structure of the network, there are holes in the network structure, which are structural holes [9]. For example, there are three types of people: A, B, and C, A has contact with B and C, but B and C have no contact, then there is a structural hole between B and C. Structural holes can form information asymmetry phenomenon, A between B and C, has an obvious competitive advantage, it can obtain more information, to benefit output. Network centrality mainly evaluates the degree to which an enterprise occupies a strategic position in the network, and is the degree to which network members are close to the core of the network system. The different positions of enterprises in the knowledge network will result in differences in knowledge acquisition advantages. The higher the centrality, the higher the status and power of enterprises [10,11].
(2) Regional innovation capability Regional innovation power refers to the power of a region to transform knowledge into new products, new technologies, and new services, to promote the development of the local economy, and to produce and determine economic performance. The regional innovation ability is mainly composed of regional leading industry clusters and regional institutional infrastructure, as well as their interaction [3]. The enterprises, public administration, innovation support infrastructure, and regional and national innovation environment are the most important components of the regional innovation system [4].
With the deepening of research on innovation capability, foreign research institutions and governments have launched several innovation evaluation index systems, among which the National Innovation Capability Index of the United States, the OECD's "Science, Technology, and Industry Scoreboard" and the EU's innovation Scoreboard have been widely concerned and cited by people. According to the division of the input-output process, foreign innovation measurement evaluation indicators generally include innovation input and innovation output indicators, such as R&D expenditure, R&D personnel, patents, papers, and so on [12]. The research on the regional innovation system in China is later than that in foreign countries, but it develops quickly. Since 2001, the China Science and Technology Development Strategy Group has published the Annual Report on China's Regional Innovation Capacity. The report takes the regional innovation capability of 31 provinces and cities in China except for Hong Kong, Macao, and Taiwan as the evaluation object. By quantifying the innovation capability, the innovation capability of each region is compared vertically and horizontally. Many scholars have used this evaluation system for empirical analysis.

Hypotheses Development
(1) Characteristics of knowledge network Knowledge is an important source of competitive advantage. In the context of open innovation, regional economic development differences are closely related to knowledge. It can be said that in a dynamic competitive environment, the regional innovation advantage cannot be maintained only through repeated labor and low-end industries. Knowledge innovation and knowledge output should also be used to promote the improvement of regional innovation ability [13]. Network embedding is the core concept of network theory, which reflects the importance of actors in the network and the connection with other nodes. The structure of actors in the network determines the knowledge and configurable resources available to actors in the network, thus affecting their ability and performance [14].
The high-density network will generate a large number of connections between enterprises, and information and resources will flow more quickly within the network. The highdensity networks are more likely to develop trusting relationships, shared norms, and common patterns of behavior [8]. However, hydrogen patents are highly technical, covering a wide range, and the core technology is difficult to break through. Excessive network density is not conducive to the protection of innovation achievements, making many enterprises lose part of the motivation for innovation. An excessively close connection is not conducive to the transmission of new knowledge and information, resulting in redundant knowledge and information, which is not conducive to the growth and innovation of enterprises. The existence of structural holes means that the subject has the opportunity to obtain two kinds of heterogeneous information flows, which is conducive to the acquisition of information and the control of the enterprise's resource acquisition is realized by constantly exploiting the structural holes in the network, and the network enterprise can win the competitive advantage by constantly changing the network structure [9,15]. The nodes in the center of the knowledge network usually have more connection channels and higher network accessibility. They are relatively free from the control of other enterprises and easier to access a large amount of external information, with fast resource flow and information update speed. It can obtain more high-quality information and gain the control advantage of knowledge resources in the network, which is conducive to innovation [16]. In summary, the density of the knowledge network in the hydrogen industry is not conducive to the improvement of regional innovation ability, while the structural hole and centrality are opposite. Therefore, this paper proposes the following hypothesis: H1a: Knowledge network density of the hydrogen industry negatively affects regional innovation capability.
H1b: Knowledge network structural hole of the hydrogen industry positively affects regional innovation capability.
H1c: Knowledge network centrality of the hydrogen industry positively affects regional innovation capability.
(2) Moderating effect of high-speed rail opening The emergence of high-speed railways speeds up the flow of production factors such as manpower, material resources, and funds between cities and breaks the barriers between cities [17]. The improvement of infrastructure such as high-speed rail will accelerate the transfer of regional elements and improve regional accessibility [18] . Therefore, the opening of highspeed rail can change the structure of the knowledge network and the influence of knowledge network characteristics on regional innovation ability.
The opening of high-speed railways promotes the regional flow of talent, which is conducive to the transmission of new knowledge and information. The mobility brought by high-speed railways weakens the importance of regional network density and reduces the dependency between enterprises in the region. At the same time, the opening of high-speed rail will accelerate the transfer of elements from surrounding cities or regions to central cities. The "congestion cost", and the rise of wages and rents brought by big cities will make industries migrate to surrounding cities, resulting in the loss of the center of the knowledge network and the weakening of the influence of network centrality. The enterprises occupying the position of the structural hole are less competitive than those in the core position, so they lack the resources of regional flow. Meanwhile, the industrial migration after the opening of the high-speed railway makes the role of structural hole position prominent, and the demand for heterogeneous information flow is even greater. To sum up, with the opening of high-speed rail opening, the effect of knowledge network density and centrality on regional innovation capability will weaken, and the structural hole is the opposite. Therefore, this paper puts forward the following hypothesis: H2a: High-speed rail opening weakens the negative effect of knowledge network density on regional innovation capability.
H2b: High-speed rail opening strengthens the positive effect of the knowledge network structural hole on regional innovation capability.
H2c: High-speed rail opening weakens the positive effect of knowledge network centrality on regional innovation capability.
The research framework of this paper is shown in figure 1.

Research Samples and Data Sources
In this paper, the patents classified as the industry category of hydrogen energy and published by the State Intellectual Property Office of China between 2000 and 2017 were selected as samples to build the basic database. Considering the high technical barriers in the hydrogen industry and the focus on regional innovation capabilities, our research focuses only on Figure 1. Research framework invention patents. Our research samples contain patents' basic information and citation information. We added up the data of 31 provinces and municipalities in mainland China from 2000 to 2017 and put the patents at the provincial level. Based on the annual number of patent backward citations among provinces, we constructed a knowledge network and a matrix of patent citations of 31*31 to measure the hydrogen patent knowledge network among provinces and import the software Ucinet6.0 to analyze its knowledge network characteristics. The regional innovation capability was proxy by the comprehensive utility value of China's regional innovation capability. Finally, a total of 539 variables in 31 provinces and municipalities were finally established as the research samples of this study.

Variables
(1) Dependent variable Regional innovation capability (Innovation). The innovation capacity of a region is determined by the potential to produce a series of related innovative products, the most important factor is the R&D stock [19]. In China, the most authoritative evaluation index system of regional innovation capability is the one adopted in the Report of China's Regional Innovation Capability issued by China Science and Technology Development Strategy Research Group. The index system includes five aspects: innovation environment, knowledge creation, knowledge acquisition, enterprise innovation ability, and economic benefits of innovation, with 174 specific indicators. We are concerned about China's regional innovation capability, so we choose the comprehensive utility value of China's regional innovation capability as an indicator of regional innovation capability.
(2) Independent variable Density (Density). Network density refers to the degree of interaction between network members, that is, the average degree of interaction between enterprises [10]. High density means that any member of the network has more connections with other members, while low density means that each member has fewer connections with each other. Knowledge network density measures the extent to which contacts (alters) are connected to each other [20].
Structural holes (SH). Structural holes indicate a kind of network position benefit. When the other two units connected by a main body are not directly connected with each other, the main body occupies the position of structural holes. Since there is often redundant information among closely connected units, the existence of structural holes means that the subject has the opportunity to obtain two heterogeneous information flows, which is conducive to the acquisition and control of information [9].
Centrality (Centrality). Network Centrality is an index used to determine whether an enterprise is in the center, the periphery, or the edge of the network. It is used to determine the network position of an enterprise in the network [21], that is, centrality represents the position of an enterprise in the network structure. (

3) Moderating variable
High-speed rail opening (HST). High-speed railway refers to the special passenger train railway designed to run at a speed of more than 250 km per hour (including reservation), and the initial operation speed of more than 200 km per hour. In this paper, the number of prefecture-level cities with high-speed rail lines is divided by the total number of prefecturelevel cities in the province to represent the rate of high-speed rail opening.

(4) Control variable
For the region, we control for their scale by using the number of the resident population (RP) as a proxy, while trying to ensure that provinces of the same level are compared [22]. Because passenger turnover (PTO) is to some extent a proxy for the flow rate of talent, so we control it. Considering that the scientific research output of teachers and students in colleges and universities has a knowledge spillover effect, which promotes regional technological innovation, we also control for the number of higher education institutions (HS) and the number of college graduates (GN) [23]. Given that technical personnel can promote knowledge spillover and flow, and human capital has a positive correlation with technological development level and innovation ability [24], we control the full-time equivalent of RD personnel in industrial enterprises above designated size (RDP). The process of independent innovation is also the process of the formation and cultivation of patent capability. Patents are closely related to scientific and technological progress. Researchers believe that patents are effective indicators to measure technological innovation [25], so we control the number of granted patents (PG) and the number of invention patent applications (IPA). Finally, year dummy variable is included in each model. A full list of variables used in our empirical study, along with brief descriptions, is provided in Definition Innovation The comprehensive utility value of China's regional innovation capability. Density The total number of identified relationships over the theoretical maximum number of relationships.

SH
The effective scale of a node is the individual network scale of the node minus network redundancy. Centrality The extent to which it occupies a central position.

HST
The proportion of the number of prefecture-level cities with high-speed rail service in each province to the total number of prefecture-level cities. PG The number of authorized patent applications. PT0* Passenger turnover. RP* The number of permanent residents. HS The number of institutions of higher learning. GN* The number of graduates from ordinary institutions of higher learning. IPA The number of invention patent applications.

RDP
The full-time equivalent of RD personnel in industrial enterprises is above the designated size.
These variables are used in logarithmic form. For variables that are measured as numbers, this takes the form, of a log(variable+1).

Model Specification
To verify the research hypotheses proposed herein, we construct the following regression model: (1) where the vector Control i represents the control variables related to firms, industry, and region characteristics, and e i is an error term. Considering that the dependent variable is the comprehensive utility value of China's regional innovation capability, and the samples are panel data, this study uses ordinary least squares regression to verify our hypotheses. After the houseman test, we adopted the fixed effect model to estimate the model parameters more reasonably. table 2 shows the results of the correlation analysis for the main variables involved. Most of the correlation coefficients are less than 0.6. Moreover, the variance inflation factor coefficients are all less than 10, indicating no multicollinearity concerns.  table 3 presents the regression analysis results of the impact of the knowledge network on regional innovation capability. Model (1) shows only the baseline model with the control variables. The independent variables density, structural holes, and centrality are inserted into Models (2). Subsequently, the moderate variables HST, and their interactions are added in Models (3), (4), and (5). Each specification includes a complete set of year-fixed effects. The results in the model (1) show that, as the number of higher education institutions and R&D personnel increases, regional innovation capacity will increase and as passenger turnover and the number of college graduates increased, the result was reversed. The results in Models (2) show that knowledge network' density has a negative effect on regional innovation capability (β = −2.891, p < 0.05), while structural holes and centrality have a positive effect on regional innovation capability (β = 3.693, p < 0.01; β = 0.006, p < 0.01). It indicates that with the increase of knowledge network density, regional innovation ability will decrease and the increase of knowledge network structure hole and centrality index can promote the enhancement of regional innovation ability. Therefore, we find support for H1a, H1b, and H1c.

Regression Analysis and Results
Next, we predict that the rate of high-speed rail opening can moderate the relationship between the knowledge network and regional innovation capability. The results of Models (3), (4), and (5) show the results. The interaction coefficients in Models (3) and (4) are positive and statistically significant (β = 0.025, p < 0.001; β = 0.830, p < 0.01). The interaction coefficients in Models (5) are negative and statistically significant (β = −0.019, p < 0.001). It indicates that the rate of high-speed rail opening enhances the positive effect of structural holes while weakening the effect of density and centrality. Hypothesis 2a, 2b, and 2c are supported.

Robustness Test
To avoid potential endogeneity problems, we make the following efforts 1 . First, in order to avoid potential reverse causality, we use the regional innovation index to conduct regression analysis on the relational and structural embedding of the knowledge network. If the coefficient is significant, there may be an endogeneity problem. The regression results showed that all the coefficients were not significant, indicating that endogenous anxiety caused by potential reverse causality did not exist. Secondly, the fixed effects model was used for the main regression analysis. In order to solve the possible bias of different models, the random effects model was used as the robustness test in our supplementary analysis. The results are basically consistent with those obtained by the original model. Third, the influence of the knowledge network on innovation capability is a long-term process. Therefore, we redefine variable Innovation using an alternative measure to lag the innovation index by two years. table 4 shows the results, which are consistent with the main findings reported in this study.

Conclusions
Based on the data of invention patents and citations published by China's State Intellectual Property Office from 2000-2017, this paper establishes the knowledge network of the hydrogen energy industry among 31 provinces and tests the influence of the knowledge network on regional innovation ability and explore the moderate effect of the high-speed rail opening. We find that knowledge network density has a negative influence on regional innovation ability. In contrast, knowledge network structure holes and centrality can promote regional innovation ability. Furthermore, we investigate the boundary condition for the influence mechanism. The results show that the rate of high-speed rail opening strengthens the effect between knowledge work structure hole and regional innovation capability while weakening the effect of knowledge network density and centrality on regional innovation capability.

Practical Implications
This study has several important implications for both policymakers and managers. First, enterprises should focus on independent innovation and cannot rely solely on knowledge network interaction and industry-leading enterprises. For an industry, realizing scientific and technological innovation through independent research is a prerequisite for the sustainable development of the whole industry, and then improving the regional innovation capability. Leading enterprises in the industry with rich resources and innovation capability should carry out core technology innovation. Moreover, they should take the initiative to take the responsibility to transfer knowledge to other enterprises through the knowledge network. In this way, small and medium-sized enterprises can make full use of the knowledge to carry out independent innovation instead of relying too much on the knowledge network. Second, companies that dominate the network have more resources and are more innovative. Therefore, we should give play to the advantages of the enterprise, actively strive for the research and development of core technology, and quickly digest and absorb internal and external knowledge. Although enterprises in remote locations have weak R&D capability, they are more responsive to the knowledge network and have more flexible management. They can acquire knowledge through the knowledge network and actively strive for cooperative R&D with core enterprises or undertake outsourcing projects. In short, by making full use of the resources and capabilities of core enterprises, peripheral enterprises gradually move closer to the center through outsourcing projects or cooperation. Third, although the opening of high-speed rail promotes the circulation of knowledge and technical personnel, it increases the risk of the loss of enterprises in the center of the knowledge network to some extent. Therefore, the government should actively take incentive measures to give certain policy subsidies and technical support to leading enterprises to prevent huge losses caused by the loss of enterprises. At the same time, in addition to providing some convenience to enterprises in the central position, the government should also pay attention to enterprises in the structural hole position. The opening of the high-speed rail makes the non-redundant connection of the knowledge network more prominent, and such enterprises control the possibility and efficiency of knowledge interaction.

Limitations and Avenues for Future Research
Even with all our efforts, there are still certain limitations in this study. First, owing to constraints such as data availability, we only used the data from 2000 to 2017. Future research can use the updated and further investigate how knowledge network affects regional innovation. Second, the research framework of this paper does not cover all industries, and other industries can also be looked at for future research. Third, restricted by the data source, we study only the impact of the knowledge network on regional innovation capability under one boundary condition. In future research, more industry-or region-level factors and their interactions can be considered. Finally, due to data limitations, this paper conducts research in the context of China. In future studies, the research background can be extended to other developing countries or developed countries to further verify our research framework.