1. Introduction
Clean energy plays an important role in reducing carbon emissions, so more and more countries vigorously develop clean energy [
1]. As a promotion agency of clean energy industrialization, clean energy incubator has become a crucial part of innovation system. With the burgeoning development of clean energy incubators in recent years, governments all over the world are introducing policies to actively encourage their establishment [
2]. Whereas, these incubators are faced with the problems of valuing quantity and neglecting quality. Some incubators still stay in the stage of property leasing and simple incubation service, but still enjoy the preferential policies of the government and waste public resources [
3]. In addition, the planning of many clean energy business incubators is mainly oriented towards the layout of hardware and space, and ignores the planning of soft capabilities such as incubation ability [
4]. In order to further play the role of incubators and realize their sustainable development, it is necessary to make the factors clear which affect the core competitiveness and propose a feasible evaluation model. The research on the core competitiveness evaluation of clean energy incubator can help managers find out the deficiencies, and then help managers to improve these deficiencies. Thus this study can provide suggestions for enhancing competitiveness of clean energy incubators.
In 1990, C.K. Prahalad and Gary Hamel, professors of Michigan University and London Business School, respectively, formally raised the concept of “core competitiveness” for the first time. They pointed out that core competitiveness was the ability to coordinate different production skills as well as organically combine various technological schools [
5]. Literature [
6] discussed how to integrate technological innovation into the core competitiveness of enterprises. The research held that it was of great theoretical and practical significance to put the core competitiveness evaluation into force for sustainable development. Reference [
7] considered that environmental adaptability was an important component of the core competitiveness in photovoltaic module supply enterprises. Thereby, it provided suggestion in business decision-making. Thirteen influential factors of core competitiveness were selected in modern circulation industry from market expansion, innovation, circulation efficiency, sustainable development and circulation effect in reference [
8]. The evaluation was achieved through study on both independent and synergistic effect of the factors. Hence, it becomes a good entry point to analyze the key resources and capabilities demanded by the sustainable development of enterprises.
At present, scholars have published their significant work on core competitiveness evaluation of incubators, but few focus on the clean energy enterprise. There are some similarities among these research subjects, such as the content and components of kernel competence. Thus the existing study also can provide reference for the research. Reference [
9] developed Principal component analysis (PCA) and SPSS based model to evaluate scientific enterprise incubators of nine cities in Pearl River Delta. The results showed the ranking and classification of incubators and put forward development plan accordingly. Literature [
10] conducted a questionnaire survey on six chain incubators. The relationship between quantitative performance and importance was revealed through IPA in competitiveness evaluation. Reference [
11] outlined the assessment approach of enterprise incubator based on PCA and DEA from the perspective of service and operation efficiency. A comprehensive technique that integrates questionnaire survey with grey correlation assesses the national incubators regarding knowledge service capability in [
12]. Reference [
13] paid close attention to social benefits in core competitiveness assessment of incubators. The evaluation is executed from five aspects, that is, public service, innovation and development, social education, labor employment and enterprise cultivation.
Evaluation methods can be separated into two categories: traditional assessment technique and modern intelligent algorithm. The former needs complex calculation despite its mature development and accurate results, such as fuzzy comprehensive evaluation method [
14] matter-element extension assessment [
15] and TOPSIS [
16]. Reference [
14] developed an approach to reveal the relationship between exposure, sensitivity and adaptive capacity for better flood vulnerability assessment, based on the fuzzy comprehensive evaluation method. Reference [
15] utilized the improved fuzzy matter-element extension assessment model to evaluate urban water ecosystem health effectively. Reference [
16] used the improved TOPSIS model to evaluate the sustainability of power grid construction projects. Modern intelligent approaches chiefly consist of back propagation neural network [
17], extreme learning machine [
18], support vector machine [
19], least squares support vector machine (LSSVM) [
20]. Thereby, this paper intends to combine the traditional evaluation model and intelligent algorithm to assess the core competitiveness of clean energy incubators. Concretely, matter-element extension assessment is first applied to obtain the results. Then, LSSVM is employed for intelligent evaluation. However, owing to blind selection of penalty and kernel coefficients, it’s essential to use an appropriate method for optimization [
21]. NSGA-II is a multi-objective model based on the first generation of non dominated sorting genetic algorithm with the advantage of excellent performance and high efficiency [
22]. For example, reference [
23] used the NSGA-II algorithm to obtain the multi-objective optimisation of an interactive buildings-vehicles energy sharing network with high energy flexibility, which achieved good optimization results. To this end, this paper exploits NSGA-II to automatically decide the appropriate values in LSSVM.
It is worth noting that there exist many influencing factors on the core competitiveness of clean energy incubators. Dimension reduction is critical for accuracy on account of strong coupling, nonlinearity and redundancy in the indicators [
24]. PCA, as a multivariate statistical approach, has the advantage over dealing with indexes with strong linear relationship [
25]. Nevertheless, the relationship among the core competitiveness indicators of clean energy incubators is mostly nonlinear. PCA is not able to acquire high-order features which ignores the nonlinear information while reducing the dimension [
26]. As a result, kernel principal component analysis (KPCA) is used in this paper to map the initial input variables to high-dimensional feature space through nonlinear transformation so that the input dimension can be reduced on the premise of retaining nonlinear information among variables [
27].
Therefore, this paper establishes the core competitiveness index system of clean energy incubator and proposes a hybrid model integrated matter-element extension assessment and KPCA-NSGA-II-LSSVM for evaluation. The rest of paper is organized as follows:
Section 2 designs the indicator system including strategic positioning ability, seed selection ability, intelligent transplantation ability, growth catalytic ability as well as service value-added ability, and introduces the preprocessing method.
Section 3 presents a brief description of the proposed evaluation technique. In
Section 4, a case study is carried out to validate the established model.
Section 5 concludes the paper.
3. Methodology
3.1. Entropy Weight Method
As a common approach in objective weight calculation, the principle of entropy weight method derives from thermodynamics. Entropy is a measure of the system uncertainty. The greater entropy means more chaotic system and less information [
28]. Suppose the system locates in
states, the probability of each state is set as
. The entropy of the system is described as Equation (2)
where
satisfies:
,
.
The following procedures are listed:
(1) Establishment of standardized judgment matrix
In accordance with standardized data obtained from Equation (1), the standardized judgment matrix is acquired.
(2) Calculate the information entropy of each index
(3) Index weighting
where
.
3.2. Evaluation Model Integrated Matter-Element Extension with TOPSIS
TOPSIS, also known as approximate ideal solution, firstly draws up two schemes. One of the scheme presents the best attribute values, called positive ideal solution, while the attributes of the other plan are all the worst, named negative ideal solution. Then, compare each scheme with the positive as well as negative ideal solutions, respectively. The optimal programme is selected which is close to the positive ideal solution and far away from the negative one simultaneously [
29]. Hence, this best scheme shows the highest core competitiveness. Whereas, TOPSIS can not judge the competitiveness level of each plan, that is, it can’t be decided whether the level belongs to very high, high, general or low [
30].
Matter-element extension evaluation method combined matter-element theory, extension set with correlation degree for quantitative evaluation. This model divides the data interval of the evaluated target into several orders and determine their levels. Correlation degree is calculated between each plan and the grade. The larger the correlation degree is, the higher the membership extent present. The level of the evaluated target depends on the grade of data interval with the highest membership degree [
31].
Thus, this paper integrates matter-element extension model with TOPSIS for core competitiveness assessment. TOPSIS is firstly used to determine the positive and negative ideal solutions and divide the interval into several grades by equal distance. Each grade corresponds to the given level of core competitiveness. Then calculate the correlation degree between each scheme and the grades. Hence, the core competitiveness level of the plan can be accordingly judged.
The main measures can be described as follows:
(1) Establishment of standardized index matrix
Suppose the scheme set including
evaluation indicators is
. The assessment indexes are presented as
. The standardized index matrix
is derived from consistent and dimensionless processing as Equation (7)
where
is the standardized indicator of
in scheme
.
(2) Establishment of weighted standardized matrix
The entropy weight method is applied to calculate the weight of indexes
. The weighted standardized matrix is shown as Equation (8)
(3) Determine positive and negative ideal solutions of each scheme
The positive ideal solution is:
The negative ideal solution is:
(4) Division of extremum interval and calculation of closeness degree
The extremum interval composed of negative positive ideal solutions is divided into layers. namely, , , consists of and .
The closeness degree between each index and evaluation interval in standardized decision matrix can be obtained according to Equation (11):
Thus, the weighted closeness degree of each evaluation plan can be further derived:
(5) Determine the grade of each scheme
The level that the maximum of belongs to can be regarded as the evaluation grade of the scheme.
3.3. KPCA
PCA is able to merge the original features and reduce the dimension to simplify computation, especially aiming at strong linear indicators. However PCA is difficult to grab high-order features, so it ignores nonlinear information during dimension reduction. With consideration of nonlinearity in core competitiveness of clean energy incubators, kernel principal component analysis (KPCA) is taken advantage of to extract key factors in this study [
32].
This approach can validly reduce the dimension of the input in condition of retaining main nonlinear information. It is achieved by using nonlinear transformation to map initial input into high-dimensional feature space. Apparently, KPCA is able to compress the information contained in a large number of indexes into some comprehensive indicators. The basic steps are presented as follows:
Set a random vector , , m is the number of input, namely the original input can be expressed as . The dataset is projected into the space F via nonlinear mapping Φ, , .
The covariance matrix is shown as follows:
The eigenvalue and eigenvector can be obtained as Equation (15)
where
is the eigenvalue,
represents the corresponding eigenvector. The symmetric matrix
K is shown in Equation (16):
can be acquired via matrix centralization:
where
In is
n ×
n matrix,
Ii,j = 1/
n, Equation (15) can be simplified as:
Thus the kernel principal component can be calculated with reference to extraction technique of traditional PCA.
3.4. NGSA-II
NSGA-II optimization algorithm is a novel genetic algorithm on the foundation of NSGA. Despite Pareto optimum can be derived from individual classification in accordance with non dominated sorting via NSGA, it needs complicated calculation and given shared radius value. Therefore, fast non-dominated sorting technique, crowding degree and elite strategy are introduced into NSGA, namely NSGA-II, to promote the robustness and operation speed [
33].
The specific procedures of NSGA-II are described as follows:
- (1)
Generate initial population Pt and achieve non-dominated sorting as well as given Pareto optimum. The genetic operation Pt is carried out to form the offspring Qt.
- (2)
Aamalgamate Pt with Qt to generate the new species Rt. Implement non-dominated ranking on Rt and calculate the crowding degree di of ri.
- (3)
A new generation Pt+1 comes into being in line with non-dominated ranking and crowding degree di.
- (4)
If t does not reach the maximum number of iteration tmax, . Otherwise, the calculation will be terminated and the output will be obtained.
The crowding degree (
di.) refers to the density of other individuals around each personality in the population. The individual crowding degree at the boundary is defined as infinite, the crowding degree at other positions can be calculated as follows:
where
di. represents the crowding degree of
i,
equals the number of objective functions,
fjmax,
fjmin are the maximum and minimum of
j-th objective function.
3.5. LSSVM
LSSVM is an extension of SVM. This method makes use of equality constraints to replace the inequality constraints and employs kernel functions to transform prediction into equation problems, which contributes to the improvement of evaluation accuracy and speed [
34].
In LSSVM, the sample is set as
,
is the number of samples. The regression model can be established as Equation (20)
where
maps the training samples into a highly dimensional space.
and
are the weight vector and bias, partly.
The converted optimization problem is shown as follows:
where
equal the regularization parameter applicable to balance the complexity and accuracy in LSSVM.
ξi is the error.
The Lagrange function is defined to address the problem:
where
is Lagrange multipliers. In the light of the Karush-Kuhn-Tucker (KKT) conditions, Equation (24) is derived:
The optimization problem is converted into the following matter through eliminating the variables of
and
ξi:
where
The final form of LSSVM is displayed in Equation (30):
where
K(
xi, x) equals the kernel function that satisfies Mercer’s condition. The paper selects radial basis function (RBF) as the kernel function of LSSVM on account of its wide convergence range and application, as described in Equation (31):
where
is the width of the kernel width.
Evidently, the performance of LSSVM is chiefly decided by two parameters: kernel parameter and regularization parameter σ2.
3.6. Approaches of Matter-Element Extension Combined with TOPSIS and KPCA-NSGA-II-LSSVM
Based on the approach that combines matter-element extension, the core competitiveness evaluation model incorporating KPCA, NSGA-II, and LSSVM is constructed for clean energy incubators where KPCA is utilized for the determination of input and NSGA-II is exploited for parameter optimization in LSSVM. The flowchart of the novel evaluation technique is illustrated in
Figure 2.
Step 1: Parameters initialization and data preprocessing. Suppose the input
consists of the aforementioned core competitiveness evaluation indexes of clean energy incubators, as
Table 1 shows. Consistency and standardized processing are implemented on the original input
xi.
Step 2: Weight Determination. The entropy weight method is used here. Besides, evaluation results can be obtained based on matter-element extension integrated with TOPSIS.
Step 3: Input selection. Crucial factors in original input
are extracted via KPCA. Gaussian kernel function is selected for nonlinear mapping as shown in Equation (32). The kernel principal components whose cumulative variance contribution is higher than 95% form the new input matrix.
Step 4: Parameter optimization. Initialize the parameters in LSSVM and NSGA-II. Due to the influence of parameters on training and learning ability in LSSVM, this paper applies NSGA-II to optimize the two parameters, that is, and . Circulation ends at the maximum number of iteration , thus the optimal parameters can be substituted into LSSVM. Through continuous retraining and testing, the two parameters are adjusted again and the optimal core competitiveness evaluation model for clean energy incubator is established.
Step 5: Output the evaluation results. According to the characteristics of the core competitiveness level of clean energy incubators, the category labels are set as 1, 2, 3 and 4, that corresponds to the grades of core competitiveness, namely very high, high, general and low, respectively.