Evaluation of Enterprise Performance Based on FLI-GA Model

Purpose: There are many kinds of methods to evaluate the performance of enterprise, but they still have some distinct shortcomings. In order to achieve a better evaluation result, we put forward a new model named FLI-GA (Fuzzy logic inference & Genetic algorithm). Design/methodology: This model, mainly based on the fuzzy logic inference method, uses fuzzy rule-based system (FRBS) to avoid the drawbacks of FRBS. Genetic algorithm is applied in this model. Findings: FLI-GA model can be used to evaluate certain enterprise performance, and its evaluation results are more accurate than fuzzy logic inference method. Originality/value: This model combines the genetic algorithm with the unclear reasoning methods so as to make the appraisal results more reasonable and more satisfying.


Introduction
Evaluating the enterprise management performance in a scientific way is helpful to obtain accurate guidance of enterprise management behaviors.This evaluation is essential to improve the management methods of enterprises.Besides, this kind of evaluation helps to strengthen the innovation of enterprise management system, as well as to enhance the achievement inspection efficiency of the operators.Such evaluation can assist the establishment of the encouragement and restraint mechanism.On one hand, it is conductive to the enhancement of the image consciousness and competitive power of certain enterprises.On the other hand, simultaneously, it also provides the basis for the macroeconomic regulation and the formulation of the economic policies as well.Therefore the evaluating performance of enterprise is playing a more and more vital role in modern enterprises.

Brief Definition of Enterprise Performance Evaluation
Enterprise Performance evaluation can make objective and fair judgment of an enterprise's operational effectiveness in a certain operating time.This will help to guide this enterprise to promote its reform and innovation, thereby enhance the competitiveness of this enterprise.
Modern enterprises cannot neglect the important role played by performance evaluation.
As a multi-index comprehensive evaluation, the popular enterprise performance evaluation methods mainly include Analytical Hierarchy Process (AHP), artificial neural network (ANN) evaluation method, grey correlation method, fuzzy comprehensive evaluation method, genetic algorithm, factor analysis, TOPSIS and so on.

Literature Review of Enterprise Performance Evaluation
After the born of modern enterprises, ownership rights was separated from management rights.In order to protect their own interests, owners of capital generally assess operators' subjective effort and their operating results.Over the years, a variety of evaluation thinking conceived and became more ripe and satisfactory.There are 3 comprehensive evaluation methods from the view of corporate finance: • Dupont Financial Analysis System.This system attaches great importance to integrity of management and operations, which also outstands the coordination and multi-layer of management.
• Wall Marking Way.Irrelevant financial indexes are assigned with different proportion, making it possible to comprehensively calculate the financial situation of enterprises.
The deficiencies of this method are unreasonable score calculation method and inadequate indexes' proof.
• EVA Evaluation System.EVA indexes overcome the traditional profit indexes' ignorance that the value of corporate assets will change over time, as well as the opportunity cost of owner's equity.Although it has a great progress, EVA evaluation system still has limitations, such as financial-oriented and short-term-oriented.
As is shown in the former section, there are also some popular methods and their improved models.
The advantages of AHP method (Golam & Akhtar, 2011;Bi & Zhang, 2010) is its extremely precise value, higher scientific basis.The more obvious shortcomings are its poor operability, the more artificial color when determine the relative weights, and less intuitive.The neural network method has strong robustness, memory capability, nonlinear mapping ability and a strong self-learning ability, but has no ability to explain its reasoning process and reasoning basis.
Artificial neural network (Lou & Kuang, 2010) has self-learning function, association and storage capabilities and high-speed ability to find the optimal solution.But its architecture is not versatile, and it is difficult to accurately analyze the performance of the neural network.
The gray relational analysis (Yang, Zhang, Wang, Wang & Zhang, 2010) requirements are lower and have less calculation.But the process of computing weights has poor objectivity, which does not meet some of the more important samples.
Fuzzy comprehensive evaluation (Ma, He & Shuai, 2011) method's results are clear and have systematic characteristics.This method can solve the problems of fuzzy and difficult to quantify, so it is suitable for a variety of non-deterministic problems.The disadvantages of this method are its duplication of information, subjectivity, and the difficulty of determine the membership function.
Genetic algorithm (Iraj, Mahyar & Fereydoun, 2011) has good convergence in the calculation of accuracy requirement, less computation time, high robustness.By contrast, the drawbacks are that sometimes it may be precocious and needs large amount of computation.Besides, it has poor stability and small data scale.
Factor analysis (Patrick, Douglas & Fabio, 2010) can use computer software to solve problems quickly and easily.Therefore, compared with other methods, factor analysis is a scientific, practical, simple and comprehensive evaluation method.However, Factor analysis is just a kind of financial analysis method which is only on the basis of financial indicators, macroeconomic factors and some special cases.
TOPSIS (Saeed, Mehdi & Mostafa, 2012) method is an ideal target sequence optimal selection technology; it's a very effective method in multi-objective decision.But the drawback of subjective weights and reverse problem limit its application.

Introduction of Fuzzy Logic Inference Method
In the process of evaluating the enterprise performance, the subjective factor of human is uncertain, which makes it difficult to compare the values of different indexes.Even the experts in this domain can hardly give exact weight of each individual index.Moreover, weight is dynamic and alterable, so the value of index's weight maybe improbable.How to deal with this problem?The answer is fuzzy logic inference method (Rustum, Omar & Emad, 2009).Fuzzy logic inference method can reflect the expert's correct judgment, so as to avoid difficulties of giving weights directly; in addition, as long as the satisfaction of the membership function (fuzzy inference rules) level is more than enough, experts' judgment can approximate the actual situation.
On account of the above advantages, our new model will be improved basing on fuzzy inference method.The fuzzy logic inference method uses fuzzy rule-based system (shortened form FRBS).It allows assessors to divide a complex system into parts in order to easily handle this system.FRBS can also use the data of various types (qualitative or quantitative data) and it allows a small amount of measurement error.

Evaluation Steps of Fuzzy Logic Inference Method
After the analysis of former section, we can refine the reasoning steps of the fuzzy inference method.The evaluation process of fuzzy logic inference method contains 7 steps: • First substitute the indexes scores into the predetermined membership function, and then get membership of each index (the fuzzy set 1) through membership function.
• Memberships are compared with the corresponding rule base.The minimum principle should be adopted during above process.
• Use the maximum principle to contrast results in order to screen better membership after the comparison within the rule base, and thus get the memberships of the middlelevel evaluation indexes-fuzzy sets2.
• Fuzzy set 2 is compared with related rule base.Then we repeat step 2.
• We use the same method as step 3 to get the membership degree of the high tier evaluation indexes-fuzzy sets3.
• The fuzzy sets of major and secondary factor are both substituted into the rule base to compare the two fuzzy sets with their related rule bases.The same maximum and minimum principles are applied as previous steps.At last, the final evaluated fuzzy sets are given out.
• Compare the degree of fuzzy sets' memberships which vary from well, ordinary to bad.
According to the maximum principle of membership, the final analysis conclusion is obtained.

Shortage of Fuzzy Logic Inference Method
Fuzzy logic inference method introduced above first uses triangular membership function in fuzzy mathematics to calculate the membership of the underlying index, and then use the rule base layers to get new membership values of upper tier's indexes.This method makes the calculus process inevitably rigid, thus the result is singleness to the rate for the same group, so that it is not conducive to the latter part of the analysis and further assessment work.

Building the FLI-GA model
We propose an improved model based on the fuzzy logic inference method and genetic algorithm method-FLI-GA model (Fuzzy Logic Inference & Genetic Algorithm Model).After calculate the lowest tier indexes' values, the improved FLI-GA model makes use of genetic algorithms to generate offspring and choose the best individual.The calculation process is relatively flexible and easy to understand, and its multiple results obtained from repeated computing can be applied for post-analysis and selection.

Brief introduction of Genetic Algorithm
Genetic Algorithm (GA) (Pond, Posada, Gravenor, Woelk & Frost, 2006) is a class of random search method.It learns from biological laws of evolution (survival of the fittest) and is evolved from this law.It was first proposed by Professor J. Holland from the United States in 1975.GA's main feature is that it directly operates the objects and there are no limits about derivation and function continuity.GA has inherently implicit parallelism and better ability of global optimization.Besides, its probabilistic optimization method can automatically access and guide the optimal search space.It can also adaptively adjust the search direction, and need not to decide the rules.These properties of GA have been widely used in combination optimization, machine learning, signal processing, adaptive control, artificial life and other important fields.

Build index system of FLI-GA model
The relate studies of enterprise performance evaluations (Zhang, Xie, Cao, 2011;Zhang & Tan, 2012;Qi & Sun, 2011) are only focus on the cost information and financial information.This kind of models ignores the quality and sense of time; neither do the non-financial information like human capital and so on.To avoid the drawbacks of traditional enterprise performance evaluation, new research achievements are included in our study.By considering the normal principle of index selection and characteristics of enterprise performance evaluation, we initially screen 13 indexes from the views of financial, business growth ability, internal management and workforce capability.The indexes are shown in Table 1.The final index system is built after the initial indexes were selected.Some standards are implemented when select the final indexes: • The selected indexes should reflect two factors: absolute and relative number; • The selected indexes should be simple, clear and practicable; • It is necessary to select the quantitative as well as qualitative indexes; -378-Journal of Industrial Engineering and Management -http://dx.doi.org/10.3926/jiem.959 • Current corporate outcome indexes and strategic targets should be both considered.
Based on the principles above and the indexes in Table 1, a new index system is screened out, which is shown in Figure 1.To clarify this further, the final index system will be divided into four parts to be explained: • Asset operational status: whether a company is profitable or not is the key factor of enterprise performance.Operation of assets mainly includes current assets turnover and total assets turnover status.
• Business growth ability of an enterprise decides its fate.Marketing ability relates to the enterprise market share and product awareness.Good corporate culture helps to establish a good public image and corporate reputation, and gives great incentives to subordinates.
• Cumulative employee productivity is just the overall efficiency of an enterprise.
• The organizational policy of an enterprise influences the enterprise's internal resources configuration.
Strictly speaking, every enterprise should have a very individualized performance evaluation system.This index system is designed on the basement of the commonality of general enterprises.In the practical application, evaluation system should be appropriately changed according to the enterprise that is evaluated.

The algorithm of FLI-GA model
The algorithm of FLI-GA model will be introduced after building the evaluation index system.

The determination of the lowest tier indexes' membership functions
There are three common methods (subjective experience method, analytical reasoning method, survey & statistics method) to determine the membership function.
According to the nature of the problem, analytical reasoning method applies some definite analyses and reasoning method to select typical function as membership function in the continuous domain.Furthermore, several principles should be considered when selecting the membership function: Membership functions should be simple; Meet the convex fuzzy set principles; Select less evaluation indexes and rules to reduce complex calculations; Try to satisfy the requirement of non-overlapping membership; Describe the transition relations intuitively among the standard values from different reviews.
So according to these regulations, triangular membership function is selected as it fits the experience and principles above.We respectively select a1 、 a2 、 a3 as 0, 2.5, 5 (the index variation range is [0,5]) and get triangular membership function of the bottom indexes of FLI-GA model.The membership function is demonstrated as follows: (1) (2) (3) After the establishment of triangular membership function, we built the regulation base for model application according to the expert experience.
Bottom tier reasoning rules will be applied to get indexes' reasoning output of the Fuzzy logic inference model and improved FLI-GA model.The rule base of the upper index applies only to the upper index of the Fuzzy logic inference model, while FLI-GA model uses improved reasoning mechanism based on genetic algorithm.The reasoning rule base of the upper indexes will be applied in the next section to realize the simulation comparison between the two models.

Heredity and variation of the upper indexes and their individual fitness
Based on Genetic Algorithms, we select three membership values from bottom tier as the male parent of the three fitness: fitnessbad, fitnessnormal, fitnesswell.Their membership values are the same and fitness values vary from 0 to 1. Particularly, when one of the male parent' value is zero, we choose another paternal fitness value as the corresponding fitness value of its offspring.We marked the selected couple of male parents as A1 and A2.During the first genetic round, their corresponding fitness values were set as fitness1 (t) , fitness2 (t).So we could calculate the probability of the selected allele: (5) Then we selected allele through A1 or A2 according to p1 and p2, the specific operation is explained as follows: Set i=1 ， 2 ， 3 … L ( L is the string length of the individual).Firstly, generate a sequence ai (i = 1, 2, 3...) which varies from 0 to 1. Then compare a1 with p1 or p2 to determine the selected allele.Here, one should notice that when p1≤p2, if a1≤p1 then select

The Input data of two models
Enterprise Z regularly invites experts to assess its performance, and each expert scores the indexes back-to-back to ensure fairness and justice.We selected the experts' scores of this enterprise in 2010 and 2011, as are shown in Table 2. Based on Table 2, we average the experts' score to get the value of indexes.The results are shown in Table 3.
of genetic algorithm is mainly used to determine an optimal evaluation standard.In addition, different theories play different roles in this evaluation model.We use genetic algorithm to refine inference rules and memberships of fuzzy reasoning method.By using genetic algorithm, the inference rules and memberships become more convincing.The results of simulations indicate that FLI-GA model is objective and accurate.
Though we have achieved some positive results, there are still some shortcomings in our research that need to be improved: • The rule base needs to be improved continuously.With the development of society, the status and connotation of enterprise performance evaluation indexes are changing, so more perfect index system should be put forward.Besides, rule base of this research is static, while practical applications need dynamic rule base, therefore, we should try to establish dynamic rule base in further studies.Moreover, the convergence properties of genetic algorithms can be used on the previous studies to summarize and temper and this will make the rule base more objective and accurate.
• In FLI-GA model simulation, we haven't taken gene mutation into account.For the gene mutation is widespread in nature, and to make the model more reasonable, we should take gene mutation into consideration.

Figure 1 .
Figure 1.Index system after screening

Figure 2 .
Figure 2. Membership function of bottom index

Figure 3 .
Figure 3. Computing the bottom index value of FLI-GA model

Table 1 .
Primary indexes of enterprise performance evaluation