Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Computer network wikipedia , lookup
Distributed firewall wikipedia , lookup
Internet protocol suite wikipedia , lookup
Airborne Networking wikipedia , lookup
List of wireless community networks by region wikipedia , lookup
Network tap wikipedia , lookup
Recursive InterNetwork Architecture (RINA) wikipedia , lookup
Research on Performance Evaluation Model Based on Improved Neural Network for Electric Power Enterprises1 YANG Shao Mei, SHEN Peng Economics and Management Department, North China Electric Power University, P.R.China, 071003 College of Economics and Trade, Agricultural University of Hebei, P.R.China, 071001 Abstract In the modern enterprise management practice, performance evaluation plays an increasingly important role. China's existing performance evaluation methods still exist on certain issues, in order to improve the business operations and management efficiency, this paper established an effective performance evaluation system as a management and control methods, and which can judge and measure the company's operating activities better. Keywords Performance evaluation, Improved BP neural network, Electric power enterprises 1 Introduction Employee distribution, performance evaluation, personnel training and incentive policy is the four critical systems to guarantee human resources management effectiveness. In the four systems, the performance evaluation is the most important, because it is the foundation of the other three systems. Appropriate performance evaluation system can play a role of the contract, it clearly demonstrates the organization's requirements to the staffing; appropriate performance evaluation is equivalent to an ombudsman, which may examine each employee's work effectiveness. Therefore, it is a kind of control system, according to the performance evaluation standards, enable managers to identify what kind of employees can mount guard, what kind of employees can stay and what kind of employees should be laid off. Similarly, performance evaluation as a control system plays a very crucial role for the staff's self-control. Performance evaluation can provide a reliable basis for the staff's promotion, demotion, job transfer, salary increase and dismission etc. The performance evaluation is a systematic evaluation process. The performance evaluation methods have been widely applied, such as: mutation base method, multi-objective decision analysis method, etc. However, these methods are subject to stochastic factors in the evaluation, and the evaluation results are influenced by subjective experience and knowledge limitations easily. In recent years, with the rapid development of the neural network that has the unique advantages, that is, self-learning, self-organizing and self-adapting ability, it can overcome the influence of subjective factors and has been applied widely. This paper will use the improved BP neural network to evaluate the electric power enterprises performance. 2 The Improved BP Neural Network Principle ∈ 2.1 The parameters determination of the improved BP neural network The input vector in the input layer is: X Rn, X=(x0,x1,x2…,xn-1)T; the input vector in the first hidden layer is: X' Rn1, X'=(x'0,x'1,x'2…,x'n-1)T, output vector is: Y' Rm1, Y'=(y'0,y'1,y'2…,y' m1-1)T; the input vector in the second hidden layer is: X'' Rn2, X''=(x''0,x''1,x''2…,x''n2-1)T, output vector is: Y'' Rm2, Y''=(y''0,y''1,y''2…,y''m2-1)T; the input vector in the output layer is: X''' Rn3, X'''=(x'''0,x'''1,x'''2…,x'''n3-1)T, output vector is: Y Rm, Y=(y0,y1,y2…,ym-1)T. The weight value between the input layer and the first hidden layer is denoted with ωij, the threshold value is θj, the weight value between the first hidden layer and the second hidden layer is denoted with ω'jk, the threshold value is θ'k, the weight value between the second hidden layer and the final output layer is denoted with ω''kl, the threshold value is θ''l, then the ∈ ∈ ∈ 1 ∈ ∈ ∈ The research was supported by the Scientific Research Foundation for Young Teachers of North China Electric Power University. The item No. is 200611034. 1070 input and output in every layer neurons are met: m2 −1 m1 −1 n −1 x ' j = ∑ ωij x i − θ j x ''k = ∑ ω ' jk y ' j − θ 'k x'''l = ∑ ω '' kl y ''k − θ ''l j= 0 k =0 , , y ' = f (x ' ) y ''k = f k (x ''k ) , yl = f l (x'''l ) , j j j The activation function f (u) is a nonlinear function. We normally choose sigrnoid function, 1 . f (u) = i =0 1 + e−u 2.2 The learning process of the improved BP neural network The study process of the BP neural network can be divided into two stages: the first stage is toward pass, namely input sample data from the input layer, transmit signal forward, and calculate the corresponding output neuron of every layer according to the above formula that is not feedback and connected among the layers. The second stage, namely the back pass, If the error between the actual output of the output layer and the goal output doesn't fall into the scheduled precision range, the error signal returns along the original pathway to amend the weight value and threshold value. By the two iterative processes, we make the network to achieve convergence at last. At this time, the error reaches the scheduled range. In the multilayer BP neural network, given a set of sample data (x,t), X Rn, t Rm, when input Pith sample, namely (xp1, tp1), we can calculate the network output yp1 Rm, relative to the xp1 using the improved BP algorithm, then the error function is defined as: 2 1 m−1 ε = ∑ (t pl − y pl ) 2 l= 0 To all samples, the total network error: ∈ ∈ ∈ 2 1 p m −1 E = ∑∑ (t pl − y pl ) 2 pl =1 l= 0 We enter into the second stage of the learning process, namely adjustment reversely the weight value and the threshold value, and revise every weight value ωnq: p ∂E ∂ε , η is the learning rate. ) = −∑η ∆ωnq = −η ( ∂ωnq ∂ωnq pl =1 The learning process of the neural network is the process to seek the smallest of error E, but the traditional BP algorithm to complex networks, it very likely falls into local minimum value. This paper introduces the Levenberg-Marquardt optimization algorithm which can enhance the network convergence speed and reach the error range rapidly. 3 Improved BP Neural Network Method Construction 3.1 The comprehensive evaluation index systems 3.1.1 Asset management performance index We can evaluate the asset management performance for the electric power enterprise from the perspective of the asset management capability, asset strength and enterprise growth ability. The evaluation index including: total asset turnover (U1), current asset turnover (U2), total asset growth rate (U3), equipment' advanced degree (U4), knowledge capital proportion (U5). 3.1.2 Human resources management performance index Human resources management performance evaluation mainly includes the quality of human resources management, human resources management activity and human resource trends. The evaluation index including: personnel labor productivity (U6), the average education level (U7), staff turnover (U8), basic quality and stability of leadership (U9), human resources training rate (U10). 3.1.3 Financial performance index In evaluating the financial performance, we can evaluate from the profitability, solvency and business development capacity. The evaluation index including: debt-to-asset ratio(U11), total assets 1071 return rate(U12), net rate return rate(U13), sales growth rate(U14), have gained interest multiples(U15). 3.1.4 Culture performance index This paper mainly from three areas to assess the enterprise culture performance: enterprise culture status, enterprise overall culture quality and the enterprise attractive. The evaluation index including: enterprise management philosophy (U16), enterprise values (U17), enterprise synergy (U18), staff team spirit (U19), enterprise culture input rate (U20). 3.2 Improved BP neural network model construction We take the 20 indicator of describing the enterprise performance as the input vector, and take the corresponding comprehensive testing results as the network expectation output. We take enough samples to train the network, make the relative error to meet the scheduled accuracy after ceaseless learning process. At this time the weight value and the threshold value hold by the neural network is the correct internal denotation acquired by the self-adaptive learning. Once the network has been trained, it could serve as an effective tool to evaluate the performance. (1) S.K.Doherty and other scholars' studies have shown that three-layer feedforward neural network, namely the neural network only contains one hidden layer can approximate any nonlinear function relation with any accuracy. So we set up a three-tier feedforward neural network, the input layer neuron number is the above 20 indicators, the determination of the hidden layer neuron number has not to reach a unified theory yet. According to the experience, the node number of the hidden lay should meet 2n>m (m denotes the input layer node number). Therefore, we select 7 network neurons for the hidden layer, and the neuron in the output layer is only one, namely the power enterprise performance comprehensive value. (2) Network parameters initialized: we endow with the link weight value ωij and the threshold value θj between the input layer and the hidden layer, the link weight value ω'jk and the threshold value θ'k between the hidden layer and the output layer. (3) Select a tier model randomly as the input signal. (4) Calculate the input x'j and the output y'j of the hidden layer neurons. (5) Calculate the input x''k and the output yk of the output layer neurons. (6) Calculate the general error uk of the output layer neurons, judge uk whether to meet demands, if met to step (9) and not met to step (7). (7) Calculate the general ion errors of the hidden layer neurons: m −1 v j = [∑ (u kω ' jk )]f ' j (x ' j ) k=0 (8) The amending weight value and the threshold value: ω ' jk (N + 1) = ω ' jk (N) + ∆ω ' jk (N) , ωij (N + 1) = ωij (N) + ∆ωij (N) θ 'k (N + 1) = θ 'k (N) + ∆θ 'k (N) , θ j (N + 1) = θ j (N) + ∆θ j (N) (9) We take the next tier model as the input signal so as to all the training models train a circumference, until the total error reaches the scheduled accuracy. The learning is terminated; otherwise we update the study frequency, and then return to training again. 3.3 Emulate experiment We take the performance evaluation based on the improved BP neural network of 12 electric power enterprises as an example, which are shown in table 1. We use the MatLab to realize the software program, establish the three-layer BP neural network structure of performance evaluation, the given study accuracy ε = 0.0001, and we select 7 network neurons for the hidden layer. We take 1-8 group evaluation data and evaluation results in table 1 as the training set, train the network, and carry through the simulation evaluation using the evaluation indicators data of the four residual groups and the trained network. In table 2, the network training results and the actual comprehensive evaluation results are shown. The simulation results about the 4 test sets and the actual evaluation results, as shown in table 3. The results in the table 2 and table 3 show that not only all the training samples is very close to the actual evaluation value, but the results of the four simulation test sets is also very close to the actual evaluation. 1072 Table 1 Expert evaluation data No. U1 U2 U3 U4 U5 U6 U7 U8 U9 U10 U11 1 0.5 0.5 0.5 0.7 0.7 0.5 1 0.7 1 0.7 0.7 2 1 0.7 1 1 1 1 1 0.7 1 1 0.7 3 0.7 0.7 0.5 1 0.7 1 1 0.7 1 1 0.7 4 0.7 0.7 1 0.7 0.7 0.5 0.7 0.5 0.7 0.7 0.7 5 0.7 0.7 1 0.7 0.7 0.5 1 0.7 1 0.7 0.7 6 0.7 1 1 0.7 0.7 1 0.7 0.7 1 1 0.7 7 0.5 0.7 0.5 0.7 0.7 0.5 0.5 0.7 0.5 0.7 0.7 8 0.5 0.5 0.5 0.5 0.7 0.3 0.3 0.5 0.5 0.7 0.7 9 0.7 0.7 1 1 0.7 0.5 1 0.7 1 1 0.7 10 0.7 0.5 0.5 0.7 0.7 0.5 0.7 0.7 0.5 0.7 0.7 11 0.7 1 1 1 0.7 1 1 0.7 1 1 0.7 12 0.7 0.7 0.3 0.5 0.3 0.5 0.5 0.5 0.7 No. U12 U13 U14 U15 U16 U17 U18 U19 U20 Score 1 1 1 0.7 0.7 0.7 0.7 0.7 0.7 0.1 0.7130 2 1 1 0.7 1 0.7 1 0.3 0.7 1 0.9310 3 1 1 0.7 0.7 0.7 0.7 0.7 0.7 0.5 0.7660 4 0.5 0.7 0.7 0.5 0.7 0.7 0.7 0.7 0.5 0.6830 5 1 1 0.7 0.7 0.7 1 0.7 0.7 0.1 0.7270 6 1 0.7 0.7 0.7 0.7 1 1 1 1 0.8610 7 0.5 0.7 0.5 0.7 0.7 0.7 0.7 0.7 0.5 0.6040 8 0.5 0.3 0.5 0.7 0.7 0.7 0.7 0.3 0.5 0.4880 9 1 1 0.7 0.7 0.7 1 0.7 1 0.5 0.8270 10 0.5 0.7 0.7 0.7 0.5 0.7 0.7 0.7 1 0.6470 11 1 0.7 0.7 0.7 0.7 0.7 0.7 1 0.5 0.8170 12 0.3 0.5 Continued table 0.5 0.3 0.5 0.7 0.5 0.7 0.3 0.3 0.5 0.4600 Table 2 The actual evaluation results compared with the network training results and the taxis No. 1 2 3 4 5 6 7 8 Actual evaluation results 0.7130 0.9310 0.7660 0.6830 0.7270 0.8610 0.6040 0.4880 Network training results 0.7185 0.9193 0.7567 0.6890 0.7196 0.8612 0.6029 0.4894 5 1 3 6 4 2 7 8 Actual evaluation results Taxis Network training results Taxis 5 1 3 6 4 2 7 Table 3 The actual evaluation results compared with the simulation results and the taxis 1073 8 No. 1 2 3 4 Actual evaluation results 0.8270 0.6470 0.8170 0.4600 Simulation results 0.8289 0.6511 0.8169 0.4509 Actual results taxis 1 3 2 4 Simulation results taxis 1 3 2 4 4 Conclusion The performance evaluation is associated with many factors, it need large numbers of statistical calculation, and the factitious factors can be mixed into easily, which make the performance evaluation work is difficult. In this paper, we build the improved BP neural network model, which can not only overcome the limitations of the traditional evaluation methods and avoid the human errors in the evaluation process, but also enhance the learning accuracy and the algorithm convergence speed greatly. References [1]Tim Hill, Leorey Marquez, marcus O'Ionnor, Artificial neural networks forecasting and deciding model. Internation journal of forecasting, pp.5-15, Oct. 1994. [2]BEI Jin-lan, WANG Chun-yu, Enterprise performance evaluation system construction. Statistics and decision, no. 12, pp.124-125, 2004. (in chinese) [3]QIU Zhi-yong, Construction of enterprise performance evaluation system. Cooperation economy, science and technology, no. 11, pp.26-27, 2006. (in chinese) [4]KONG Jun-hua, FENG Yan, Construction of China's enterprise performance evaluation system. Market modernization, no. 4, pp.69, 2006. (in chinese) The author can be contacted from e-mail : [email protected] 1074