BP neural network on power enterprise Informatization Evaluation Index System

Abstract: The correct analysis and evaluation of the level of enterprise Information power, the power industry for improving the overall level of Information, enhance the competitiveness of power enterprises has important practical significance. Through the application of BP neural network theory, the power level of enterprise Information comprehensive evaluation, to establish the mathematical model of the evaluation system. The model convergence speed and ability to adapt to the network is feasible and applicable. The experimental results show that: The model overcomes the level of Power Enterprise Information complicated evaluation process, with easy accurate and reliable. Features fast, accurate identification.

Paper Keywords: BP neural networks, power companies, the level of Information, evaluation, index system
0 Introduction
Power industry is the application of information technology one of the earlier industry has experienced a production process automation, information management and other construction stage. Now showing the power of information infrastructure in place, data and numerous widely used features, from the quantization areas to improve the quality level. the power of information technology for effective business evaluation and management, is to raise the level of information and implementation of enterprise information important guarantee for sustainable development.

How to actively carry out information technology to reduce operating costs? Scientific evaluation of indicators by which China's power enterprises of information technology development? This is the current power industry must solve a problem, and now does not have a complete power of enterprise information Evaluation Index System of the official release. by building scientific, practical and effective evaluation of power level of enterprise information system, using a learning and memory, induction, tolerance and self-learning and adaptive ability of BP neural network algorithm, scientific, effective, objectively evaluate the level of power enterprise information will help standardize and improve the electricity enterprise information construction, and promote the health power of information and rapid development.

1 Power Enterprise Informatization Evaluation Index System
1.1 The principles established
(1) simple scientific principles. Evaluation index system should clearly reflect the power of enterprise information and indicators of the level of the relationship between the indicators related to avoid inclusion of the size of the indicator system should also be appropriate. If the evaluation index system is too large, index levels over more targets too small, then the evaluator is bound to attract the attention of a small issue, but if the evaluation index system is too small, index level is too small, too rough indicator, it can not fully reflect the power of enterprise information and evaluate the overall design and usage.

(2) just and equitable principles. That evaluation should be objective, fair and reasonable to reflect the level of enterprise information power dynamic.

(3) easy to operate principle. In the practical application of evaluation index system should be operable indicator means clear, reliable, easy data collection, did not understand index system for the personnel to operate the process and application.

(4) quantitative indicators, supplemented by some qualitative indicators. Index-based quantitative indicators as far as possible, quantitative indicators but also all use electric power companies can not fully reflect the overall situation of the level of information, so be supplemented by some descriptive of qualitative indicators.

Contents Index System 1.2
The level of power enterprise information integration evaluation index system of principles, to build from three levels of the level of power enterprise information evaluation index system, mainly by the level of business support, IT performance levels, information technology, IT management, IT sustainable development capacity the five major aspects, as shown in Table l.


1.3 Determination of indicator values and the normalized
55 in the above three indicators, there are qualitative and quantitative indicators of the points. According to the evaluation criteria for indicators can be divided into 3 categories of indicators: indicators of positive and negative indicators and optimize the index. Due to the different indicators reflect different aspects power level of enterprise information, and because between indicators of different dimensions, it can not be compared. Thus, in order to facilitate determination of final evaluation values, each index needs to be dimensionless, ie, evaluation, normalization, normalize in order to eliminate the impact of index dimension. Considering the convergence problem of neural network training, all indicators of sub-3 treatment normalized the situation.

(1) qualitative indicators. Qualitative indicators construction engineering, information security, information security degree of perfection of the system, the implementation of information security systems, enterprise IT staff qualities. Evaluation of the value of these indicators used to evaluate the expert scoring method The value is 0.0 ~ 1.0.

(2) quantitative indicators positive. Is a better indicator of the larger index values, including: equipment operation, safety run-time, material supply protection rate, labor productivity, network coverage, Unicom rate. Because of such indicators bigger and more Well, it used the maximum value for all power enterprises the ideal value for the index, the non-dimensional treatment.

(3) negative quantitative indicators. Is an indicator of its value as small as possible, including procurement costs, production costs, the average response time. Such indicators are as small as possible, therefore, choose the minimum value of all electricity companies ideal value of the index, and the non-dimensional treatment.

(4) optimization index. The indicator with an optimal range, not too big or too small for good indicators, including the power of enterprise assets-liability ratio indicator, that if too much invested enterprises will be in the information there will be insolvent circumstances, is not conducive to the development of power enterprises, if the index value is illustrated in a small building in the enterprise information does not play limited capital value. generally take 40% of the target to 60% more ideal, then dimensionless. dimensionless as follows:
(1)-dimensional transformation to non-dimensionless. To take a second large-scale distribution of parabolic pArtial description of the mathematical model:

(2) dimensionless index treatment. To take a linear increasing function description:

2 Power Enterprise Informatization Evaluation of Neural Network Expert System
2.1 BP neural network principle
Artificial neural network (ANN) is composed of a large number of simple nonlinear processing units, adaptive, self-organization system, which is in modern neuroscience based on the results, try to the nervous system by simulating the human information processing, memory and processing methods, to design a style with a human brain information processing system. It can be widely used in prediction, classification, pattern recognition and data processing process control and other occasions, as opposed to traditional data analysis and processing methods, but also suited to deal with vague, non-linear and pattern characteristics are not specific issues.

BP neural network is one-way transmission of multilayer feedforward neural network, the network can be divided into input layer, middle layer (hidden layer) and output layer, in which the input and output have only one layer, middle layer can have one or more layers layer. the same level no connection between network nodes, each network node represents a neuron, its transfer function Sigmoid type commonly used functions. each of the connection between neurons has a weighting factor W, which can enhance or reduce the output of a neuron on the next neuron stimulation. The weighted coefficient is usually referred to as weight, weight change rule called the weight algorithm. based on BP neural network expert system based on the algorithm under certain By learning the sample data to determine network weights. neural network expert system to determine the weight, structure and stability, you can handle the new data, given the corresponding output.

2.2 BP Network Based on Power Enterprise Informatization Evaluation of the learning process
BP neural network power level evaluation of enterprise information model, the input layer contains 55 neurons were treated with 55 Power Enterprise Information Evaluation of three indicators of the level of the sample data input, the middle layer contains 26 neurons, output layer 1 neurons, information is power enterprise level evaluation results, the corresponding BP network structure is shown in Figure 1.

Links to download free paper http://www.hi138.com the initial layers of the neural network connection weights are arbitrary, we must first train the neural network so that the power of enterprise information evaluation results of the actual output level with the expected value of the deviation as small as possible. BP neural network through training to learn the true value of the sample and the error back-propagation network output to the layers of neurons, continue to adjust the gradient descent method the weights of each layer, reducing the result of the right values of bias, so that real output and the network of training samples the output of the error control error in the setting of the 0.001 range. specific BP neural network learning process is as follows:
(1) According to the power level of evaluation of enterprise information requirements of training set. Selection of the enterprise information network output power level that affect three indicators of x1, x2, ... ..., x55 as input variables in order to determine Enter the number of nodes (the network has 55 input nodes),
(2) to initialize. Home Ownership is random arbitrarily small, the learning accuracy given a 10 £ objective error is 0.001, the initial weights read into the network and the learning samples. There can be 300 employees of power companies carry out survey, the first 100 randomly selected group of records (sample number 1 to 100) as the neural network identification model of training samples
(3) train the network by BP algorithm. Learning process shown in Figure 2,

(4) to determine whether the learning precision requirements, such as implementation of reached into the next step, otherwise go back and continue to learn,
(5) to store and output weights. Using the trained network to test (using 10 sample records for example), the output power level of enterprise information final result.

3 Experimental results and analysis
BP neural network using the information on the level of power enterprises to identify, input layer, hidden layer and output layer nodes were 55 �� 26 �� 1. Based on experience and testing, the first group of records as a study of 100 samples for training neural element with the connection weights, learning precision £ = 1 �� 10, after the 10 groups (sample number is 291 300) samples used as the test examination. After repeated learning, the learning outcomes (test) as shown in Table 2.


4 Conclusion
BP neural network based on information level evaluation of power enterprises to take full advantage of information and data about the sample, through a highly nonlinear mapping, revealing the level of power enterprise information and its influence factors inherent mechanism to overcome the power enterprises Evaluating the information modeling and solving difficult problems, weaken the power of enterprise information to determine the level of the process of evaluation index weights the influence of human factors to ensure the level of power enterprise information evaluation results of the objectivity and impArtiality. Links to free Download Center http://www.hi138.com paper