The Improved BP Neural Network In The Research On The Application Of Evaluation Index System

Since the computer and the emergence of the Internet, make the world since the information technology to share and rapid development of up, day of thousands of new technology, new knowledge, and the emergence of constantly being applied to different fields in. Each in the field of industry enterprise is in constant pursuit of the information technology, so as to the steps in leading position in the field of industry. So the situation this will lead to all the competition between enterprises is becoming more and more fierce. Any enterprise want to the fierce competition in steady development, must first to master its competitive level, can we take the right measures to enhance the enterprise the competitive ability. Therefore, more and more companies are paying attention to its competitive power evaluation study to establish proper evaluation index system is the basic way of research enterprise competitiveness, and so on it is particularly important.The evaluation index system is set up in the process of each index set weight, for the traditional methods of mainly has:the hierarchical analytical method, and fuzzy comprehensive evaluation method, the Delphi method, etc. The use of these methods is dependent on the experience, that is most by the influence of the artificial factor is bigger, and not be avoided. And they ignore all indexes and the competitive power of enterprise in the nonlinear relation between, so the result is difficult to reflect true enterprise competitive situation.In recent years, more and more researchers will BP neural network as a set of weights a new method. Because the BP neural network as artificial neural network of an important branch, have since learning and the organization ability. It uses the existing data records for its own network, training, through the input and output data, and of the expected output data, the relationship between the constant adjustment of network weights to complete nonlinear objective function approximation. However, due to the BP algorithm uses is the gradient descent method, forecast from a sure right value of PM, namely the local optimal search. So because of the complexity of the error function, the training is easy to converge to a local minimum points as the network weights. Although the local minimum value point also can to a certain extent target is approximated function, but it is not the best power point, which will result in the network generalization ability reduce. Therefore, the researchers began to continuously the research of the neural network optimization problems.Genetic algorithm is a kind imitates the evolution process of the natural selection and chromosome information exchange mechanism of combining the global optimal search method. That is it can with the larger probability search to the global optimal solution. So if the genetic algorithm global optimization of characteristics of fixed BP network weights with deviation, whether can avoid the BP neural network into the local minimum dilemma? Because neural network and genetic algorithm, many researchers different mechanism of its most of the ideas are about the research in their field or will the use of combination method are also different. This paper is to use genetic algorithm to optimize the BP neural network weights adjustment for the purpose, and its feasibility research, and through the empirical analysis result to the argument.The first part of this paper mainly introduces the background, significance, expatiates the neural network and genetic algorithm is proposed in this paper and the research content and method.The second part describes the BP neural network algorithm and genetic algorithm the detailed design of the detailed design. It focuses on the design of each step and on one of them involved key content discussed in detail.The third part explain the genetic algorithm to optimize the BP neural network weights method of adjusting principle and algorithm procedures, and set up optimization model.The fourth part in China's software outsourcing services industry as an example, the characteristics of the establishment conforms to the evaluation index system, based on the configuration of the improved BP neural network model of various parameters, so it can better training for example, come to the correct weights matrix.The fifth part is to establish the optimization model of the right to have value optimization ability for verification. Namely into the sample data respectively, the not improved neural network with the improved neural network is trained, get training results and with the real results were compared, and concluded that two model predictive power of the pros and cons.This paper expounds the most part of the main conclusion and further research prospects for the future.In this article, the traditional genetic algorithm to the BP neural network of weitht training is optimized, better improve the traditional BP neural network function, make it better and more accurate for each index set weight, so that the index system to more accurately reflect the enterprise the competitive ability.To sum up, the genetic algorithm to search out through its global optimal solution and has strong robustness, can avoid the characteristics of the traditional BP neural network is easy to fall into the local minimum faults. But, for the two network itself need parameter setting, so far no specific rules and standard to measure quality of parameters, but also to the evaluation results have great influence, so we can only through the practical problem to determine the network parameters. So how to find out the regularity of the set parameters remains to be to further research.