Research On Scientific And Technical (S&T) Credit Assessment Of Private S&T Enterprise Based On Neural Network

As the social economy is developing, our country more and more regards the development and innovation of S&T importance. It encourages enterprises to participate in research and development (R&D) so that enterprises become principal subject of technical innovation. Private S&T enterprises are new economic enhancing power. Consequently, the troop of private S&T enterprises is stronger and stronger. the whole national S&T power is increasing, However, there are a few enterprises that don't develop completely and their R&D capability is not enough. How to discriminate the level of them depends on the problem evaluating their S&T credit. The effective S&T project implement is related to the capability and quality of the enterprises applying S&T projects. It objectively demands a set of management system, assessment system to manage and evaluate the S&T credit. Because private S&T enterprises have unbalanced development so that a set of method fitted to S&T credit assessment must be found How to assess the S&T capability of enterprises is the core of this thesis.It's also a hotpot for national S&T project credit management how to applying scientific method to assess the S&T credit objectively and effectively. This thesis mainly sums up some kinds of credit or credit risk assessment models and analyzes the mean and method. This research applies ANN technology to found the S&T credit model connected to the feature of private S&T enterprises and studies the real situation. Research Team did some investigation about 'Guangdong private S&T credit assessment system'. It takes 59 enterprises as a research object. According to the contents of the poll, the status of Guangdong S&T enterprises developing and their S&T credit status are analyzed. The S&T credit model of Guangdong private S&T enterprises is created using ANN technology with its feature self-study, self-organization and extension. The model have been trained and tested. Simulation research is created using Matlab7.0 software. 40 enterprises is selected randomly in the 59 to train the model and the 19 remained is used to test it. In the research, study algorithm, the amount of the hidden layer neural unit and training epochs are studied and tried many times for the predicting capability of the model. The improvement of enhancing the study speed is proposed also. Nice studying effect and predicting effect are gained. The result reveals the ANN method is easier than traditional method to construct model that adapts the situation. The research on this project has practical application and spread value to management of S&T project and project implement.