Study On Evaluation And Sensitivity For The Power Transmission And Station Engineering Cost Feature Showing Small Sample

As the scale of power construction become larger and larger, the power engineering cost management has been brought forward higher demands. On one hand, the traditional cost methods of engineer based on budgetary estimate and budget system has been not satisfied for new type cost management. On another hand, the engineering involved many factors, the analogized projects are limited, which make utilization of the history data so ineffective . What's more,because the present fashion big-sample learning method for mass data has no clear definition for the data capacity,it is so difficult to carry out practical problem accurately. Therefore, the paper emphatically studied the small-sample learning problem from learning small data.It is more difficult for machine learning of small sample data than machine learning of massive data, so it become hot issue in artificial intelligence study. Recently, such study mainly emphasis on per-processing technique of small sample data, convergence ,robust and parameter optimization performance of learning algorithm. From these way, the train of thought in the paper is as following: Based on the history project data of power engineer cost in some region,firstly a whole set of specific data per-processing scheme is proposed,it is used for simplifying the data based on the respective specific characteristics of power transmission and power station project data. Aiming at the characteristic of transmission construction data that the amount of data and the differences of data is large and the optimal value by searching optimization has no representativeness, the FCM (Fuzzy C-Means)clustering algorithm is used to make clustering process for the simplified data. Aiming at the characteristic of power construction whose data capacity is small and the differencies of different voltage grade is large, the basis-sub module is proposed to classificate the data set. Through the two method , the suitable small-sample is produced. Sencondly, aimed at the small-sample data , the PSO-LSSVM model which is fast and effective in the aspect of small sample learning is presented. In the model, the LSSVM(Least Squres Support Vector Machine) put a good performance in small-sample learning and it is also suitable for learning the data of limit amount and complex influence factors, And the PSO(Particle Swarm Optimization) as a great algorithm of parameters optimization optimates the SVM(Support Vector Machine) parameters in the model.Campaired with the tranditonal crossover algorithm for the SVM parameters optimization, it shows that the optimization algorithm based on PSO is more effective. Finally, on the basis of the engineering estimating model, the corresponding sensitivity analysis modle is presented, which is able to make a fast review for the project cost.By the practice application analysis,it suggests the modle could control the project cost effectively during the designing process.A whole cost management system is designed in the paper, a kind of new rapid estimation and review method is presented which is different from traditional method through design documents, quotas and service fee standards. Simulations results based on transmission line and power station project show such new method can effectively realize project cost management and provide technical support for project construction...