The Design And Implementation Of Power Grid Data Mining Platform Subsystem

With the rapid development of Smart Grid, on sections of electricity generation, distribution, usage had produced huge amounts of data, and still dance grows quickly. These data are invested with information flow,power flow and rvices flow information. The enormous amount of power grid data contained potential information of users' power consuming behavior and operation condition of the distribution network. But simple mathematical statistics can hardly dealing with massive data and discover new knowledge of power grid. So that, the integration of smart grid and data mining technology has become an inevitable development trend.Reasonable application of grid data mining for scientific research and Application, whether it is effective for the power sector to optimize the grid structure, power resource allocation, or to improve the experience of power users, there is a very important significance.In order to solve the problem of lack of data mining platform system in the process of data mining in power grid. In this paper, we design and implement a data mining platform subsystem for grid data. The system is a HDDFS and Spark based distributed cluster of B/S architecture Web platform. Through this system, we can quickly realize the data mining,such as clustering, classification, association analysis and so on. At the same time, it can also facilitate the rapid integration of power grid data mining scene and grid knowledge discovery.Firstly, according to the current situation of data mining research and the development of data mining platform, this paper puts forward the requirement of grid data mining platform in function and non function.According to the requirement, design the overall system architecture, data storage, external interface, workflow and deployment view. This paper also designed and implemented a programmable solution for remote use of Spark, allows developers to use object-oriented programming in the Web framework for distributed data sets. Then, the core modules of the system detailed design, illustrates the integration and operation mode of the system of grid scene. Base on the data of 2015, this paper design the remote control results of distribution network scene, through the improvement of random forest algorithm to improve the quality of the model.The experiment proves that the module prediction results can provide effective reference for dispatchers. Finally, through unit testing, integration testing and performance testing, the system can provide an effective data mining platform for grid data mining research and application.