Ammeter Examination And Information Management System Based On BP Neural Networks

With the rapid development of electric power enterprise, especial with the popularization of one door one ammeter recently, the output and workload of examination of ammeter are increased sharply. These make the examination of ammeter confront an austere situation, and the examination is hard to deal by conventional examination method. Domestical conventional examination platform is not so automatic. During the process of test the examination and management are separated, and it is difficult to accomplish entirely automation and satisfy the real requirement. Under such circumstance, it is especially important to develop a new intellectual ammeter examination system associated with information management system.The content of this paper firstly analyzes the ammeter examination system's actuality domestic and abroad and analysis the main methods of error examination as well. Then the theories are elaborated which are used in this topic, such as the basic theory of back-propagation neural network (BPNN), the problems existing in BPNN, the arithmetic in improvement and so forth. After these, these are expatiated: design objectivity, principle of designing and each function module of this system. Finally both the hardware and software of this system are implemented, and an analysis is got on incertitude measurement. The outcome shows that:this improvement measure can be used to accomplish the examination of ammeter, the management of ammeter's information, the print of ammeter's information, and it improves the automation and intelligence of ammeter's examination.This paper emphasizes the application, and makes further research and discusses mainly in following aspects:1. This paper studies the theory of BPNN, the modeling mechanism of BPNN,and systematically analysis the arithmetic of BPNN, the structure of network topology and improved arithmetic of BPNN, so as to find out a BPNN arithmetic adapting to the examination of ammeter—automatic-adaptive BPNN Arithmetic. The selection of the layer number of BPNN and the selection of each layer's node number both have a great influence on the astringency of the network.2. In theory, three layers BPNN can approximate to any non-linearity function. However, the data provided for BPNN may not be real and unprocessed, quality of processing data becomes a important factor to the access to the maximal proximity to get data. This paper adopts the methods of data pre-processing, scientific measurement times, the elimination of the illogical data, secondary-elimination of the disturbing data and data normalization for data processing.3. Ammeter Examination and Information Management System is designed and developed, which integrates intellectual examination and management as a whole. This system is based on Windows operating system, with human GUI, adopting large data based on SQL Server 2000 and Microsoft VB6.0, which assurances that the system has the ability of data processing and the system is stable. The programs for data processing and BPNN are produced by MATLAB software, which can compute array and matrix expediently.