Performance Degradation Evaluation And Intelligent Fault Diagnosis For Hydropower Generator Unit

With the continuous development of electric power industry in China, The hydropower has entered a new era with the larger unit, high voltage transmission and more intelligence. On the one hand, hydro energy has been increasing gradually in energy structure of electric power, and undertakes the more and more task of load frequency control. On the other hand, large hydropower station has been increasingly growing to electric power supply to industrial directly, due to the construction of the uhv dc transmission. Hydroelectric generator units (HGU) which are the key equipment of the hydropower production process are becoming more and more large-scale, complex, high-speed and high-power. Meanwhile, the sudden and seriousness of the HGU failure show prominent increasingly and the safe and stable operation of which is more and more important to the power grid is also growing. Therefore, the conventional hydropower state assessment and fault diagnosis methods which are based on the artificial experience can no longer be well adapted to the practical engineering application. It is a great of an urgent to construct the online health condition evaluation model and fault diagnosis method to accurately identify the fault type and its severity trends for hydropower station equipment. Finally an efficient and reasonable unit overhaul scheme and the corresponding policy recommendations should be provided. In view of the hydropower unit vibration fault diagnosis and some key scientific problems in engineering application, this paper focus on the hydropower unit vibration fault feature extraction and the optimal feature subset choice by introducing the theory of nonlinear signal identification and information entropy method and reveal multidimensional complex mapping relationship between fault symptom and different fault types of hydropower unit; it also construct online performance degradation evaluation method based on transient start-up process data of HGU; finally intelligent hybrid fault diagnosis method is introduced to achieve the goal of accurate diagnosis in this paper which is based on two-phase extreme learning machine integrated structure and parameters self-learning strategy. The main research content and innovative results in this paper are as follows:(1) Due to the fault signals of hydroelectric generating units tend to present the non-stationary and nonlinear characteristics. Research work will be integrated empirical mode decomposition method and the permutation entropy method, puts forward the intrinsic mode for feature extraction of permutation entropy method, this method can fully identify fault under the action of unit characteristics of chaos signal change, fully characterize the fault characteristic information unit. In this paper, aiming to fault feature extraction and optimal feature subset selection, we introduced the nonlinear signal identification and information entropy method, research on the contribution factor of the different fault feature. The method fully consider different correlation between model parameters and feature subset, by using continuous and discrete hybrid coding, the model parameters and feature subset could be optimized by the group of artificial bees colony algorithm with the is the search strategy, nectar source code and update strategy. Finally, the high correlation and low redundancy feature subset could be acquired.(2)Nowadays, field test and simple limit comparison for condition assessment and monitoring of HGU are widely used in the HGU. Nevertheless, this method is vulnerable to operation condition change, unit structure and measurement equipment, et.al and it is lack of early warning capacity of potential failure to fully reflect the operation state of HGU. Aiming at the problems above, this paper analyzed the similarity between start-up process and unit performance test and proposed a performance evaluation method based on the start-up process of HGU. Firstly, the basic trend of primitives was defined and start-up process was divided into several segments by the least squares algorithm. Each segment is represented by the primitive by using the first-order and second order derivative information of fit curve. will algorithm in the process of different of the trend of the trend of the first defines the basic primitive; secondly, start-up process is represent by primitive sequence which effectively avoid strong noise and shock signal interference in the HGU; at the same time, to build accurate quantitative indicators which measure of performance degradation degree of HGU, this paper proposed a start-up process similarity index with modified dynamic time warping distance, which can get the unit actual performance degradation of quantitative numerical and finally realized the online status of performance degradation assessment method.(3) Aiming at small sample, high dimension problem in fault diagnosis for HGU, this paper combined with the artificial bee colony algorithm and extreme learning machine, and proposed a classification model with the structure and parameter adaptive optimization named two stages evolutionary extreme learning machine. At the first stage, optimal number of hidden nodes are adaptive defined by greedy algorithm and cross validation which overcomes the problem that hidden nodes is difficult to determine in the training phase of original extreme learning machine learning; at the second stage the weights and bias parameters of extreme learning machine are determined by artificial bees swarm algorithm which make the parameters selection is adaptive and fault diagnosis method can be self-learning evolution. At the same time, this paper At the same time this paper proposed a improved artificial bee colony algorithm to adaptively balance exploit and explore ability and not to easily fall into local optimal solution by introducing the two search strategies with global guide and Levy flight. Finally it further improves the classification of extreme learning machine by the experiment test and practical diagnosis application of HGU.(4) Aiming at the problem that various business systems operated in the current hydropower plant differ in the development enterprise, development language, deployment platform, and different communication protocols due to the operation period is longer, the various business systems, moreover, external data exchange format cannot be compatible with each other, we proposed a kind of distributed remote fault diagnosis system design scheme in this paper based on service-oriented architecture (SOA) theory. It can handle data access and interoperability function between fault diagnosis system and external system in the heterogeneous environment and the external system effectively. The hydropower multi-source information could be integrated by constructing various applications of remote monitoring and fault diagnosis system on different levels and different abstraction granularity, with design and implementation of various core function module of the plan. Finally, an effective intelligent fault diagnosis could implement the goal of operation safety and automation and provide strong technical support on the hydropower station management. The distributed fault diagnosis system based on the SOA has been successfully launched in SongJiang river hydropower plant and has been applied for the operators in the power plant...