Yong-Wei Li

Application of multi-sensor information fusion technology in the power transformer fault diagnosis

This paper is aimed at applying BP neural network and Dempster-Shafer (D-S) evidence theory to realize the real-time monitoring and the fault diagnosis by taking power transformer as the object of fault diagnosis. We make use of the neural networks ability of better fault tolerance, strong generalization capability, characteristics of self-organization, self-learning, and self-adaptation, and take advantage of multi-source information fusion technology to realize comprehensive processing for uncertainty information. Combining with BP neural network and D-S evidence theory, a characteristic layer fusing model of power transformer fault diagnosis has been established. As high-voltage electric equipment has complex structure and works in harsh environments, the fiber bragg grating (FBG) temperature sensors are used to monitor the real-time thermal characteristics of the power transformer hotspot. The simulation results of power transformer fault diagnosis shown that this method is effective.

Research of DSP-Based SVPWM Vector Control System of Asynchronous Motor

This paper is aimed to expound and optimize the principle of SVPWM and then realize the optimized SVPWM by using the TMS320LF2407 digital signal processing (DSP) which is specially for the motor control. It has introduced a control system of asynchronous motor based on the digital SVPWM through the hardware and software design, and makes a simulation by using the Matlab of Simulink for asynchronous motor SVPWM control system.

Research of path planning based on adaptive dynamic programming for bio-mimetic robot fish

In robot path planning, artificial potential field (APF) is used to describe complex environment information. This paper proposes an artificial potential field-based adaptive dynamic programming (APFADP) method and applies it to the bio-mimetic robot fish path planning. In ADP, according to Bellman optimal theory, system cost is conventionally used to present control action cost. In APFADP, a novel potential field is defined according to system cost which is based on APF and the description of environment information can be given through the learning process. In the proposed method, we use action-dependent heuristic dynamic programming (ADHDP) that consists of two neural networks: the critic network and action network. The critic network is designed to approximate system cost by learning from position variables and angle between robot fish movement and target. The action network is designed as a controller to find the optimal path, which produces control outputs by learning from position variables. Verification has been conducted to illustrate the good performance of the proposed method by experiment results on bio-mimetic robot fish path planning.

Neural Network Based Visual Tracking with Multi-cue Adaptive Fusion

Visual tracking has been an active area of research in computer vision. However, robust tracking is still a challenging task due to cluttered backgrounds, occlusions and pose variations in the real world. To improve the tracking robustness, this paper proposes a tracking method based on multi-cue adaptive fusion. In this method, multiple cues, such as color and shape, are fused to represent the target observation. When fusing multiple cues, fuzzy logic is adopted to dynamically adjust each cue weight in the observation according to its associated reliability in the past frame. In searching and tracking object, neural network algorithm is applied, which improves the searching efficiency. Experimental results show that the proposed method is robust to illumination changes, pose variations, partial occlusions, cluttered backgrounds and camera motion.

The study of enhancing temperature sensitivity for FBG temperature sensor

In this paper, the basic sensing principle of fiber bragg grating (FBG) temperature sensor is introduced, and the general packaging methods are showed. On that basis, a new method of packaging FBG is advanced, which increases the thermal sensitivity coefficient and eliminates the impact of stress. Through experiments, the thermal characteristics and stress effect of the naked FBG and the packaged FBG are analyzed, the results show that the trend line of the packaged FBG has good linearity, and the method basically eliminates the stress effect, the sensor can be used accurately to monitor states of the thermal equipments, and the thermal range sensed is from −15°C to 200°C.

The temperature-variation fault diagnosis of high-voltage electric equipment based on information fusion

As high-voltage electric equipment has complex structure and works in harsh environment, FBG (fiber Bragg gating) sensors were applied to realize the real-time monitoring of some characters in which temperature was taken as the main factor. Using neural network to recognize and classify fault types, making a further fusion of fault information by expert system. After simulation and experiment, it shows good results, and provides a effective way to realize the monitoring and exact diagnosis of temperature-variation fault on high-voltage electric equipment.

Applied Research in Fuzzy Neural Network Predictive Control

Being aimed at the problem that non-linear and long lag-time technological processes are difficult to control, this paper puts forward a predictive control method which combines BP neural network with fuzzy control. After having tested real-time control of the synthetic ammonia decarburization calcinations processes in a decarburization factory, this method is proved to be effective.

Research on the Congestion Control of Broadband Integrated Service Digital Network Based on ATM

Congestion control technique is important for B-ISDN. In terms of the latest recommendation of ITU-T, the basic ideas, operational mechanism and functions are analyzed. Flow control and congestion control are the two levels included in the congestion control mechanism of B-ISDN. Congestion control of B-ISDN is completed by cooperating flow control with congestion control. Congestion control mainly includes selective loss of cell and congestion indication, which is an important and complex problem to be solved

A kind of utility function in adaptive dynamic programming for inverted pendulum control

The inverted pendulum is a complex, classic nonlinear system. In this paper, we apply adaptive dynamic programming that is a learning control method designed by neural network for inverted pendulum control. To achieve a good control performance, a kind of utility function is studied by means of action-dependent heuristic dynamic programming. According to the type of utility function based on power function and exponential function, a novel utility function will be designed in this paper. Using the utility function to design a controller in inverted pendulum system, and comparing with other utility function used in controller for inverted pendulum control problem, improvement experiment result is obtained.

Research on modelling and optimisation of RBF neural network based on particle filter

The BP neural network algorithm used in approximation of function has two shortcomings: the one is its slower convergence rate and the other is its falling into the local minimum. In order to solve the problems, a new method is proposed in this paper and is complicated on a typical complex system – the synthetic ammonia decarbonisation industrial process. The main issue of the proposed approach is on modelling and optimisation of the radial basis function (RBF) neural network based on particle filter algorithm. The learning capability and the advantage of particle filtering algorithm on processing non-linear system are used and the modelling and optimisation of RBF neural network based on particle filter is present. Furthermore, some simulation studies with calcination section have been done. The simulation result shows the superiority of the RBF neural network based on particle filter algorithm. It provides an efficient way for the complex system modelling and optimisation control research. Both the experimental results and the application validate the feasibility of the proposed algorithm.