Kailong zhang | Northwestern Polytechnical University (original) (raw)

Papers by Kailong zhang

This paper proposes a novel parameter learning algorithm for a self-constructing fuzzy neural net... more This paper proposes a novel parameter learning algorithm for a self-constructing fuzzy neural network (SCFFN) design. It concludes dynamic prior adjustment (DPA) which is employed to adjust parameters according to the distribution of the input samples and group-based symbiotic evolution (GSE) which is applied to train all the free parameters for the desired outputs. DPA considers the relevance between input samples space and the IF-part parameters, which intends to accomplish coarse adjustment. Then, GSE is adopted to search the global optimum solution. Unlike traditional GA with each gene representing a whole fuzzy system, GSE divides the population into several groups that each one only represents a fuzzy rule. The full solutions can be generated by all possible combinations of the groups. The simulations results have verified that the proposed algorithm achieves superior performance in learning accuracy.

Robot, spaceship and other applications that can be called autonomous Cyber-Physical Systems (CPS... more Robot, spaceship and other applications that can be called autonomous Cyber-Physical Systems (CPS) are all smart embedded systems, which usually run in open environment by themselves with inhered intelligent capabilities. Besides the application intelligence, the researches of adaptive platform for such systems have also become topics recently. Based on the analysis of current work, a new environment adaptive scheduler (α-S) for multiprocessor platform of a cyber-physical system is proposed in this article. And then, its related structures, models, and reasoning methods are studied deeply. Different from the existing ones, α-S can schedule all tasks according to the environment change by two layered functions. The upper one uses not only the properties of a task but also the state of environment and system resources as input variables, which is defined as Universal Environment (UE), and then calculates the Quality of Scheduling of all tasks with a customized fuzzy logic. The lower one allocates these tasks onto processors automatically under the dynamic performance evaluation of computing resources, such as load and success rate of each processor. Further, some allocation policies are studied and defined to make α-S be more flexible to adapt different requirements. Finally, studied methods are simulated and then experimented on an obstacle-avoiding robot system.

To improve the resources utilization of complex hardware/software system, and response to real-ti... more To improve the resources utilization of complex hardware/software system, and response to real-time task instantly, this paper proposed a scheduling model of task resources and its fuzzy membership function based on fuzzy clustering analysis algorithm, and gave a method using average distance to calculate the distance between two membership functions. In addition, we also designed `Fuzzy Resources Clustering Algorithm, FRCA', which could find out the resources occupied by tasks. A simulation test was applied on a certain complex embedded real-time system, the results illustrated that the proposed algorithm was able to efficiently schedule task resources, and make the system has a quick response to tasks.

Because of the resource constraints and high reliability requirement of Embedded Distributed Syst... more Because of the resource constraints and high reliability requirement of Embedded Distributed System (EDS), some new fault-tolerance means, which are different from the traditional hardware- redundancy ones, should be studied. In this article, a fault-tolerance method that based on similar resources and related technologies are proposed and discussed. First, several mathematical models of key elements, such as computing nodes, similar nodes and tasks, are constructed. Then, the similarity computation methods and evaluation criteria are evinced by two different views: tasks and resources. Supported by theories above, numerous methods, such as similar nodes auto- discovery (SNAD) and its optimization one (oSNAD), redundant tasks auto-deployment, and reconfiguration policies of fault tasks and nodes are highlighted respectively. Simulation results show that these approaches and schemes can improve the adaptive fault-tolerance abilities of complicated embedded distributed systems.

Unmanned vehicles (UVs) have become more and more widely used. And augmenting their autonomy is a... more Unmanned vehicles (UVs) have become more and more widely used. And augmenting their autonomy is an urgent demand in order to make them play a bigger role. This paper presents a multi-agent architecture for UVpsilas distributed real-time embedded (DRE) system. It can support the autonomous operation of UV from two aspects: (1) at the global, task scheduling among multi UVs can be completed autonomously based on a mixed negotiation mechanism with less communication overhead, taking various dynamic and stochastic factors into account, such as arrival of new task, risk of UV loss; (2) at the local, an adaptive resource manager (ARM) is introduced. On basis of the feedback loop theory, it can improve the local autonomy by allocating resources efficiently and enabling the system to adapt to fluctuations in input workload, resource availability and operating conditions.

As a complex nonlinear system, greenhouse can not be controlled perfectly by traditional control ... more As a complex nonlinear system, greenhouse can not be controlled perfectly by traditional control strategies. This paper proposes a self-organizing fuzzy neural network controller (SOFNNC) with group-based genetic algorithm (GGA) to drive the internal climate of the greenhouse. SOFFNNC is a hybrid control strategy which combines fuzzy control and neural network organically. It generates or prunes neurons automatically by the structure learning algorithm, which can adaptively strike a balance between the rule number and the desired performance. In other to avoid the shortage of the original learning algorithm to SOFNNC, we come up with an improved structure learning method and a new parameter learning method with GGA. Based on a greenhouse model established by an Elman neural network (ENN), we test the performance of SOFNNC. Simulation and comparison results prove that SOFNNC can achieve outstanding control effect with high efficiency.

Today,the development of sewing technology has shown the seriate intelligent trend, and one impor... more Today,the development of sewing technology has shown the seriate intelligent trend, and one important factor is the embedded system technology. After careful research and analysis, this paper brings forward a customizable embedded system architecture, which is made up of customizable embedded hardware platform, customizable embedded OS and component-based embedded software. We have used this architecture to design new electro-pattern sewing machine successfully.

Failure detection is an essential part to build high dependable distributed real-time embedded sy... more Failure detection is an essential part to build high dependable distributed real-time embedded systems. However, the performance degradation of production work due to the execution of failure detection cannot be omitted. We present an adaptive performance management method for failure detection based on feedback control theory. This method is autonomous and thus allows the system to self-manage the CPU resources allocated for failure detection, with only the high-level policy input. Therefore, it limits the performance impact due to the execution of failure detection, even in the dynamic environment.