An Optimization Method for the Initial Parameters Selection of Fuzzy Cerebellar Model Neural Networks in Parametric Faul
- PDF / 2,948,941 Bytes
- 12 Pages / 595.276 x 790.866 pts Page_size
- 68 Downloads / 157 Views
An Optimization Method for the Initial Parameters Selection of Fuzzy Cerebellar Model Neural Networks in Parametric Fault Diagnosis Qiongbin Lin1 • Shican Chen1,2,3 • Chih-Min Lin2
Received: 15 February 2020 / Revised: 24 April 2020 / Accepted: 15 June 2020 Taiwan Fuzzy Systems Association 2020
Abstract When the initial parameters of the fuzzy cerebellar model neural network (FCMNN) are not properly selected, there is a great possibility that the error function converges to the local minimum region or diverges under the gradient descent back propagation (BP) algorithm, which will affect the classification ability of FCMNN. Aiming at this problem, GA is used to optimize the initial center positions and width of the activation function and weight of FCMNN. After obtaining the optimal initial parameters, the internal parameters of FCMNN can approach the convergence value of minimum error more quickly and accurately through further training of the network, so as to obtain better network learning performance. The introduction of GA to optimize the initial value of FCMNN can effectively reduce the blindness and time cost of manual selection of initial parameters, and further improve the intelligence of neural network diagnosers. The simulation and experimental results show that the classification ability of the GA-FCMNN and GA can effectively find the optimal combination in the set data domain. Keywords Fault diagnosis Fuzzy cerebellar model neural network (FCMNN) Genetic algorithm (GA) Parameter optimization
& Chih-Min Lin [email protected] 1
College of Electrical Engineering and Automation, Fuzhou University, Fuzhou, China
2
Yuan Ze University, Taoyuan, Taiwan
3
Quanzhou Power Supply Company, State Grid, Quanzhou, China
1 Introduction Cerebellar model neural network (CMNN) is widely used in control systems and has the advantages of low computational complexity, strong generalization ability and fast learning speed [1–3]. In the classification problem, Lin et al. used FCMNN in the auxiliary diagnosis of breast cancer [4]; in the fault diagnosis of power electronic circuits, we have also obtained good diagnostic results using FCMNN [5]. However, in the research, we found that the FCMNN initial parameter selection directly affects the model learning ability, which is related to the BP parameter update algorithm. Once the initial weight and learning rate are not well selected, the model is easy to fall into the local minimum. To better select the initial parameters of the neural network, many optimization methods have been used to select the initial parameters of the back propagation neural network (BPNN). Usually, the optimization methods first select the optimal initial parameters of the neural network, and then carrying out further training, the internal parameters (weights, deviations) of the neural network will more accurately and quickly approach the solution that produces the smallest error, thus obtaining a fault diagnoser with superior performance. When studying the influence of parametric fau
Data Loading...
