Ziwen Leng

Short-term traffic flow forecasting model based on wavelet neural network

Short-term traffic flow forecasting plays an important role in the urban traffic control and guidance system. In the paper, the advantages of wavelet transform and artificial neural network are introduced. Focusing on the characteristics of time-variation and uncertainty of urban traffic flow, the paper adopts the combination of wavelet analysis and artificial neural network, establishes the short-term traffic flow forecasting model of wavelet neural network (WNN) and carries out independent test by rolling forecasting based on the measured data from traffic library. Simulation results indicate that, compared with the forecasting model of BP neural network, the WNN model has better forecasting precision and faster convergence speed, and wavelet neural network could be better applied in the short-term forecasting of traffic flow.

Application of GA-LSSVM in Fault Diagnosis of Subway Auxiliary Inverter

Focusing on the fault diagnosis precision of subway auxiliary inverter, the diagnosis method based on genetic algorithm (GA) and least squares support vector machine (LSSVM) is proposed in this paper. First, the optimal parameters of LSSVM are obtained by GA with global search capability and the diagnosis model of the optimized LSSVM is established, then the empirical mode decomposition (EMD) is introduced to decompose the fault signal into several intrinsic mode functions (IMF), and finally we will extract the approximate entropy of each IMF as the fault feature which will be applied to test the performance of the diagnosis model. Simulation results have proved that the proposed diagnosis method is feasible to recognize each fault and has achieved higher precision.

Short-term forecasting model of traffic flow based on GRNN

Urban traffic flow has the characteristics of nonlinearity and time-variation, and how to accurately forecast short-term traffic flow has been an essential part in traffic field. Taking advantage of the Generalized Regression Neural Network (GRNN), the paper establishes the short-term forecasting model of traffic flow based on GRNN. The GRNN model selects the cross validation algorithm to train the network, takes the root mean square of forecasting error as the evaluation criterion of the network to determine the smoothing factor and uses the method of rolling forecasting to forecast the traffic flow. Compared with the forecasting models of RBF and BP neural network, GRNN has stronger approximation capability and higher forecasting accuracy.

Fault diagnosis of subway auxiliary inverter based on EEMD and GABP

Focusing on the non-stationary characteristic of the fault signal of subway auxiliary inverter, this paper proposes the method that combines ensemble empirical mode decomposition (EEMD) with genetic algorithm to optimize BP neural network (GABP) to diagnose the fault categories of subway auxiliary inverter. Firstly, this paper extracts feature vectors from the original fault signal by EEMD, then establishes the multi-fault diagnosis model by GABP. The genetic algorithm (GA) is introduced to search the optimal solutions of initial weight and thresholds of BP neural network (BPNN), so as to improve the convergence and precision of diagnosis of network. Simulation results show that this method we proposed can identify these faults more accurately and higher efficiently.

Short-Term Forecasting of Traffic Flow Based on Genetic Algorithm and BP Neural Network

This paper focuses on the short-term traffic flow forecasting which plays an important role in traffic control and guidance system. To improve the efficiency and precision of short-term forecasting, the method of BP neural network optimized by genetic algorithm (GA) is put forward. Considering the shortcomings of traditional BP neural network, genetic algorithm is used to optimize the weight and threshold of BP neural network by training sample, then the optimized network is tested by testing sample and the network conducts short-term forecasting for 30 min by means of the rolling forecasting. The validity and accuracy of GA-BP forecasting model is tested on the basis of traffic flow. The simulation results demonstrate that the proposed model has higher superiority in convergence rate and forecasting precision.

Application of EKF Training RBFNN in Fault Diagnosis of Subway Auxiliary Inverter

Fault signals usually have the characteristics of nonlinearity and instability, and RBF neural network (RBFNN) has the defect of lacking of generalization capability when it is used for fault diagnosis. To improve the precision of fault identification and network convergence, the paper introduces the extended Kalman filter (EKF) as the learning algorithm of RBFNN to estimate the center vector of hidden layer and the network connection weight, extracts the feature vectors of fault signals in subway auxiliary inverter by wavelet package, which will be taken as the input samples of the optimized network. Simulation results show that, compared with traditional gradient descent algorithm, RBFNN trained by extended Kalman filter has better effect in improving the diagnostic precision and speeding up the network convergence, and the proposed method can be applied in fault diagnosis of auxiliary inverter.

The application of maximum differential algorithm in adaptive ocean sampling

This paper presents a technique for adaptive ocean sampling using ocean sampling platforms equipped with multiple sensors. The virtual environment of 2D ocean sampling is established, so as to simulate the ocean sampling region by means of the sampling platforms. There are three important phases which can be written as collecting scientific data, drawing the sampling area, and utilizing the maximum differential algorithm (MDA) in ocean sampling. By analyzing the sampling data and using the maximum differential algorithm, the sampling platforms achieve the optimizing sampling path. The simulation results by adaptive ocean sampling of single sampling platform and multiple platforms show that the proposed approach is effective and feasible. This method can be applied to conduct the moving direction based on the ocean sampling platforms.

Fault Diagnosis of Subway Auxiliary Inverter Based on PCA and WNN

Taken the nonlinearity of fault signals in subway auxiliary inverter and the diagnostic precision into consideration, the paper proposes the fault diagnosis method on the basis of principal component analysis (PCA) and wavelet neural network (WNN). Firstly, extract the initial feature vectors of fault signals by the decomposition and reconstruction of wavelet package, then use PCA to reduce the dimension of initial feature vectors, so as to eliminate redundant data information. Finally, the processed feature vectors will be taken as the input samples of wavelet neural network for the fault diagnosis. Experiment results have tested and verified the feasibility and effectiveness of the method. The proposed diagnostic method has higher precision and stronger convergence than the network directly using initial feature vectors.