Narrow Results

This paper presents a repair algorithm for the design of a Radial Basis Function (RBF) neural network. The proposed repair RBF (RRBF) algorithm starts from a single prototype randomly initialized in the feature space. The algorithm has two main phases: an architecture learning phase and a parameter adjustment phase. The architecture learning phase uses a repair strategy based on a sensitivity analysis (SA) of the network's output to judge when and where hidden nodes should be added to the network. New nodes are added to repair the architecture when the prototype does not meet the requirements. The parameter adjustment phase uses an adjustment strategy where the capabilities of the network are improved by modifying all the weights. The algorithm is applied to two application areas: approximating a non-linear function, and modeling the key parameter, chemical oxygen demand (COD) used in the waste water treatment process. The results of simulation show that the algorithm provides an efficient solution to both problems.

Combined with RBF neural network and sliding mode control, the synchronization between drive system and response system was achieved in module space and phase space, respectively (module-phase synchronization). The RBF neural network is used to estimate the unknown nonlinear function in the system. The module-phase synchronization of two fractional-order complex chaotic systems is implemented by the Lyapunov stability theory of fractional-order systems. Numerical simulations are provided to show the effectiveness of the analytical results.

In order to improve the fault diagnosis rate and efficiency of diesel engine, the PCA-RBF neural network as a new algorithm was constructed by combing the character extraction ability of PCA with the nonlinear approximation ability of RBF neural network. Firstly, eight factors which affected the fault types of diesel engine were analyzed and three principal components were extracted by PCA. Secondly, the data obtained from the three principal components were taken as the input of RBF neural network which was trained and tested. Finally, the PCA-RBF neural network was verified through simulation. The simulation results show that the network has fewer training steps, less training and higher training accuracy.

This paper introduces a novel method for the recognition of human faces in two-dimensional digital images using a new feature extraction method and Radial Basis Function (RBF) neural network with a Hybrid Learning Algorithm (HLA) as classifier. The proposed feature extraction method includes human face localization derived from the shape information using a proposed distance measure as Facial Candidate Threshold (FCT) as well as Pseudo Zernike Moment Invariant (PZMI) with a newly defined parameter named Correct Information Ratio (CIR) of images for disregarding irrelevant information of face images. In this paper, the effect of these parameters in disregarding irrelevant information in recognition rate improvement is studied. Also we evaluate the effect of orders of PZMI in recognition rate of the proposed technique as well as learning speed. Simulation results on the face database of Olivetti Research Laboratory (ORL) indicate that high order PZMI together with the derived face localization technique for extraction of feature data yielded a recognition rate of 99.3%.

This paper introduces an efficient method for the recognition of human faces in 2D digital images using a feature extraction technique that combines the global and local information in frontal view of facial images. The proposed feature extraction includes human face localization derived from the shape information. Efficient parameters are defined to eliminate irrelevant data while Pseudo Zernike Moments (PZM) with a new moment orders selection method is introduced as face features. The proposed method while yields better recognition rate, also reduces the classifier complexity. This paper also examines application of various feature domains as face features using the face localization method. These include Principle Component Analysis (PCA) and Discrete Cosine Transform (DCT). The Radial Basis Function (RBF) neural network has been used as the classifier and we have shown that the proposed feature extraction method requires an RBF neural network classifier with a simpler structure and faster training phase that is less sensitive to select training and testing images. Simulation results on the Olivetti Research Laboratory (ORL) database and comparison with other techniques indicate the effectiveness of the proposed method.

To reconstruct two-dimensional computerized tomography (CT) images from a small amount of projection data is a very difficult task. In this paper, two methods based on radial basis function (RBF) neural network are investigated to perform such a work. In the first method, we take projection data as the input and original image as the output of the network, after trained with some samples, the network can be applied to reconstruct CT image in the same class. In the second method, we adopt coordinate and a cross-section image as the input and the output respectively. For converting the image to its projections, an additional integral module is cascaded with the network. To evaluate these two methods, a comparative study is presented. A pixel-wise error estimator is adopted to calculate the overall error of the reconstructed images. Experiments show that the second method is the best for moderate projection data in practice.

A novel semi-blind defocused image deconvolution technique is proposed, which is based on RBF neural network and iterative Wiener filtering. In this technique, firstly a RBF neural network is trained in wavelet domain to estimate defocus parameter. After obtaining the point spread function (PSF) parameter, iterative Wiener filter is adopted to complete the restoration. We experimentally illustrate its performance on simulated data. Results show that the proposed PSF parameter estimation technique is effective.

In this paper, a new methodology for optimal trajectory planning of robotic manipulator in the joint space has been described. The RBF neural network is used for general approaching problem of nonlinear mapping in the joint space. A single-input and six-output RBF neural network model is built and trained. The data got from the inverse kinematics equation are used as training samples and the interpolating calculation was completed in 6-dimension joint space. With characters of rapid convergence and well approximation, this new algorithm is fault tolerant and irrelative with order of inputs, which can ensure the result trajectory is smooth enough. The algorithm has been tested in simulation in the software ADAMS, yielding good results by studying the kinematics and the dynamics performance of the robot.

In this paper, a fish swarm behavior based RNA genetic algorithm (fsRNA-GA) is proposed for the modeling problem of 2-Dimensional overhead crane. Inspired by fish-swarm algorithm, a neighborhood evolve operation is designed to improve the local search performance. Moreover, an adaptive permutation probability is introduced to hold potential individuals. To demonstrate the efficiency of fsRNA-GA, the proposed algorithm is employed to optimize the parameters of RBF neural network for modeling the 2D overhead crane. The experimental results show that the RBF network model is consistent with the actual experiment data.

In this paper, an adaptive intelligence sliding mode method was studied to improve the maneuverability and safety of ship motion control. In the design process, a ship nonlinear model was established, the control law of SMC and the structure of RBF neural network were proposed. By adopting an online learning algorithm, the parameters of the RBF neural network were optimized. In the intelligence controller, sliding model variable structure was utilized to solve the parameter perturbation and environment disturbances in ship sailing, and RBF was utilized to enhance the controller performance and to reduce vibration caused by the SMC. At the end of the paper, course keeping simulations were done to test the applicability of the designed controller. The results show that the intelligence controller can reduce the vibration trouble in SMC, and achieve high stability and fairly strong robustness.

Infrared gas sensor detecting gas concentration inaccurately is mainly because infrared gas sensor is susceptible to temperature, wind and other factors. This paper proposes an improved RBF neural network which can substantially overcome the environmental influences. Its idea is to optimize the parameters of RBF neural network by gradient descent method. Experimental results show that compared with RBF neural network the error of the improved RBF neural network is smaller.

Against such characteristics as uncertainty of controlled object in the main steam temperature control of thermal power plants, a novel PID control strategy with radial basis function (RBF) network tuning based on quantum-behaved particle swarm optimization (QPSO) algorithm is proposed. The QPSO algorithm is applied to optimize the initial parameters of RBF network, thus achieving dynamic control of the main steam temperature. The proposed controller has a self-learning ability, which can strengthen the system of uncertainties adaptability. Simulation results show that the control system performance is obviously better than the conventional cascade control.

In view of the traditional Kalman filter algorithm the shortcoming which the probable error fill-out even diverges when the model errors exist, considering the RBF neural network has the strong non-linearity to approach ability, propose assisting the Kalman filter with the RBF neural network the new algorithm, and apply it to transfer alignment of carrier aircraft’s INS. Simulation results indicate this algorithm surpasses traditional the Kalman filter algorithm.

The paper proposes the RBF neural network control method of reactor temperature, according to the PWR nuclear power plant units of 900MW. Firstly, the characteristic of control system of reactor temperature is tested on the simulation platform of PWR nuclear power plant; then, depending on the data from the experiment, the system is analyzed and transfer function fitted, and the transfer function of reactor temperature comes out. Basing on the traditional PID control, the RBF neural network combines with traditional PID control to achieve composite control, which has better control effect than traditional PID and fuzzy PID control.

Although Continuous Emission Monitoring System (CEMS) has been applied to measure the emissions from the thermal power plant, there are still some disadvantages including significant capital investment, excessive costs of maintenance and so on. This paper studied the soft senor methods for SO₂ emission from the thermal power plant based on RBF and BP neural networks. Studies illustrate that the relative errors of predictions based on RBF and BP neural networks are 8.8% and 6.9% respectively. The simulated results of the actual data show that the methods can objectively reflect the real situations of the flue gas emissions and can meet the needs of practical projects.