Papers by Keyword: BP Network

Abstract: According to the measured gas content in power transformers, we use BP neural network to accomplish the pattern recognition of transformer fault. The recognition effect of BP network pattern was studied from the aspects of adding over-fitting operation and genetic algorithm. Four kinds of neural network models, BP model BP & over-fitted identification model GABP model and GABP & over-fitted identification model, were constructed respectively, making the pattern recognition effect further enhanced.

Abstract: Schedule control is the major issue in project management, and to predict the construction schedule effectively is important practically. The article mainly predicts the schedule of a project based on BP neural network. The result shows that the predicted value is more accurate than the value calculated by linear method.

Abstract: A three layer feed-forward neural network is developed for speech recognition for industrial robots. The signal is made up by four kinds of command sound. Based on the steepest descent method, the BP network is trained by an improved learning algorithm in which momentum factor and learning rate adjustment is adopted. The simulation results indicates that the error in classification small also with high recognition correct in terms of a percentage.

Abstract: This paper proposed on-line detect method of eliminating the influence of sample order and adapting adjustment parameters, according to BP network was slow convergence and computationally intensive. Improving the speed and efficiency of training, the recognition process was more reliable and recognition results were more accurate. The experiment showed that improved BP network could meet the requirements of online testing, and recognition rate reached more than 91%.

Abstract: This paper designs the multilayer feed-forward neural network based on the immune genetic algorithm to solve the problem that BP algorithm is prone to get the local minimum in the failure diagnosis system. It is of both the learning ability and robustness of the neural network, as well as the strong global random searching ability of the immune genetic algorithm. The simulation results indicate the neural network can fulfill failure diagnosis of the complicated production better.

Abstract: In order to improve the depth performance of AUV in parking, a PSO-BP algorithm for the depth control is presented. The algorithm can use the standard particle swarm (PSO) as BP neural network learning method, and which can be evolved in the AUV depth adaptive control. The adaptive controller has adopted the double neural network unit. One of controllers is made use the input terminal to output control quantity on the basis of current displacement and vertical acceleration of the AUV. The other can be recognized on-line by the AUV model identifier. The numerical simulations are given to verify the AUV depth adaptive control by the controller. The results show that the proposed algorithm can significantly improve the AUV depth control performance. The convergence speed of AUV depth control is 4.5 times than the PID algorithm, so the efficiency of the AUV depth is vastly perfected.

Abstract: Artificial neural networks were introduced in the area of laser milling. The prediction model of surface quality in laser milling parts, including the width, depth of cladding layer, was proposed based on the back propagation (BP) neural networks. The model combined the global optimization searching performance of the genetic algorithm and the localization of the back propagation (BP) neural networks. Five technical parameters were selected to test the reliability of the model. The simulation and experimental results show that the evolutionary neural network based on genetic algorithm can effectively overcome the problem of falling into local minimum point. This method can get higher accuracy of prediction. It improves the measurement precision with the maximum relative error 2. 21% between the predicted content and the real value.

Abstract: There are different voltages in car electrical parts, and power switches are achieved by pushing. Electromagnetic radiation which is caused by switches has a great effect on the test components and close power. The voltage caused by Electromagnetic interference could be up to 20V. Controlling panel is established by LabView and power is controlled by Matlab. The test of Car electrical parts is achieved by power. Power output is controlled by panel. Wavelet analysis is used for graphics. Collecting data is more reliable through the BP Network. The error is reduced by 30%. Amplification circuit, anti-mixing and interference circuit and triode switching circuit are designed for remote controlling power and reducing electromagnetic radiation. It raises anti-interference ability. Electromagnetic radiation is mainly focused on 0-60MHz from experiments. According to this frequency brand, we adopted measures for reducing electromagnetic radiation by 8.6dB.

Abstract: The best is to read these instructions and follow the outline of this text. The effectiveness evaluation of Network Countermeasure (NCM) is a multi-level and multi-index problem. In order to objectively and accurately evaluate the NCM effectiveness, an index system for the NCM effectiveness evaluation is set up, and a SOM-BP cloud neural network evaluation method which combined cloud theory and neural network is presented. This evaluation method has not only randomness and fuzziness of cloud theory, but also the learning and remembering ability of neural network. Using this Method to evaluate the NCM effectiveness, we find this method has the advantage of fast convergence, few iterations, and high accuracy.

Abstract: A new structure of 6-RSPS simulator platform which can be rotated infinitely is presented. The solution of direct kinematics problem of parallel structure is the fundamental problem. For the solution of this problem, according to the links vector diagram, the relationship of the position and orientation of the platform between the length of links and the rotate angle of the under-hinge is derived. BP network is applied, then using the previous data obtained by inverse kinematics as the training data and testing data of BP network, get the direct kinematics solution. Simulation results verified that the solution of direct kinematics applied the BP network can be meet the requirements completely, and the error percentages are within the acceptable range.