ugg milano Improve the sensor accuracy of neural n

Improve the accuracy of the neural network sensor

) And the sensor output Y (?) continue to enter the neural network learning, the network will continue to link the right to decline by the gradient algorithm (1) the learning process refresh this operation through multiple regression. When the square error reaches a minimum value at this time could be the end of the learning process the weights of each neural network will reach a stable factor. Completed the construction of neural networks digital filters digital filters to establish when the neural network has been completed, can be used to improve the accuracy of the lower precision of the sensor can be seen that the use of neural networks to improve the accuracy of the sensor is divided into two steps to achieve completed. First, a certain number of data for evolutionary learning neural network, the data include only the sample data with high precision and low accuracy sensor output in two parts. In which the sample data with high precision determines the precision of the effect of neural network filters. Generally be obtained from the high-precision sensors to complete the second is evolutionary neural network learning and training, as a filter to achieve the improvement of sensor accuracy to be processed. At this point the neural network has become an independent online using filters. When the sensor itself due to environmental changes and led to the original nan sensor accuracy drops, may at any time as needed to amend the neural network to enhance the relevance of the sensor on the environment and the virtual nature of the reliability of their own. Total of 804 experimental and summarize the experiment,[link widoczny dla zalogowanych], the filter model of artificial neural network input layer has 6 neurons, 5 hidden layer neurons, output layer is a neuron. Found that the use of artificial neural network filter has good convergence and stability, the time when the learning coefficient for the 0.2l. Recursive about 1600 times,[link widoczny dla zalogowanych], you can make the filter output and the standard sample mean square error MSE to achieve a 90dB: adjust learning coefficient r /,[link widoczny dla zalogowanych], can also change the convergence rate of neural network algorithm. Experiment, we use the non-compensated M0Iorola silicon pressure sensor MPX10 as subjects. MPXIO no internal compensation is a type of pressure sensor. No account is simple. Low price, but the accuracy is low. For simplicity, in general, the accuracy of the sensor can be the maximum error obtained with the ratio of full scale. Is a _ which n1axl a lP. For measuring value for measuring the value of the arithmetic mean, y for the full scale value. The experiment, MPX10 precision p 0.35 0.42. As a standard neural network learning sample of data from the internal temperature compensation and correction with high precision silicon pressure sensor MPX2010. Since the sensor response and the film temperature compensation, calibration resistor network integrated on the same piece of silicon. Laser correction technology enables the sensor has high accuracy, the precision of P was 0.032 ~ 0.043. Way in Figure 2, neural network after learning, after completion of training,[link widoczny dla zalogowanych], the neural network filter for filtering the output signal MPX10. The results show that the accuracy of p MPX10 to 0.051 ~ 0062. Figure 3. 05n402, q0.2010.02002 King in January withdrawal 03D60.9I_2I_41.618W practice, thousands of changes in phthalocyanine see Figure 3MPX accuracy by using neural network filtering technology can greatly improve the accuracy of the low accuracy of the sensor. The low-precision sensors with high precision and improve the sensor performance and low cost. The method is applicable not only to pressure sensors. Also applies to other nonlinear sensors. References IGaoXZ, GaoXMA / DConv ~ erResolutionEnhancementUsingNeuralNetworks [A'ProeIEEEInstrumentandMea ・ su ~ mentTechnolo ~ 'Conference [C]. Ottawa, Canada: 1997 『2] SHaykin. NeuralNetworks, AComprehensiveFoundationrMJNewY: MacmillanCollegePublishCompany \: National Defense University Press,[link widoczny dla zalogowanych], 1997L6 gold seal Chang, Ming when the modern sensors. M] Beijing: Electronic Press, 1995

相关的主题文章：

ugg boots günstig Colored measurement noise Sinus

[link widoczny dla zalogowanych]

[link widoczny dla zalogowanych]

The post has been approved 0 times