Design Of Neural Network Data Estimator Based On FPGA

With the development of Poultry industry, many farms use the sensor to measure the gases in the barn in order to control the waste gases concentration. Because of the fault of sensor, some measured value is distorted. For the sake of the dependability of the control, we need to supervise some important data to avoid the mistake. When the sensor is in trouble, we should do something to renovate the distorted data.In this article, we designed a estimator to predict the data measured from the sensor. It can obtain the more approximate data from the actual one. It can provide a reasonable management to the waste gases.Firstly, considering with the structure of the neural network, we found a new method to estimate the number of hidden neurons. It can improve the capability of the estimator. Secondly, we compared with the structure of the hardware, chose one method to carry out the activation function. Finally, we used the Altera FPGA chip EP1C6Q240C8 to design it. The design takes top-down method.In the experiment, we deal with the NH3's concentration to test the properties of estimator. Comparing with the difference of the actual data and predicted one, we can evaluate the properties of the estimator. In this article, the estimator can recognize the distorted data and predict the actual data, we can find that the estimator can work at the frequency of 33Mhz and the error is no more than 7.8%. It can be used in actual application. It is useful to improve the reliability and accuracy of the measure system.