Application Of BP Neural Network In Prediction Of Starch Moisture Content

Until now,starch has been widely used in various fields.In the production of starch,in order to have a better transport and storage.The water contained in our state’s orders that it should not exceed 14%.But in the actual production,the methods of measurement commonly are off-line measurement,which is in a specific period of time to collect starch,then the collected samples were send to the laboratory and to be measured.This method has long period and low precision.So in the production process we often use the method of artificial judgment,by virtue of the experience of workers to judge whether the starch is more than the national standard.Although this method can reduce the loss which is caused by starch does not meet the standards of our country,its credibility is not high.Imported instruments can achieve online real-time measurement with high precision,adjustable cycle and other features,but the price is very high.In order to solve the problems of long measuring period,unreliable artificial measurement and expensive cost.Therefore virtual measurement technology has been proposed.The core is computer software that using BP neural network algorithm.The discharge temperature and humidity as the input of the network,and the moisture content as the output of the network.Through training the network,we can get the relationship between input and output.This approach is divided into two steps,one is offline training,and the anther one is online computing.Offline training is to select the sample data,then through the mean square error to compare the corresponding error in different hidden layer node number,and finally determine the optimal structure of the neural network.Online computing is to use the PLC or WinCC configuration software to provide input data for the network,the output results are sent back to the WinCC to display after the MATLAB calculation.So we can get a real-time prediction curve.In the whole control system,the PID control is adopted at home and abroad.The only difference lies in the choice of the hardware of the detection unit on the feedback channel.In order to achieve a higher quality of automatic technology.The paper gives three different control methods.Through the analysis of advantages and disadvantages of each method.Finally we select the control method which combined neural network with PID.Virtual measurement instrument with BP neural network algorithm,which has the function of real-time measurement,adjustable cycle,high reliability.It can replace the expensive online measuring instrument and the artificial judgement.The training network can also be used as a virtual sensor,which can be used in the feedback channel of the closed control loop.So the whole drying control system can be automatically adjusted.