Predicting Intelligent Instrumentation Controller based on Supervised Learning

Abstract

This research article establishes a novel approach for realizing controller of a flow process
station. This certifies ultimate optimization by replacing the conventional PID controller by a
smart controller in a closed loop process station. The smart controller is designed using
machine learning concepts. The flow process station flow transmitter training dataset is
obtained by running the process using traditional PID controller from initial condition to steady
state condition, the report generated is used as input to train the neural network. The
performance of the flow process is evaluated for minimum error. The Neural network in weka
tool trains the data fed to for the corresponding Controller output there by implementing the
controller using weka software, the performance is evaluated MSE criteria. The performance
analysis gives us a clear perception that the output is nearer to the look up table values gained
by running the process by the traditional method using PID controller. The numerical value
prediction is done using linear regression, Multilayer Perceptron, SMOreg and Tree category
M5P are the selected classifiers to predict the mean square error, RMSE and Absolute error
percentage. The performance of predictors for the intelligent controller for measuring the flow
is found to be similar to the conventional flow process PID controller. Multilayer perceptron
and SMOreg classifiers have established an error percentage very similar while the same
behaviour is established by the pair Linear regression, M5P.