Regression-Based Optimization and Control in Autonomous Systems

Machine learning techniques are currently used in many industries, smart factories for predictive maintenance, fault tolerance, and optimization. One of the biggest advancements in the field of automation is the ability of the machines to adopt machine to machine (M2M) or edge computing technologies to provide a secure layer of communication and superior control over different units. Regression is a commonly used machine learning technique for analytical insights, estimating performance metrics, prediction, etc. and most of the industrial embedded appliances have automated machines driven by controllers which can be tuned using machine learning algorithms to have an effective control over its functional units. This paper focuses on developing an autonomous module which integrates M2M and machine learning techniques between two units (MSP430—a TI-based microcontroller unit and a PC) with the ability to predict, analyze the data, optimize different functional metrics using a neural network model, and establish proficient communication and control. The prediction model is designed using Spyder IDE which imports an open source dataset to predict the CO2 emission values using multiple linear regression and evaluate the accuracy of the predicted output. The predicted output is sent to the TI-MSP430 MCU via serial communication which uses the predicted values to control the hardware peripherals that cool down the source of emission. The mean absolute error, residual sum of squares, r-2 score, and loss values are calculated and it is observed that the model with ReLu activation function predicts the CO2 emission level accurately.