Room Temperature Control using LSTM

Room Temperature Control using LSTM emulating PID control

CS-539 Machine Learning Project 2022

Team:

Vaibhav Nandkumar Kadam

Prasanna Natu

Pratik Kumbhare

Omkar Bharambe

Objective of the project

We emulate classical PID controller used in HVAC system for indoor room temperature control using LSTM network such that PID can be completely replaced and exhaustive tuning of PID parameters can be avoided.

Introduction

Why temperature control?

Without temperature control there will be no warm, cozy rooms and refrigerators.

What is a PID controller?

A Proportional-Integral-Derivative controller is a mechanism that strives to achieve a goal set point by applying a correction to the measured variable (feedback) based on proprtional, integral and derivative terms.

Why Machine Learning?

Instead of tuning the PID parameters again and again, we employ a machine learning method to achieve the same goal with less effort.

Why LSTM?

LSTM can efficiently learn the trends and features of the data and has higher prediction accuracy than the traditional methods. On top of that, LSTM eliminates the need to implement and tune the PID parameters in the controller.

Steps taken while implementing the project

Identifying the need for room temperature control
Data generation and pre-processing
Building a LSTM model for the temperature controller
Training and Testing the data by tuning hyper-parameters
Comparing the results

Training LSTM Model - Hyperparameter Selection

window

layers = 2

batch_size = 100

drop = 0.1

units = 100

Besides is the loss trend of training LSTM network

LSTM Heater Value Predictions

LSTM based controller completely emulates the classical PID output for heater value percent.

We get RMSE of LSTM with PID about 0.107

Conclusion

We found that the LSTM model can replace the PID controller having almost same performance and efficiency.
LSTM model requires less efforts to implement and it eliminates the tedious task of tuning PID parameters.
The complicated nonlinearity between indoor temperature and other parameters were overcame by LSTM model due to its dynamic memory.
When compared with traditional methods, the model based on LSTM can better describe the features of HVAC system data, which leads to higher prediction accuracy.

Here is a small presentation to give you the gist of the project:

Room Temperature Control using LSTM based PID Control

GitHub link to source code:

Data set used (United States Department of Energy of Richland, Washington):

Page updated

Google Sites

Report abuse