Smart Control of HVAC using Machine Learning, IoT and Data Analytics

  Smart Control of HVAC using Machine Learning, IoT and Data Analytics

1.   Introduction:

The basic purpose of heating, ventilation and air conditioning (HVAC) systems is to offer thermal comfort and maintain good indoor air quality through proper ventilation and filtration. HVAC systems regulate the entire climate of a facility. They also make the necessary modifications whenever the outside temperature changes. During the winter, HVAC systems (mostly boilers or heaters) struggle to maintain a suitable indoor temperature. Because it can increase both energy efficiency and indoor environmental quality, artificial intelligence has promising applications in HVAC. AI can optimize variables like airflow, temperature, and humidity, preserving air quality while consuming the least amount of energy. IoT is very valuable in the HVAC business for tracking vibration, airflow, pollutants, the number of occupants, weather conditions, and other variables. Monitoring key stress indicators and remote analysis enhances troubleshooting and makes preventative maintenance easier to implement.

Result: Movement of Actuator based on Predictions

1.1.         Sensor and Actuator Layer:

Sensors and other data-gathering devices serve as the foundation of any IoT system. They serve as the link between the physical and digital worlds, converting analogue signals to digital signals.

Sensors are categorized based on the following factors:

·       Active/Passive

·       Invasive/Non-Invasive

·       Contact / No-Contact

·       Absolute or Relative

·       Area of Application

·       Measurement

·       What it measures

1.1.1.     Humidity Sensor:

Humidity sensors detect humidity (amount of water vapour) in the air or a mass. Humidity levels can be measured in various ways: absolute humidity, relative humidity, mass ratio, and so on.

1.1.2.     Temperature Sensor:

Temperature sensors measure the amount of heat or cold that is present in a system. They can be broad of two types: contact and non-contact. Contact temperature sensors need to be in physical contact with the object being sensed. Non-contact sensors do not need physical contact, as they measure temperature through convection and radiation.

1.1.3.     Chemical Sensor (Carbon dioxide):

Chemical sensors measure the concentration of chemicals in a system. When subjected to a mix of chemicals, chemical sensors are typically selected for a target type of chemical (for example, a CO2 sensor senses only carbon dioxide).

1.1.4.     Actuators:

Actuators receive some type of control signal (commonly an electric signal or digital command) that triggers a physical effect, usually some type of motion, force, and so on.

Actuators are categorized based on the following factors:

·       Type of motion: Actuators can be classified based on the type of motion they produce (for example, linear, rotary, one/two/three-axes).

·       Power: Actuators can be classified based on their power output (for example, high power, low power, micro power)

·       Binary or continuous: Actuators can be classified based on the number of stable-state outputs.

·       Area of application: Actuators can be classified based on the specific industry or vertical where they are used.

·       Type of energy: Actuators can be classified based on their energy type.

1.2.         SANET and Smart Object:

This project aims to develop a SANET. Sensor/actuator network (SANET) is a network of sensors that sense and measure their environment and/or actuators that act on their environment. SANET can communicate and cooperating in a productive manner.

1.2.1.     SANET Advantages:

•       Greater deployment flexibility (especially in extreme environments or hard-to-reach places)

•       Simpler scaling to many nodes

•       Lower implementation costs

•       Easier long-term maintenance

•       Effortless introduction of new sensor/actuator nodes

•       Better equipped to handle dynamic/rapid topology changes

1.2.2.     Smart Objects:

A smart object is a device that they consist of the following four defining characteristics in a single unit.

1.        Processing unit
2.        Sensor(s) and/or actuator(s)
3.        Communication device
4.        Power source

We deployed our edge nodes as smart objects that include a processing unit, controller, and sensor.

1.2.3.     Communication Patterns for Sensor Network:

Periodic: Transmission of sensory information occurs only at periodic intervals.

1.3.         Low Power Processing Unit:

The goal of Low Power is to lower the device's power consumption by managing its behaviour to extend its operational lifetime. Electronic equipment that is powered directly from a power source normally do not require the use of Low Power or similar strategies to extend its life. On the other side, it is vital to reduce power consumption to increase the operating lifetime of devices that use a power source such as batteries.

1.3.1.     Low-Power Hardware Technique:

Figure 1. is the representation of the Arduino input voltage schematic. The input pin in the controller and the dc jack input passes through the 5 volts regulator that supplies 5 volts to the universal board. Whereas the voltage input through the USB port passes through the 3.3 voltage regulator that rectify the voltage to 3.3 volts only. Therefore, this project deployed the smart object techniques for the edge nodes.

1. Arduino Low Powered Input Modes

 

1.1.1.     Low Power Microprocessor Libraries:

To enable the Low Power feature on Arduino boards, we can download the Arduino Low Power library[1] from the libraries manager in the Arduino IDE. This package lets you specify how long to sleep, from a few milliseconds to indefinitely. It also lets you define which elements of the micro to disable, making it a very powerful and adaptable low-power library.

1.1.2.     Deep Sleep mode:

Deep Sleep mode allows the device to turn off several internal modules to save maximum power. The Deep Sleep mode can be programmed to wake up after a predetermined amount of time.

 

#include "LowPower.h"

 

void setup()

{

  pinMode(13,OUTPUT);

}

 

void loop()

{

  digitalWrite(13,HIGH);

  delay(2000);

  digitalWrite(13,LOW);

  LowPower.powerDown(SLEEP_2S, ADC_OFF, BOD_OFF);

}

 

2. Sleep Mode Stats [2]

 

1.1.1.      disable reporting ():

The disable_reporting() is a method used to disable the pin reporting method of pyfirmata. It eventually shuts the power supplied to receive and transmit data. It has same effects as deep sleep mode.

2.     Data Analytics:

2.1.          Data Source and Acquisition:

Accurate occupancy detection of an office room from light, temperature, humidity and CO2 measurements using statistical learning models. Luis M. Candanedo, Véronique Feldheim. Energy and Buildings. Volume 112, 15 January 2016, Pages 28-39.

2.2.          Data Preprocessing:

The bulk of the datasets obtained include missing values due to manual errors, lack of information, and other factors. Although there are several methods for dealing with missing values, there are occasions when dropping such rows from the dataset is the only choice. The pandas 'dropna' function is a popular way to remove rows and columns.

Purpose: To remove the missing values, duplicates and unused columns from a Data Frame.

DataFrame - drop() function

The drop() function removes specific labels from rows or columns.

Remove rows or columns by specifying label names and corresponding axes, or by directly specifying index or column names. Labels on different levels can be removed when using a multi-index by specifying the level.

2.3.         Feature Extraction:

2.3.1.     Heatmap ():

A correlation heatmap is a graphical representation of a correlation matrix that shows how different variables are related. Correlation can have a value ranging from -1 to 1.

cmap — sets the specific colormap we want to use,

annot — when set to True, the correlation values become visible on the colored cells.

2.3.2.     Train Test Split Procedure:

A model validation process that allows you to simulate how a model would perform on new/unseen data is known as train test split. This is how the process works.

1.     Ensure that your data is organized in a format suitable for train test split. This is done by splitting the entire dataset into Features and Targets in scikit-learn.

2.     Split the dataset into two pieces: a training set and a testing set. This consists of random sampling without replacement about 30% (you can vary this) of the rows and putting them into your training set and putting the remaining 70% to your test set.

3.     Note that the colors in “Features” and “Target” indicate where their data will go (“X_train”, “X_test”, “y_train”, “y_test”) for a particular train test split.

4.     Train the model on the training set. This is “X_train” and “y_train” in the image.

5.     Test the model on the testing set (“X_test” and “y_test” in the image) and evaluate the performance.

3. Test and Train data splitting

 

1.1.         Decision Tree Classifier:

A decision tree is a flowchart-like tree structure where an internal node represents a feature(or attribute), the branch represents a decision rule, and each leaf node represents the target.

tree_classifier.predict will take predictions from our testing set (X-test) and we will assign those predictions to a variable name y-tree-predicts.

Now we compute the confusion matrix to evaluate the accuracy of classification.

1.2.         Results:

The confusion matrix is a N x N matrix used to evaluate the performance of a classification model, where N is the number of target classes (0,1). The matrix compares the truth goal values to the machine learning model's predictions. This provides us with a comprehensive picture of how well our classification model is working and the kind of errors it is making.

After finding the confusion matrix we will plot these predicted and truth values.

4. Confusion Matrix

Translation Algorithm:

Python-Arduino Interface:

The strength of the python language and the supportive environment of Arduino is the key feature of the project. The project is the practical implementation and real-time integration of the python environment and the Arduino. The framework that connects the other environments with Arduino is called Firmata. Firmata is a protocol that is used to connect the Arduino serial interface with other programming languages. Python requires the installation of Pyfirmata.

pip install pyfirmata

Pyfirmata is the serial connectivity between Python and Arduino. The Base class for any board for pyfirmata is the following:

Class pyfirmata. pyfirmata.Board(port, layout=None, baudrate=57600, name=None, timeout=None)

Following are some functions that are to be used in the project.

Table 12. Pyfirmata Commands

SR. No.	Function	Functionality
1.	get_pin(pin_def)	Returns the activated pin given by the pin definition. May raise an InvalidPinDefError or a PinAlreadyTakenError. Parameter’s pin_def – Pin definition as described below, but without the Arduino name. So for example a:1: I. ‘a’ analogue pin PIN ‘i’ for input ‘d’ digital PIN ‘o’ for output ‘p’ for PWM (Pulse-width modulation)  All are separated by:
2.	servo_config (pin, min_pulse=544, max_pulse=2400, angle=0)	Configure a pin as servo with min_pulse, max_pulse and first angle. min_pulse and max_pulse default to the arduino defaults
3.	setup_layout(board_layout)	Setup the Pin instances based on the given board layout.
4.	disable_reporting ()	Disable the reporting of an input pin. Disable the reporting of the port
5.	enable_reporting ()	Set an input pin to report values. Enable reporting of values for the whole port.
6.	Mode ()	Mode of operation for the pin. Can be one of the pin modes: INPUT, OUTPUT, ANALOG, PWM. or SERVO (or UNAVAILABLE)
7.	write(value)	Set the output pins of the port to the correct state. Output a voltage from the pin Parameters value. Uses value as a Boolean if the pin is in output mode, or expects a float from 0 to 1 if the pin is in PWM mode. If the pin is in SERVO the value should be in degrees.

 

If you use a pin more often, it can be worth it to use the get_pin() method of the board. It lets you specify what pin you need by a string, composed of ‘a’ or ‘d’ (depending on whether you need an analogue or digital pin), the pin number, and the mode (‘i’ for input, ‘o’ for output, ‘p’ for PWM). All are separated by :

Flow Chart:

 

Conclusion:

The tests performed have provided important validation that the rule-based algorithms developed for self-correcting controls for building air handlers are a viable solution, capable of detecting, diagnosing and correcting. The primary goal and benefit of HVAC controls are to provide building occupants with a comfortable environment. The data acquired from the sensors convert into testing and training data. The data trains to model to predict the accurate output that will eventually drive the actuator to open and close the HVAC.

References:

1.      https://www.arduino.cc/reference/en/libraries/arduino-low-power/

2.      https://learn.sparkfun.com/tutorials/reducing-arduino-power-consumption/all