SPARC: IoT Smart Meter with AI Load Identification | Smarter Energy

追加: 2026-05-30

[00:00:08]The bill is too expensive. This is because this man of the bill is too expensive.

[00:00:15]It is because of you because you did not switch off the light.

[00:00:18]>> I know they forgot to switch off the light. Thank the bill.

[00:00:22]>> You should know.

[00:00:23]>> That will be great if we have system that can monitor our electricity in our house.

[00:00:29]Great idea.

[00:00:45]Project focus on developing a low IoT smart meter with AI load identification to help user monitor and manage their electricity usage more effectively.

[00:00:57]Current smart meter usually show only total energy consumption and do not identify which appliance use most electricity. Because of of this user may find it difficult to control their energy usage and reduce waste. To solve this problem, the project shift load monitoring nilm and machine learning tiny ML using an ESP32 microcontroller and PZM 0040 sensor. The system can analyze the power data from a single sensor, identify active appliance, monitor electricity usage in a real time, estimate billing and allow remote monitoring or control. The purpose system is affordable, easy and use and suitable for home. It supports smart energy habits by helping user understand where electricity is being used and take action to reduce unnecessary power consumption.

[00:02:00]Problem statement. There are three major factor that current residential electrical system face on which impact their effectiveness in promoting energy efficiency. First traditional meter only show the total of electricity has been used. They cannot distinguish which appliance is consuming the most energy especially when different device have similar power levels.

[00:02:28]They cannot know where has been wasted.

[00:02:36]Second advanced techniques such as combinational optim optimization co and factorial hyen markoff model FHMM making a high level processing power level making them not suitable for low cost and AI based deployment. Third, while IoT monitoring system exists, view integrate real time or remote control function that enable users to switch their appiance when they cannot be present physically.

[00:03:12]Uh three objective for this project.

[00:03:14]First to build a IoT smart meter with AI model with capable of load identification with more than 90% accuracy real-time energy monitoring and remote control of an ESP32 microcontroller. Second objective to simulate the functionality of smart meter like load identification and realtime monitoring using protails.

[00:03:39]Objective and the last one to evaluate the performance of the system in term of accuracy and respond time of lot identification realtime energy realtime energy monitoring automatic height usage alert and cloud logging.

[00:03:55]Spark is an IoT based system designed to help users monitor their electrical energy consumption in real time. At its core, the system uses the BZM000040 sensor to accurately measure the key electricity parameters such as voltage and current, power and energy consumption. The ESP32 microcontroller serve as the brain of the system since it handle the data processing and enabling wireless connectivity through Wi-Fi for appliances control. The system includes a relay module which acts as a switch to turn devices on and off as needed. The Bling mobile application provides a userfriendly interface for remote monitoring and control, enabling users to access the system from anywhere via the smartphone.

[00:04:53]As I turn on this electrical socket, Spark system will continuously monitor the electrical consumption in real time.

[00:05:02]Total consumption also can be seen in the blink.

[00:05:11]For the remote control function, I will control this lamp by turning it on and off by just using virtual switch on blink apps.

[00:05:27]Similarly, I can control the charging of this laptop remotely by just using virtual button or blink app.

[00:05:43]Next, Spark also have notification feature which set automatic alert when the total consumption of electric has exceeded the preset threshold. The consumption threshold can be adjusted according to customer did by using Blake app or Blake platform on laptop.

[00:06:13]Last but not least, Spark System also automatically record the consumption data and log it into Google sheet for further analysis.

[00:06:27]This process is started by collecting the electrical data from the several household appliances using the Piz EM000040 sensor. This sensor will be measured some of the important parameters such as voltage, current, power, power factor and also the frequency.

[00:06:48]The appliances tested including the laptop charger, the table fan, haird dryer and no lot condition. This data is collected with the sampling rate of two hertz which is very suitable for the general household lot identification.

[00:07:05]After the data is collected so it will display the raw data graph here as what we want. So we have the voltage, current, power per vector and also the frequency.

[00:07:19]After that after we done with the process so we will have the DSP result and the filter response. This is the filter respond before. This is the after the filter and also we can see the successful of accuracy for our model by using looking at this confusion matrix. So this is the step where we can see it is 100% accuracy of our training and only loss of 0.2.

[00:07:52]So for the data explorer also we can look at this part. So we have the correct of the accuracy data acquired from the system which is has the sampling rate of the two hertz. So this is the AI model that we use to train our data that is using the S impulse. Based on the prototype testing the AI based load identification achieve an overall accuracy of 95%.

[00:08:24]The system successfully identifies several loads including laptop, table fan and hand dryer. For system respond, the relay switching time was below 500 milliseconds under strong Wi-Fi condition. The AI load identification respond was below 2 seconds under strong Wi-Fi. This result show that the system is functional, responsive and suitable for realtime energy monitoring.

[00:08:57]>> Additional method that only shows the electricity user. This project shows the combination of provide the realtime monitoring and AI based blood identification to help users understand which appliance consume energy.