PLAID_nilm_classification

Non-Intrusive Load Monitoring (NILM) with the PLAID Dataset: Time Series Classification for Appliance Detection

-Fasna Sharaf, Adithya Thonse, Tushar Sharma, Abhijeeth Pal, Jaswanth Jadda

---

Overview

Non-Intrusive Load Monitoring (NILM) is a smart machine learning technique used to identify and monitor the usage of individual electrical appliances without the need for separate sensors on each device. Instead, NILM analyzes the overall power consumption data collected from a single point, such as a smart meter.

By studying patterns in this time series data, NILM can detect which appliances are active at any given time. For example, it can determine if a washing machine, refrigerator, or air conditioner is running—just by analyzing the power usage patterns.

This sensorless approach has wide-ranging applications:

1. Factories: Monitor energy usage to optimize costs and improve efficiency.
2. Electric Vehicles: Track power distribution across components.
3. Households: Analyze appliance usage to reduce electricity bills and promote energy efficiency.

NILM is a powerful tool for sustainability, helping reduce energy consumption and carbon footprints, all while relying on easily measurable data, without the need for expensive or intrusive sensors.

About Dataset

The PLAID dataset is a publicly available dataset designed for appliance identification tasks. It contains voltage and current measurements sampled at 30 kHz from 11 different appliance types commonly found in households.

The classes here correspond to 11 different appliance types:

1. Air Conditioner
2. Compact Flourescent Lamp
3. Fan
4. Fridge
5. Hairdryer
6. Heater
7. Incandescent Light Bulb
8. Laptop
9. Microwave
10. Vacuum
11. Washing Machine

In this dataset, each class directory contains files corresponding to the voltage and current measurements when only that appliance class is turned on. While the dataset is extensive, we have taken a subset of it to use with ModelMaker. You can find the dataset subset here: PLAID_submetered_dataset

Usage in Tiny ML ModelZoo

Here is the command to run the yaml file with TinyML ModelMaker:

./run_tinyml_modelzoo.sh examples/PLAID_nilm_classification/config.yaml

Users can configure the model pipeline using a YAML configuration file (like shown in the command above), where different stages (dataset loading, data processing and feature extraction, training, testing, and compilation) can be enabled or disabled based on requirements.

Configuring the YAML File

common and dataset section

common:
    target_module: 'timeseries'
    task_type: 'generic_timeseries_classification'
    target_device: 'F28P55'
    run_name: '{date-time}/{model_name}'
dataset:
    enable: True
    dataset_name: plaid_submetered
    input_data_path: https://software-dl.ti.com/C2000/esd/mcu_ai/datasets/plaid_nilm_submetered_dataset.zip

The important parameters here are: task type generic_timeseries_classification has been selected. The target device is set to F28P55, ensuring that ModelMaker generates a compiled output that runs on this device. Dataset loading is enabled, and the correct dataset path to the zip folder is provided.

data_processing_feature_extraction section

We have configured to include feature extraction before training so that the model can better predict the correct appliance.

data_processing_feature_extraction: 
    feat_ext_transform: ['FFT_FE', 'FFT_POS_HALF', 'WINDOWING', 'BINNING', 'NORMALIZE', 'ABS','LOG_DB', 'CONCAT',]
    frame_size: 32
    feature_size_per_frame: 8
    num_frame_concat: 8                                                            
    variables: 2

training, testing and compilation section

training:
    enable: True
    model_name: 'CLS_13k_NPU'
    model_config: ''
    batch_size: 256
    training_epochs: 30
    num_gpus: 0
    quantization: 2
    learning_rate: 0.04

testing:
    enable: True

compilation:
    enable: True
    keep_libc_files: True

In this configuration, we have used the CLS_13k_NPU model, which is a generic classification model available in our model zoo with approximately 13k parameters. This model consists of 6 Conv+BatchNorm+ReLU layers followed by a Linear layer.

We have configured other training parameters, such as an appropriate learning rate (0.04), batch size (256), and the number of training epochs (30). The quantization is set to 2, which means the model uses TI-NPU quantized operations for efficient inference on TI hardware.

Both testing and compilation are enabled. We are compiling this example using the ti-npu hard preset, which runs on the TI-NPU with full quantization.

Results

On running the YAML file, the following accuracies were observed:

I. Float Train

• Best Epoch: 29
• Accuracy: 98.521%
• F1-Score: 0.985
• AUC ROC Score: 0.985

Confusion Matrix:

Ground Truth \ Predicted As	Air Conditioner	Compact Fluorescent Lamp	Fan	Fridge	Hairdryer	Heater	Incandescent Light Bulb	Laptop	Microwave	Vacuum	Washing Machine
Air Conditioner	3222	1	39	0	0	13	11	14	0	0	0
Compact Fluorescent Lamp	0	3218	1	0	13	6	10	27	5	20	0
Fan	35	0	3250	0	0	0	12	3	0	0	0
Fridge	1	4	4	3226	24	13	1	3	2	12	10
Hairdryer	0	9	9	0	3255	1	14	9	0	0	3
Heater	1	5	12	0	0	3275	0	7	0	0	0
Incandescent Light Bulb	2	0	4	1	4	1	3286	1	0	0	1
Laptop	0	16	3	0	2	0	7	3268	0	1	3
Microwave	0	2	0	0	7	0	0	0	3288	3	0
Vacuum	0	5	11	18	1	0	1	0	1	3263	0
Washing Machine	0	4	7	62	4	0	0	8	1	2	3212

---

II. Quant Train

• Best Epoch: 2
• Accuracy: 98.518%
• F1-Score: 0.985
• AUC ROC Score: 0.985

Confusion Matrix:

Ground Truth \ Predicted As	Air Conditioner	Compact Fluorescent Lamp	Fan	Fridge	Hairdryer	Heater	Incandescent Light Bulb	Laptop	Microwave	Vacuum	Washing Machine
Air Conditioner	3198	1	63	0	0	12	15	11	0	0	0
Compact Fluorescent Lamp	0	3228	1	0	13	8	10	17	4	19	0
Fan	8	1	3283	0	0	0	4	4	0	0	0
Fridge	0	5	4	3225	25	13	1	1	2	12	12
Hairdryer	0	15	11	0	3246	2	15	7	0	0	4
Heater	2	9	14	0	0	3271	0	4	0	0	0
Incandescent Light Bulb	0	1	9	1	5	0	3283	1	0	0	0
Laptop	0	18	2	0	1	0	10	3265	0	1	3
Microwave	0	2	0	0	6	0	0	0	3288	4	0
Vacuum	0	7	11	19	0	0	0	0	0	3263	0
Washing Machine	0	4	5	66	6	2	0	6	1	4	3206

---

III. Test Data

• Evaluation Accuracy: 97.72%
• AUC ROC Score: 0.996

Confusion Matrix:

Ground Truth \ Predicted As	Air Conditioner	Compact Fluorescent Lamp	Fan	Fridge	Hairdryer	Heater	Incandescent Light Bulb	Laptop	Microwave	Vacuum	Washing Machine
Air Conditioner	1076	0	23	0	0	0	1	0	0	0	0
Compact Fluorescent Lamp	0	1074	0	0	3	0	1	20	0	1	1
Fan	27	5	1064	0	0	0	4	0	0	0	0
Fridge	0	0	0	1099	0	0	1	0	0	0	0
Hairdryer	0	0	0	0	1100	0	0	0	0	0	0
Heater	5	0	0	0	0	1090	2	3	0	0	0
Incandescent Light Bulb	0	6	0	0	55	5	1016	2	0	15	1
Laptop	0	10	5	0	7	7	23	1047	0	0	1
Microwave	1	2	6	0	0	0	1	8	1068	8	6
Vacuum	0	0	0	0	1	0	0	9	0	1090	0
Washing Machine	0	0	0	0	0	0	0	0	0	0	1100

You can find the compiled model at: tinyml-modelmaker/data/projects/{dataset_name}/run/{date-time}/{model_name}/compilation

Running on Device

After successfully running Modelmaker, you will get four main files:

1. Artifacts:

   • mod.a and tvmgen_default.h are generated and stored in:

     data/projects/{dataset_name}/run/{date-time}/{model_name}/compilation/artifacts
     

2. Golden Vectors:

   • user_input_config.h and test_vector.c are stored in:

     data/projects/{dataset_name}/run/{date-time}/{model_name}/training/quantization/golden_vectors
     

These four files will be needed while running on device.

In this example, we will use the following setup:

• Device: LAUNCHXL-F28P55X
• C2000Ware Version: 6.00
• Code Composer Studio (CCS): Version 20.3.0

Steps to run this example on-device can be found by following this guide: Deploying Classification Models from ModelMaker to Device

Upon flashing and running the project we can see the model output matches the golden vectors.

---

Performance Metrics

Here are the key performance metrics for the model running on the device:

Metric	Value
Device Name	F28P55x
AI Model Cycles	190571
Inference Time	1270.47 µs
Results Match	TRUE
Model Size	14036 bytes
Model Flash	13032 bytes
Model SRAM	1004 bytes
Application Size	719 bytes
Application FLASH	580 bytes
Application SRAM	139 bytes

Flash Usage

SRAM Usage

---

Update history: [23rd Dec 2025]: Compatible with v1.3 of Tiny ML Modelmaker