Temperature Prediction

10/2025

Full french report

Hardware used

• Arduino Nano 33 BLE Rev2
• DHT11 temperature sensor

Project architecture

.
├── compare_lstm.ipynb
├── data
│   ├── data.csv
│   ...
│   └── last_year.csv
├── data_gathered
│   └── raw.csv
├── eval_model.ipynb
├── gru_generator.ipynb
├── load_modelcpp.py
├── lstm_generator.ipynb
├── mlp_generator.ipynb
├── models
│   ├── arduino_model
│   │   ├── 2layers.tflite
│   │   ├── gru_light.tflite
│   │   ├── model.h
│   │  ...
│   │   └── temperature_model_light.tflite
│   ├── gru
│   │   ├── gru_temperature_model.h5
│   │   └── gru_temperature_model.keras
│  ...
│   └── MLP
│       ├── mlp_temperature_model.h5
│       └── mlp_temperature_model.keras
├── README.md
├── report.pdf
├── scripts_arduino
│   ├── data_gathering
│   │   ├── bluetooth
│   │   │   └── bluetooth.ino
│   │   ├── bluetooth_communication_4canaux.ino
│   │   ├── bluetooth_communication_async.ino
│   │   ├── bluetooth_led.ino
│   │   ├── code_temp.ino
│   │   ├── communication_bluetooth.ino
│   │   ├── sender_other_architecture.ino
│   │   └── senderv1_final.ino
│   ├── main
│   │   └── main.ino
│   ├── noBLE
│   │   └── noBLE.ino
│   └── prediction
│       └── script
│           ├── model.cpp
│           ├── model.h
│           └── script.ino
├── test_quantization_problem.py
└── to_tflite.py

WorkFlow

Data gathering

• Data gathering script
• Data gathered

Models

Making the models

• LSTM
• LSTM comparison
• GRU
• MLP

Evaluating the models

• Evaluation of models

Converting the models

• Convert models

Debugging

• Test quantization
• Test generated C array
• Hello world example

Final: prediction & communication script

• Without BLE
• With BLE

Datasets

Obtained via 3 different ways:

• Historical Weather API
• EC&D
• By hand, leaving the DHT sensor outside