Power Consumption Monitoring & Prediction System

This project presents a smart solution for real-time monitoring and forecasting of power and energy usage using an ESP32-based IoT setup integrated with cloud logging, machine learning, and a web dashboard.

---

🔌 Hardware Setup Guide

This section covers the hardware components, connection schematics, firmware deployment on ESP32, and linking the system to Blynk, ThingSpeak, and Telegram Bot for real-time monitoring and alerting.

---

🧰 Required Components

Component	Purpose
ESP32 Dev Board	Main microcontroller (Wi-Fi enabled)
ACS712 Sensor	Current measurement
ZMPT101B Sensor	Voltage measurement
16x2 LCD + I2C	Real-time display of voltage, current, power
Breadboard, Wires	Circuit connections
5V Power Supply	Powers ESP32 and peripherals

---

📌 Pin Connections

Component	ESP32 Pin	Notes
ACS712 (OUT)	GPIO 34 (A0)	Analog input for current
ZMPT101B (OUT)	GPIO 35 (A1)	Analog input for voltage
LCD (SDA, SCL)	GPIO 21, 22	I2C communication
GND	GND	Common ground
VCC (all)	5V / 3.3V	Use logic-level compatible

---

⚙️ Step-by-Step Setup

1. 🚀 Upload Code to ESP32

File: sketch_mar7a_copy_20250325014610.ino

Steps:

1. Open Arduino IDE.

2. Install board support: ESP32 by Espressif via Board Manager.

3. Select board: Tools → Board → ESP32 Dev Module

4. Install necessary libraries:

   • WiFi.h
   • HTTPClient.h
   • LiquidCrystal_I2C
   • EmonLib

5. Edit the code to insert:

   • Your Wi-Fi SSID and password
   • Your Google Apps Script URL (for Sheets logging)
   • Blynk Auth Token (see next section)
   • ThingSpeak API Key
   • Telegram Bot token and chat ID

6. Connect ESP32 via USB and click ✅ Upload

---

2. 📱 Connect to Blynk (IoT Mobile App)

Steps:

1. Download Blynk IoT app from Play Store/App Store.

2. Create a new project.

3. Add widgets:

   • 1x LCD display
   • 1x Gauge for current
   • 1x Gauge for voltage
   • 1x Gauge for power
   • 1x Notification (for alerts)

4. Copy the Auth Token from project settings.

5. Paste it into your ESP32 sketch (char auth[] = "your_token";)

6. Upload code again and monitor values on the app.

---

3. 🌐 Send Data to ThingSpeak

Steps:

1. Sign in at https://thingspeak.com

2. Create a new channel with fields:

   • Voltage
   • Current
   • Power

3. Copy your Write API Key

4. In your ESP32 code, update:

   String apiKey = "YOUR_API_KEY";

5. ThingSpeak will now log and plot data automatically.

---

4. 🤖 Telegram Bot Integration (Alerts)

Steps:

1. Open Telegram and search for BotFather

2. Run /newbot and follow prompts to get your bot token

3. Note the Bot Token and update in code:

   String botToken = "YOUR_BOT_TOKEN";

4. To get your chat_id:

   • Search for userinfobot on Telegram
   • Send /start, it will return your chat ID
   • Add that chat_id to your ESP32 code

5. Your ESP32 can now send alerts when:

   • Power exceeds a threshold
   • Sudden surges/drops are detected

---

5. 📊 Google Sheets Logging

Steps:

1. Open Google Sheets → Extensions → Apps Script

2. Paste code from logSensorData.gs

3. Deploy as web app:

   • Deploy → New deployment
   • Choose type: Web App
   • Execute as: Me
   • Who has access: Anyone
   • Click Deploy → Copy URL

4. Update this URL in your ESP32 code:

   const char* scriptUrl = "YOUR_DEPLOYED_SCRIPT_URL";

5. The ESP32 now pushes live data to your Google Sheet!

---

💻 Software Architecture & Usage Guide

This section explains how real-time power consumption data is collected, processed, used to train a machine learning model (LSTM), and how it is then used for forecasting future power usage through an interactive web app interface.

---

🗂 Project Directory Structure

Software-Setup/
│
├── app.py                       # Flask web app to run prediction & UI
├── model_training.py            # LSTM model training script
│
├── monitor_data.csv             # Historical training data (from Google Sheets)
├── current data.csv             # Real-time uploaded input for prediction
├── power_units_lstm.keras       # Trained LSTM model file
├── scaler.pkl                   # MinMaxScaler used to normalize input/output
│
├── static/                      # Auto-generated output files and assets
│   ├── training_history.png     # Loss curve of model training
│   ├── prediction_plot.png      # 48-hour forecast graph (PNG)
│   ├── uploaded_data.csv        # Saved original user-uploaded data
│   ├── prediction_data.csv      # Saved forecast data
│
├── templates/                   # Web UI HTML templates (Flask)
│   ├── index.html               # Main upload interface
│   ├── results.html             # Forecast display + AI Assistant chat
│
├── uploads/                     # Stores incoming user CSV files
└── README.md                    # Documentation (you are reading it)

---

🔁 Data Flow (End-to-End)

🟩 1. Data Collection from Device

• Source: The ESP32 reads voltage and current values using ACS712 and ZMPT101B.

• Processing:

  Power = Voltage × Current
  Units = Power × (duration / 1000)
  

• Logging: Data is sent to Google Sheets every few seconds using logSensorData.gs.

• Export the Sheet manually as monitor_data.csv for model training.

---

🧠 2. Model Training with model_training.py

Features Used:

• POWER
• UNITS

Preprocessing:

• Combines DATE and TIME into a datetime index
• Interpolates missing values
• Normalizes values with MinMaxScaler
• Uses 24-hour sliding windows for time series

Model:

• LSTM (3-layer)
• Dropout regularization
• Optimizer: Adam
• Loss: Mean Squared Error
• Early stopping enabled

Outputs:

• power_units_lstm.keras: Saved model
• scaler.pkl: Scaler object
• training_history.png: Loss curve

Run with:

python model_training.py

---

🌐 3. Web App (Prediction Interface)

Backend: app.py (Flask)

Workflow:

1. User uploads a .csv (e.g. current data.csv)

2. System:

   • Validates columns:

     DATE, TIME, VOLTAGE, CURRENT, POWER, UNITS
     

   • Extracts last 24 hours

   • Scales data using scaler.pkl

   • Predicts next 48 hours of POWER and UNITS

3. Saves:

   • Predictions to prediction_data.csv
   • Graph to prediction_plot.png
   • User data to uploaded_data.csv

4. Displays output on results.html with AI assistant

---

🧠 AI Assistant (Optional)

• Integrated via Gemini 1.5 API

• Handles queries like:

  • "What’s the expected usage tomorrow?"
  • "When is peak load?"
  • "How can I reduce power?"

---

🖥️ How to Run the Web App

Install Dependencies:

pip install flask pandas numpy tensorflow scikit-learn joblib matplotlib google-generativeai

Launch:

python app.py

Visit: http://localhost:5000

---

📤 Input File Format (Required Columns)

Column	Type	Description
DATE	string	Format: DD/MM/YYYY
TIME	string	Format: HH:MM:SS
VOLTAGE	float	Measured voltage
CURRENT	float	Measured current
POWER	float	Calculated power (V × I)
UNITS	float	Energy units (kWh)

• Minimum rows: 24 (hourly records)

---

📊 Outputs

• Graph: Forecast of POWER and UNITS for 48 hours
• Table: Timestamped values
• CSV: Saved prediction results
• Assistant: Interprets trends and gives suggestions