Solar Power Generation Anomaly Detection

Project Overview

This project implements anomaly detection for solar power generation systems across two locations (A and B), analyzing sensor data to identify irregular patterns and potential system issues.

Data Description

The project uses two main datasets:

• solar_sensor_data.csv: Contains power generation metrics
• weather_sensor_data.csv: Contains environmental measurements

Key metrics include:

• DC/AC Power
• Daily/Total Yield
• Ambient Temperature
• Module Temperature
• Irradiation

Features

1. Data Preprocessing

• Data preprocessing: Addition of seasonal and time period columns
• Feature engineering: Creation of dummy variables for categorical data
• Sensor-specific analysis: Separate handling for sensors from Locations A and B
• Anomaly detection: Calculations based on z-scores for flagging anomalous readings

2. Model Implementation

Four machine learning models were implemented:

• Linear Regression
• Random Forest Regression
• Gradient Boosting Regression
• Convolutional Neural Network (CNN) Autoencoder
  • Used for anomaly detection through reconstruction error
  • Separate models for each sensor
  • Capable of capturing complex temporal patterns in power generation

3. Anomaly Detection

• Uses multiple approaches:
  • Z-score based calculations
  • CNN Autoencoder reconstruction error
  • Flags readings that deviate by more than 3 standard deviations
• Separate analysis for each sensor location

Model Performance

Linear Regression

• Location A
  • Average Mean Squared Error (MSE): 986.0062
  • Average R² Score: 0.9923
• Location B
  • Average Mean Squared Error (MSE): 15839.6510
  • Average R² Score: 0.8606

Random Forest Regression

• Location A
  • Average Mean Squared Error (MSE): 417.4443
  • Average R² Score: 0.9967
• Location B
  • Average Mean Squared Error (MSE): 317.8780
  • Average R² Score: 0.9974

Gradient Boosting Regression

• Location A
  • Average Mean Squared Error (MSE): 425.1359
  • Average R² Score: 0.9966
• Location B
  • Average Mean Squared Error (MSE): 223.5140
  • Average R² Score: 0.9982

CNN Autoencoder

• Implemented in separate notebooks for each sensor
• Uses reconstruction error to identify anomalies
• Features:
  • Input: Time series of power generation data
  • Architecture: Multiple convolutional layers for feature extraction
  • Output: Reconstructed power generation patterns
  • Anomaly Detection: Based on reconstruction error threshold

Key Findings

1. Location A generally showed better model performance than Location B
2. Random Forest and Gradient Boosting models outperformed Linear Regression
3. Multiple anomalies were detected across different sensors
4. Weather conditions significantly impact power generation patterns

Project Structure

• Data_Preparation.ipynb: Initial data processing and cleaning
• CNN_Autoencoders sensor*.ipynb: Sensor-specific analysis
• Models and results/: Contains model implementations and analysis reports
• LR-RF-GBR.ipynb: Implementation of the three main regression models

Usage

1. Load and preprocess the data using Data_Preparation.ipynb
2. Run individual sensor analysis notebooks:
   • Traditional models: LR-RF-GBR.ipynb
   • CNN models: CNN_Autoencoders sensor*.ipynb
3. Review results in the Reports directory

Results

The project successfully identified anomalies using multiple approaches:

• Traditional Models:
  • Best performance: Gradient Boosting (R² > 0.99)
• CNN Autoencoder:
  • Effective at capturing complex patterns
  • Visual confirmation of anomaly detection through reconstruction plots
  • Sensor-specific analysis providing detailed insights