M.Tech-Thesis-Project

M. Tech. Thesis Project on Detection and Identification of Hybrid Distribution System Using Wavelet Transform and Artificial Neural Networks is documented here.

Abstract

In this thesis, Islanding and Power Quality(PQ) Issues in Hybrid Distributed Generation (DG) System consists of Photovoltaic(PV) system and Wind Power Plant connected to grid through a Point of Common Coupling(PCC), are detected and classified, using Wavelet Transform and Artificial Neural Networks. Wavelet Transform indices are extracted from the Negative Sequence component of the voltage signal at PCC to detect the disturbances. A feature vector is modeled with WT indices and loading of DG system to train Artificial Neural Network. . The proposed method is compared with a conventional method. The results demonstrate the advantages of Wavelet over conventional method in detection and classification of disturbances in the system and robustness in application of Machine Learning (ML) Classifier. The trained ANN is deployed as a Web Service using Microsoft Azure Machine Learning Studio. It enhances the implementation feasibility of proposed method.

Find the published papers in following links

Hybrid Distributed Generation System

Simulation Diagram in MATLAB Simulink

Work Flow

Simulate the simulation_diagram.mdl using script_get_training_data.m for following situations and collect the training data for ANN
- Grid Connected (Normal)
- Islanding
- L-L Fault
- L-G Fault
- Nonlinear Load Switch
From negative sequence voltage at PCC during these events, extract the following features
- Loading of DG
- Standard Deviation of detail coefficients of Wavelet Transform of negative sequence voltage signal at level 3 (SD3)
- Standard Deviation of detail coefficients of Wavelet Transform of negative sequence voltage signal at level 4 (SD4)
- Energy content of detail coefficients of Wavelet Transform of negative sequence voltage signal at level 3 (E3)
- Energy content of detail coefficients of Wavelet Transform of negative sequence voltage signal at level 4 (E4)
ANN is trained at Microsoft Azure Machine Learning Studio by uploading the training data to cloud as a .csv file
- Train the ANN locally using main_program_ann.m from ann_code folder for testing purpose
Sample training data would be as follows

Loading	SD4	SD3	E4	E3	Label
0.91	0.00016	0.000123	0.001321	0.001398	Normal
1.155	0.000165	0.000105	0.001364	0.001199	Normal
1.05	0.046179	0.035238	0.381349	0.403484	Islanding
1.12	0.046168	0.041096	0.38108	0.471473	Islanding
0.98	0.011971	0.004943	0.098714	0.056379	L-G Fault
1.085	0.011999	0.004949	0.098944	0.056448	L-G Fault
1.365	0.029071	0.009113	0.239756	0.10396	L-L Fault
1.715	0.028943	0.009125	0.238704	0.1041	L-L Fault
1.225	0.025061	0.012793	0.20666	0.146597	Non-Linear Load Switch
1.435	0.025036	0.012789	0.206452	0.146566	Non-Linear Load Switch

Using Python GUI, consume the web service as a POST request to predict the states of the system (as mentioned above)
- Python GUI can be with gui_and_azure_api_consumption.py

sleebapaul/masters_thesis_project