Ninad Kiran Gaikwad

The project involves developing and comparing computationally inexpensive black/grey-box developing models (neural network architectures and Bayesian estimation methods) for residential/commercial buildings where data comes from EnergyPlus and other open-source building data repositories like PecanStreet. Then a simulation framework has to be developed to co-simulate these building models at scale with OpenDSS (along with HELICS) to aid the development of both single-building and aggregator-level intelligent controllers which can optimize the energy consumption of buildings for grid support. Currently, we are pursuing model estimation and development of the co-simulation platform.

Course project for Analysis of Power Systems (EE521). A Julia-based package is being developed to perform Newton-Raphson-based power flow, continuation power flow, power system static state estimation, and basic power system optimization. Currently, power system stability analysis and transient simulation capabilities (EE523) are being implemented.

Course project for Introduction to Network Science (CPTS591). A power system transients simulator with closed-loop control and inter-node communication capabilities was developed in Python in a modular fashion. A comparison was done of the capability of degree centrality, PageRank, and eigenvalue-based analysis for accurately predicting the criticality of the power system nodes about their impact on the performance of a distributed frequency control algorithm.

Where during grid outage scenario smart houses with PV, Battery storage, EVs and smart loads will be capable of managing their energy based on optimal control and reinforcement learning. MPC and RL-based central controllers for a single house have been developed. Currently work on centralized and distributed architectures based on MPC and RL for energy resiliency of community of houses is being pursued.

Course project for Machine Learning (CAP6610). Generative Adversarial Networks and Variational Autoencoder networks were trained on the MNIST dataset to generate handwritten digits. Two types each of the GAN and VAE were trained one with dense layers and the other with CNN layers, the implementation was done using the TensorFlow library in Python.

Course project for Optimal Control (EML6934). The Linear Tangent Steering Control and Robot Arm Control problems were formulated as optimal control problems and solved numerically using MATLAB. For the indirect method, a Hamiltonian Boundary Value Problem (HBVP) was formulated through optimality conditions arising from the calculus of variations, and for the direct method, Collocation was used by formulating a Nonlinear Program (NLP). The NLP was formulated in MATLAB and solved using IPOPT.

Course project for Control Theory (EML5311). System identification of an unknown plant with sensor noise was conducted using the Sine-Sweep technique through simulations in MATLAB. The estimated transfer function was converted to a minimally realized state-space model for designing a Linear Quadratic Regulator (LQR) for set-point tracking using MATLAB.

Course project for Optimal Estimation and Kalman Filtering (EML6352). ARMA models based on Least Squares and Maximum Likelihood Estimation techniques were developed and implemented in MATLAB and compared against the ARMA models of MATLAB’s Econometrics toolbox, for forecasting solar power generation from a real-world dataset. The effect of different ARMA models, amount of training data, and prediction on different timescales was studied.

Worked with Dr. Anupama Kowli in the Electrical Department of IIT-B to develop data fault detection algorithms for real-building data collected using Raspberry-Pi-based sensors deployed in one of the lecture halls. The methods applied were SVM, ANN, Wavelets, PCA, and a hybrid PCA-Wavelet. All the algorithms were developed in MATLAB in a modular manner.

Master’s Thesis project, in which an entire software for solar and wind energy estimation and forecasting was created in MATLAB using GUI. The software can generate plant-level energy estimation capability for both wind and solar generation plants. The software also has a weather and generation data preprocessing system. Forecasting using ANN and ARIMA can be done using their respective GUI interfaces. Forecasting using WRF (NWP model) is also automated by developing BASH Shell scripts and running it on a cluster of four RaspberryPi-2 micro-computers.

Individual project, in which a research paper on the ANN-based DTC control strategy for the DFIG was studied, a simulation on the same was created in Sim PowerSystems Matlab, and an IEEE-style report was prepared. Gained valuable experience in decoding a research paper and simulation methodology. A seminar on the same was presented before the faculty of the electrical department.

Individual project, in which a research paper on the vector control strategy for the DFIG was studied, a simulation on the same was created in Sim PowerSystems Matlab, and an IEEE-style report was prepared. Gained valuable experience in decoding a research paper and simulation methodology. A seminar on the same was presented before the faculty of the electrical department.