This dissertation develops and implements a methodology for integration, and real-time control of battery energy storage and grid-tie inverter to provide grid services and compensate for the intermittency of renewables in the power system with the minimum use of battery capacity.

Deploying Renewable Energy Sources (RES) has attracted significant attention in recent years. However, integrating RES into the grid makes the planning and operation of power systems more challenging due to the poor controllability and predictability of these resources. Technical concerns associated with large-scale integration of intermittent RES include power quality and power stability issues. In order to facilitate the deployment of RES into the power grid, the dissertation proposes control and integration of a battery energy storage system to compensate for the stochastic nature and rapid fluctuations of RESs. To this end, a new control methodology and integration platform for battery energy storage system is developed which is unique due to its plug and play operation capability which is independent of system parameters, and fast acting characteristic for applications such as real-time solar power smoothing, dynamic voltage regulation, and real-time load uncertainty compensation, while it minimizes the use of battery capacity.

The developed methodology incorporates control techniques, machine learning algorithms, discrete signal processing, and optimization methods for smart operation of battery energy storage systems in the grid. A convex optimization approach is utilized to find the global optimal operating point for battery energy storage system to maximize the performance of the system and minimize the use of storage capacity. The optimal operating point is tracked by a model-free adaptive controller. A discrete signal processing approach adapts the control technique for practical implementation. Machine learning algorithms enable capturing and predicting renewables’ and loads’ stochastic behavior. In this context, in order to capture both linear and nonlinear patterns, Discrete Wavelet Transform (DWT), Auto-Regressive Moving Average (ARMA) models, and Recurrent Neural Networks (RNN) are combined as a hybrid method for super short-term solar power prediction. Second-order cone programming (SOCP) is employed to design a minimum-length, low-group-delay digital Finite Impulse Response (FIR) filter to compensate renewable intermittency and load uncertainty. An extremum seeking algorithm and adaptive PI controller are used to improve the grid voltage profile in the presence of RES and EVs without an explicit model of distribution circuit.

To prove the performance of the proposed concept, a testbed platform including 64kWh battery energy storage system and 36kW grid-tie inverter interacting with 25kW solar PV and 50kW DC fast charger is designed and implemented. Also a prototype of a hardware/software control unit is built which incorporates the proposed algorithms/techniques for real-time managing the power flow of BESS and inverter in four quadrants. The control hardware is configured appropriately for interaction of the controller with the grid-tie inverter, battery management system, and measurement units. The control software is structured to manage the communication, safe operation and monitoring of the system. The main advantage of the proposed methodology is to minimize the use of battery capacity for applications such as voltage regulation, renewable compensation, and load compensation. The proposed controller has been tested over several dynamic scenarios including smoothing the solar power generation, voltage regulation and shaping power demand. The results obtained show a 45 percent reduction in the deployed battery capacity for solar smoothing, an 11 percent reduction in the mean absolute value error of the super short-term solar prediction, and 20 percent improvement in voltage regulation for defined size of storage, compared with their counterparts in the literature.

This thesis investigates the applications of non-parametric approaches for probabilistic demand forecasting in power distribution systems. This thesis develops two probabilistic short-term load forecasting models. We implement and evaluate two type of probabilistic forecasting methods namely: kernel density estimation and mixture density networks. In particular we are interested in the study of the features and (any) advantages of using machine learning approaches over the more traditional approaches in probabilistic demand forecasting.

This thesis gives a short-term load forecast of the residential demand with respect to the outside temperature using the probabilistic forecasting methods. The factors impacting the performance and accuracy of the forecasts are evaluated. The historical data for energy consumption generally has multiple seasonality’s associated with it. For more accurate demand forecasting, it is critical to take into account the different seasonality’s in the data and the effect of exogenous variables (temperature) while developing different models. Both the models are trained separately for yearly and seasonal datasets to study the effect of seasonality on forecasting.

Various tests are performed on the models to assess their statistical significance when compared to one another. The comparative assessment between Mixture Density Networks and Kernel Density Estimation also advances the knowledge of applying these techniques to STLF. The proposed approaches are compared with other benchmark models like ARIMA (1,0,0) model and a neural network which are also trained separately for yearly and seasonal datasets.