A battery energy storage systems (BESS) with power electronics capable of reactive power control can be used as a fully controllable 4-quadrant load/generator, and, when controlled intelligently, it can be used to help mitigate a myriad of distribution grid issues. The research goal in this thesis is to find the optimal way to practically model and use BESS as a distributed power resource in various situations. Each situation has brought its own set of real-world constraints such as battery capacity, efficiency, electrical billing, battery state of balance, and operational issues.
Full scale and lab scale experiments were done to explore the operational issues in real hardware battery energy storage systems. In the full-scale results, a stochastic optimization-based framework is developed and implemented, to demonstrate conducting peak load reduction at a distribution feeder using customer owned batteries, under both offline and online control settings. Multiple experimental tests were performed by operating a 1 MWh / 200 kW battery at UCR’s Center for Environmental Research
amp; Technology’s microgrid. They verified the considerable reduction of the feeder peak load achieved based on the proposed framework. They also showed many operational issues that may not be foreseen in computer simulations, such as the need to carefully calibrate battery management system (BMS) state of charge (SoC) estimates and to consider BESS operational efficiency and SoC drift.
In order to more fully study and understand the operational issues a power hardware in loop (P-HIL) testbed is designed and successfully tested to enable accurate modeling and testing of grid-connected BESSes in power system applications. The testbed architecture, hardware components, software systems, communications, and computer controls are explained. The key advantage of the developed testbed is the ability to examine the physical battery-cells, battery-packs, and the BMS, where the modeling is most challenging and the existing models lack accuracy the most. The lab setting also allows for much more control, and the ability to force a bad situation to see how the system responds without a downside. The lab scale tests provide significant insight into the battery cell operation and issues caused by balance and capacity mismatches.
In both scenarios the frameworks were able to noticeably improve operation of the distribution grid. However, both the lab-scale and full-scale experiments showed the need for BESS operation optimization algorithms to use per cell battery models due to variance in the cell’s charge level and capacity. This caused the system to overestimate the BESS capacity. Thus, a pack-based BESS model is developed capable of modeling differences between each cell while still being usable in a BESS scheduling framework. It is shown how this cell-level circuit model, consisting of a voltage source and internal resistance, can be used to construct a battery pack model that is usable in a BESS scheduling framework. Two methods are presented, namely CALVS (Circuit-based Approach with Linear Voltage Source) and CANVS (Circuit-based Approach with Non-linear Voltage Source), to tackle the non-linear relationship between open circuit voltage and SoC. The models are trained using P-HIL test data from a Lithium Ion battery pack, and parameter estimation challenges were identified and resolved. The model is then validated in a peak load shaving experiment. The models are demonstrated to have much more realistic results when comparing the modeled and actual battery operation using the P-HIL testbed. The model presented is general enough that it can be applied to many BESS scheduling frameworks to significantly improve BESS operation on the distribution grid.