A FAST QUASI-STATIC TIME SERIES SIMULATION METHOD USING SENSITIVITY ANALYSIS TO EVALUATE DISTRIBUTED PV IMPACTS

SUMMARY

The desire to reduce carbon footprint coupled with government incentives has led to a massive deployment of renewable energy resources, solar photovoltaics (PV) in particular. Electric utilities in the United States face the challenge of numerous solar PV interconnec- tion requests filed by customers, which seems to increase every day. The approval process for an interconnection request often requires a detailed impact analysis to ensure that the installed resource will not adversely affect the reliability of the distribution grid. Currently, impact analysis for a single PV system using high resolution Quasi-Static Time Series (QSTS) simulation can take anywhere between 10-120 hours to complete, which becomes an infeasible option for utilities, considering the ever increasing number of interconnection requests. To streamline the process, utilities use static scenario-based simulations to de- termine the maximum hosting capacity of the distribution feeders. However, these hosting capacity estimates don’t take into account the voltage regulation (VR) equipment within the feeder. The VR devices, such as tap changing transformers and switched capacitor banks, have the capability to maintain the feeder voltage profile within the ANSI C84.1 limits. Therefore, excluding their impacts while determining the hosting capacity estimates can produce overly conservative results. Consequently, states with aggressive renewable port- folio standards, such as Hawaii and California, are inadvertently limiting their solar PV deployments by relying solely on these conservative estimates.

The goal of this dissertation is to develop a fast, robust, scalable and accurate QSTS analysis tool that can overcome the limitations of existing PV impact evaluation techniques. This work starts by performing a critical analysis of the currently used methods for PV integration studies, and then makes an argument as to why the QSTS analysis is the way forward. The novel algorithm developed in this work leverages the local linearity of the AC power flow manifold to compute the sensitivity coefficients, using a regression-based framework. Furthermore, due to the discontinuities caused by the VR equipment in the manifold, these coefficients are recomputed for every new observed state of the VR devices. A unique aspect of the proposed algorithm is its ability to decouple the estimation of the states of VR devices and the bus voltage profiles. As a result, the PV impact metrics that are sensitive to the operation of VR devices are estimated at a very high resolution time- step, while a coarser resolution is used for the rest. By using an abstract notion of a VR device, the algorithm can also model the smart inverter dynamic real and reactive power control including Volt-VAR and Volt-WATT operational modes. Apart from the voltage- related impacts of PV systems, the proposed algorithm is capable of accurately estimating the maximum thermal loading, line losses, and overload duration of the distribution system infrastructure.

The novel algorithm developed in this dissertation is evaluated on a variety of 3-phase, unbalanced distribution feeders with multiple VR devices including load tap changing transformers, line voltage regulators, switched capacitor banks and smart inverters. Both the standard IEEE test circuits and utility-scale distribution feeders, with low voltage sec- ondary side modeled, are used for evaluating the robustness of the algorithm. In addition, actual feeder SCADA data and on-field irradiance sensor measurements were used to syn- thesize the 1-second resolution, yearlong load and PV time series profiles used in this work. The proposed algorithm shows an average speed improvement of around 150 times, when compared to the traditional brute-force QSTS method, and is able to maintain high accu- racy levels across a variety of different PV impact metrics. Finally, the scalability of the algorithm is also established in terms of its ability to simulate any number of input time series profiles and VR devices.