Control of a utility scale photovoltaic generator for frequency response and fault ride through capability

Abstract

A utility scale i.e. large sized PV generator is integrated in a high voltage power system through an inverter and appropriate step up transformers. This thesis addresses the problem of deriving supports from the inverter of such a PV generator during the phenomena of moderate change in frequency and low voltage in the grid system. These features termed frequency response and low voltage ride through (LVRT) are needed respectively when a moderate change in load or conventional generation occurs suddenly or a fault occurs in the vicinity of PCC (Point of Common Coupling) at which a PV generator is inverter-interfaced with the grid. Also the thesis considers real power supports from the neighbouring conventional generators in case a cloud sweeps over a large PV generator curtailing its output. Maintaining grid frequency and PCC voltage at target values are important if a PV generator has to remain connected with the grid satisfying the criteria of IEEE Standard 1547. Part of this responsibility needs to be shared by a PV generator itself.

The existing works reported in literature mainly addressed the LVRT capability of grid embedded large PV generators for a fault duration and/or voltage sag less than that prescribed by IEEE 1547 standard. However, a few works focused on deriving supports from conventional generators and load management instead of PV generators for any amount of frequency deviation. In both cases the reported works did not consider the control aspect at all or did not mention clearly how the input control signals to the inverter or conventional generator’s governor are obtained in real time.

In this research a new methodology has been proposed (i) for updating real and imaginary components of the reference voltage of a PV generator inverter and (ii) for determining selected conventional generators’ governor actuation time. The first one ensures that the PV generator provides precise real or reactive power supports respectively for moderate frequency deviation in the system and low voltage at PCC. The second one ensures that the conventional generators will increase their output to compensate for the decrease in PV generator output due to cloud sweeping. The method focuses on developing linear regression of a set of four input variables with each of the output signal. The input variables (signals) selected have the impact on the output variables and are easily measurable and/or on-line obtainable from control centre. The three input variables are pre-disturbance real and reactive power outputs of the PV generator, real power demand of the system while the fourth input variable is deviation in frequency or PCC voltage or change in PV generator output depending upon the disturbance. The output signals are real and imaginary components of the reference voltage of a PV generator inverter for frequency response and low voltage phenomena while actuation time of conventional generators’ governor for cloud sweeping phenomenon. One of the advantages of the method is that three input signals are common for each output signal. This will reduce real time data acquisition cost and processing time. The regression equations have been developed using a set of training patterns having diverse range of values regarding input and output signals.

The proposed method has been applied considering a 200 MWpk PV generator embedded in the standard IEEE 39-bus test system i.e. New England System with an approximate day peak load of 6097 MW day peak and in a real-life system i.e. 490 bus 156-generator Bangladesh Power System (BPS) with a day peak of about 9963 MW. All three events were considered i.e. frequency deviation, low voltage arising from a fault for the IEEE 1547 prescribed duration, and cloud sweeping. The training and test patterns are created at simulation stage using DIgSILENT 15.1 software.

The dynamic stability analysis using output control signals obtained from proposed regression method in one go has been compared with the dynamic stability analysis of the same test patterns setting output signals (i.e. inverter reference voltage or governor actuation time) through trial error. The regression based method was found much faster i.e. real time compatible than a trial-error based method and more effective in achieving the target for respective events.