**1. Introduction**

In recent years, the penetration of dispersed generations (DGs) such as photovoltaic and wind energy in electricity systems has significantly increased. Some of the distributed generations are natively DC. The developments of DC loads and storage technologies have also considerably increased [1]. The DC grid has much more advantages such as higher system efficiency, easier integration and lower losses due to reduced conversion devices than the AC grid [2–5]. Therefore, the DC grid has been rapidly increasing in the power system. One of the major challenges with DC grid-connected DG is the detection of the unintentional islanding phenomenon. According to the IEEE 929-2000 standard [6], islanding problem occurs when the distributed network isolated from the main utility and the loads are supplied by DGs only. Based on this standard, the islanding phenomenon must be detected within 2 s to prevent damage to electrical devices and ensure the safety of maintenances. Nowadays, islanding detection has received considerable attention through recent studies.

In [7], a novel islanding detection method based on two additional compensators and positive feedback of active and reactive power control loops. The method does not have a non-detection zone, no affect on power quality and normal operation of synchronous DG. Because of it need additional compensators, it may cause the method to complicate and costly.

A novel islanding detection method for micro-grid was proposed in [8]. The method can detect islanding phenomenon in hardest condition (low active and reactive power mismatches) by using a differential morphological filter (a dilation–erosion differential filter (DED)) of the RMS signal (DEDFOR) at the point of common coupling. The method has a small non-detection zone (NDZ), fast detection time, and high accuracy. However, the method is used for double-fed induction generators (DFIGs) system.

A new passive islanding detection includes in Smart Islanding Detector (SmartID) was proposed in [9]. It combines with classical voltage and frequency relays to make a better anti-islanding protection system. The method is used to backup in case of the classical method fail. However, the combination of SmartID and classical relays also complicated.

In [10], a new islanding detection strategy for low-voltage inverter-interfaced microgrids was presented. This strategy was based on adaptive neuro-fuzzy inference system (ANFIS). The ANFIS method monitors seven inputs measured at the point of common coupling (PCC) such as root-mean square of voltage and current (RMSU and RMSI), total harmonic distortion of voltage and current (THDU and THDI), frequency, and active and reactive powers based on practical measurement in a real microgrid. The ANFIS method detects islanding condition by using its pattern recognition capability and nonlinear mapping of relation between input signals.

A novel hybrid islanding detection method for grid-connected microgrids in the case of multiple inverter-based distributed generators was proposed in [11]. This method was based on the slope of linear reactive power disturbance (RPD) and four passive criteria, namely, voltage variation, voltage unbalance, rate of change of frequency and correlation factor between the RPD and frequency variation. Islanding is detected when the frequency exceeds its thresholds. However, the cancellation issue in the case of multiple inverter-based distributed generators was ignored in this study.

In [12], a new method used the transient response of voltage waveform to detect the islanding condition. This method was based on two new criteria, namely, the peak of the transient index value (TIV) and the positive sequence superimposed phase angle at the point of common coupling.

In [13], a method based on Kalman filter (KF) was proposed to extract and filter the harmonic components of the measured voltage signal at distributed generation terminals. The islanding detected by the selected harmonic distortion (SHD) was calculated by the KF after the different changes in the power system. This changes was detected by a residual signal.

An unintentional islanding detection method based on the combination of the main criterion of switch status and the auxiliary criterion based on the no-current and difference-voltage criteria was proposed for "hand-in-hand" DC distribution network in [14]. However, the power supply reliability of the DC distribution network must be considered.

A novel islanding detection approach based on the relationship between the angular frequency and the derivative of the equivalent resistance observed from the small-scale synchronous generators in microgrids, including small-scale synchronous generators, was proposed in [15]. The derivative of the equivalent resistance observed from the small-scale synchronous generators approaches to zero when the islanding occurs in small-scale synchronous generator side.

In [16], we proposed an active islanding detection method by injecting perturbation signal to make the power imbalance and voltage fluctuation exceed the thresholds faster and easier in the islanding condition. This method can rapidly and efficiently detect the islanding condition. Howerver, the cancellation problem when using this method in multi-PV operation has not been considered.

Many converters are connected in parallel because of their capacity limitation in a large-scale solar system. Therefore, the cancellation issue can occur in a multi-PV operation.

On basis of the literature, no research has been conducted on the cancellation issue in multi-DG operation when using active islanding detection method. In this study, the cancellation issue when using the active Islanding Detection Method (IDM) in [16] is analyzed. Moreover, the solution improves the islanding detection method in [16] to eliminate the injected signal cancellation is proposed and verified by Matlab/Simulink. The system in this paper is the DC grid connected-photovoltaic system. The rest of this paper is organized as follows: The system description and cancellation problem are presented in Section 2. The simulation results of the cancellation problem is analyzed in Section 3. The solution for this problem is shown in Section 4. The conclusions are presented in Section 5.
