Design of robust intelligent protection technique for large-scale grid-connected wind farm

2.1 Mathematical modelling

As shown in Fig. 1, the DFIG wind turbine consists of different main components such as wind turbine rotor, gearbox, generator, power electronic converters, and transformer for grid connection. The wind turbine is used to convert the wind kinetic energy into mechanical energy, where it is converted to electrical power via electrical generator. The available wind power that can be extracted from the wind is expressed as follows [40, 41]:

$$ {P}_w=0.5\kern0.5em \rho\ A\ {V}_w^3 $$

(1)

where P _w is the wind power, ρ is the air density, A is the cross-sectional area of the wind crossed the blades, and V _w is the wind speed. The developed mechanical power by the wind turbine is given by:

$$ {P}_m=0.5\ {C}_P\left(\lambda, \beta \right)\ \rho\ A\ {V}_w^3 $$

(2)

$$ \lambda =\frac{\omega_r\ {R}_b}{V_w} $$

(3)

where P _m represents the mechanical power, C _P represents the power coefficient, λ represents the tip-speed ratio, β represents the pitch angle, R _b represents the blade radius, and ɷ _r represents the angular speed of turbines rotor.

The stator windings of DFIG are connected to the grid, while the rotor windings are fed via AC/DC/AC converters such as RSC and GSC. In addition to the converters, DC-link capacitor is connected between them to acts as energy storage and keeps the DC voltage ripple with some little variations. The RSC works at variable frequencies depending on available wind speed and the GSC works at grid frequency. The power flow direction via converters depends on the operation mode of the electrical generator. In the super-synchronous operation mode, the stator output power and the rotor slip power are delivered into electrical grid. In the sub-synchronous operation mode, the stator of DFIG delivers power to the grid and the slip power to the rotor via the slip rings and the converters.

2.2 Control systems

The control system strategy of DFIG wind turbine is designed based on controlling of RSC and GSC. Also, it is designed to regulate the blades pitch angle which regulates the speed of wind turbine and protects the wind turbine mechanical parts from damage. Usually, the control systems are designed using proportional integral (PI) controller due to simple structures, clear functionality, low cost and reliable performance [42–46]. Firstly, the RSC is utilized to control active power and reactive power (or voltage level) measured at generator terminals. The control of active power is used to adjust the angular speed of turbine rotor to follow power-speed characteristic of turbine for tracking the maximum power point. The reactive current which is flowing in a converter is used to control voltage level or reactive power at the DFIG terminals. Secondly, voltage level at the DC-link capacitor is controlled using the GSC controller. Also, it can be controlled to absorb or generate reactive power for supporting the grid voltage. Thirdly, wind turbine is provided with a pitch angle control system to limit the extracted power during the condition of high wind speed. The pitch angle control system is activated only when the rotor speed increasing than the rated value due to increasing of wind speed or fault occurring.

3.1 Studied large-scale wind farm

The rated of studied large-scale wind farm is 120 MW, where it consists of 60 wind turbines driven by DFIG, where each wind turbine has a capacity of 2 MW. The wind turbines are positioned in regular matrix of 10 by 6 where each 10 wind turbines are located in a row as shown in Fig. 2. The wind turbines are simulated by a separated 6 generators each of them has a capacity of 20 MW (2 MW multiplied by 10). The generated power is exported to a 220 kV electrical grid via internal feeders and a 25 kV transmission line has a length of 30 km. A single line diagram of the studied wind farm connected grid is illustrated in Fig. 3. Also, the main data of DFIGs, transformers, internal feeders, and transmission line are described in Table 1.

Parameters of DFIG
Generator rated power (MW)	2
Generator rated voltage (V)	690
Resistance of the rotor (pu)	0.005
Leakage inductance of the rotor (pu)	0.155
Resistance of the stator (pu)	0.00706
Leakage inductance of the stator (pu)	0.1709
Mutual inductance (pu)	2.9
Lumped Inertia Constant (s)	5.04
Frequency (Hz)	60
The parameters of the power transmission line (25 kV)
Zero sequence resistance (Ω/km)	0.4129
Positive sequence resistance (Ω /km)	0.11529
Zero sequence inductance (H/km)	0.00331
Positive sequence inductance (H/km)	0.001049
Zero sequence capacitance (F/km)	5.01e-9
Positive sequence capacitance (F/km)	11.329e-9
Internal feeders (Z 1 , Z 2 , ……, Z 6 ) parameters
Zero sequence resistance (Ω /km)	0.3963
Positive sequence resistance (Ω /km)	0.1153
Zero sequence inductance (H/km)	0.00273
Positive sequence inductance (H/km)	0.00105
Wind turbine transformer parameters
Voltage ratio (kV)	0.69/25
Resistance (pu)	0.0081
Reactance (pu)	0.0453
PCC bus transformer parameters
Voltage ratio (kV)	25/220
Resistance (pu)	0.0051
Reactance (pu)	0.065

3.2 The proposed protection technique based on ANFIS

ANFIS is the fuzzy system represented in a framework of the adaptive networks. Moreover, ANFIS combines the concepts of fuzzy inference systems (FIS) rule base and the learning benefit of artificial neural networks (ANN) to form hybrid adaptive systems with learning capabilities. ANFIS uses FIS rule base to describe the relationship between the input/output parameters and the ANN to train the data then find best parameters for the FIS membership function to get the suitable rules. The principles of ANFIS architecture consists of five layers, namely fuzzification layer, product rule layer, normalization layer, defuzzification layer and total summation layer. With input/output data for a given set of parameters, the membership function parameters of FIS model are adjusted using adaptive algorithm. The adaptive algorithm uses a back-propagation algorithm alone or a hybrid learning algorithm for training the parameters of MFs to emulate a given training data set where the hybrid algorithm applies a combination of the least-squares method and the back-propagation gradient descent method.

The inputs of ANFIS approach are phase voltage and phase current, which are fuzzified with membership functions and trained according to a training data measured at normal and abnormal conditions to get the best membership functions parameters. Figure 4 presents the flowchart for training and testing of ANFIS approach.

The general strategy of the proposed technique is to detect fault occurrence, classify fault type and determine the fault location of the studied wind farm based on ANFIS approach. The input signals of the proposed technique are measured magnitude values of voltages (V _a , V _b , V _c ) and currents (I _a , I _b , I _c ) at each row terminal. As shown in Fig. 5, the proposed protection technique consists of eighteen ANFIS networks dividing into six groups, where each group is responsible to protect one row using the measured values of voltages and currents at the row terminal. Each feeder of wind farm is equipped with two circuit breakers (CBs) to isolate it during fault occurrence, where the CBs are controlled by ANFIS. As illustrated in Fig. 5, the collected output of each ANFIS group is a control signal to turn off the CBs for wind turbine generators of faulted zone to isolate and protect them. During grid faults, the CBs are triggered depending on fault location and fault type to limit the high over-current and isolate the faulted row wind turbines.

The designed ANFIS networks are trained and tested using various sets of field data. The parameters of each three ANFIS structure are identical and selected as follows, two inputs and three MFs for each input, the inputs type for MF is Gaussian, the output is constant, the error tolerance is chosen to be zero, the number of epochs is 300, grid partitions, and the optimization method is hybrid algorithm. The structure and parameters of ANFIS are adjusted to perform three targets, where it can be used to detect, classify and determine the location of different faults for the studied wind farm. Figure 6 shows a flowchart of different steps for proposed protection technique.

4.1 Impacts of fault types

In this subsection, the behaviours of wind farm generators during occurrence of different faults such as single-line to ground fault, double-line to ground fault, and three-line to ground fault are investigated in cases of isolating and un-isolating of the faulted wind turbine generators by controlling of CBs trip signals. The studied faults are occurred close to row 1 at fault location (F1) as shown in Fig. 3.

Figure 8 shows the variations of active power, reactive power and their percentage errors at PCC bus of wind farm during occurrence of single-line to ground fault. As shown in Fig. 8a, when the faulted wind turbine generators are tripped off, the active power is fluctuated between 88.08 MW and 91.08 MW after fault occurrence, while it is fluctuated between 115.5 MW and 107.8 MW after fault clearance. On the other hand, in case of CBs not tripping of faulted wind turbine generators, there is no variation in the measured active power at PCC bus. The percentage errors of active power fluctuation in case of CBs tripping and not tripping are shown in Fig. 8b. Furthermore, the reactive power is fluctuated between − 5.795 MVAR and − 3.315 MVAR after fault occurrence, while it is fluctuated between − 3.65 MVAR and − 8.019 MVAR after fault clearance in the case of CBs tripping as shown in Fig. 8c. Also, in the case of CBs not tripping, there is no variation in the measured reactive power at PCC bus. The percentage errors of reactive power fluctuation in case of CBs tripping and not tripping are shown in Fig. 8d.

Figure 9 shows the variations of active power, reactive power and their percentage errors at PCC bus for studied wind farm in the case of double-line to ground fault. As shown in Fig. 9a, the active power is fluctuated between 79.08 MW and 106.6 MW after fault occurrence, while it is fluctuated between 118.1 MW and 106.4 MW after fault clearance in the case of CBs tripping. Also, in case of CBs not tripping, it is fluctuated between 80 MW and 98.67 MW after fault occurrence, while it is fluctuated between 130.7 MW and 99.16 MW after fault clearance. The active power fluctuation percentage errors in case of CBs tripping and in case of CBs not tripping are shown in Fig. 9b. As shown in Fig. 9c, it is noticed that, the reactive power is fluctuated between − 52.1 MVAR and 0.5216 MVAR after fault occurrence, while it is fluctuated between − 13.13 MVAR and − 3.173 MVAR after fault clearance in the case of CBs tripping. Also, in case of CBs not tripping, it is decreased to − 77.35 MVAR after fault occurrence and it is increased to 4.762 MVAR after fault clearance. The reactive power fluctuation percentage errors in case of CBs tripping and not tripping are shown in Fig. 9d.

Figure 10 shows the variations of active power, reactive power and their percentage errors at PCC bus in the case of three-line to ground fault. As shown in Fig. 10a, the active power is fluctuated between 49.05 MW and 119 MW after fault occurrence, while it is fluctuated between 115.5 MW and 105.1 MW after fault clearance in case of CBs tripping. Also, in case of CBs not tripping, it is fluctuated between 43.91 MW and 55.15 MW after fault occurrence, and it is fluctuated between 115 MW and 76.06 MW after fault clearance. The active power fluctuation percentage errors in case of CBs tripping and in case of CBs not tripping are shown in Fig. 10b. As indicated in Fig. 10c, the reactive power is fluctuated between − 122.9 MVAR and 3.906 MVAR after fault occurrence, while it is fluctuated between − 13.48 MVAR and 2.278 MVAR after fault clearance in the case of CBs not tripping. Also, in the case of CBs not tripping, it is decreased to − 136.8 MVAR after fault occurrence and it is increased to 16.44 MVAR after fault clearance. The reactive power fluctuation percentage errors in case of CBs tripping and not tripping are shown in Fig. 10d.

Based on the fluctuations of active power, reactive power and their percentage errors of the studied system for different fault types, it can be suggested that, the proposed protection technique isolates the faulted wind turbine generators in cases of double-line to ground fault and three-line to ground fault by tripping the faulted zone CBs.

4.2 Impacts of fault locations

To study the impacts of fault locations, the behaviour of wind farm generators is studied in case of three-line to ground fault occurs at various locations inside wind farm. The impacts of fault locations are studied for three cases, when the fault occurs at the location of F1 as first case, when the fault is occurring at the locations of F2 and F3 as second case, and when the fault occurs at the locations of F4, F5, and F6 as third case. Figure 11 shows the active power and reactive power variations at PCC bus when the fault occurs at different fault locations.

As shown in Fig. 11a, when the fault occurs at F1, the active power is fluctuated between 49.05 MW and 119 MW and it stabilizes at a value of 90 MW before fault clearance. Also, it is fluctuated between 29.58 MW and 104.5 MW, then it stabilizes at the value of 71.6 MW during fault occurrence at F2 and F3, while it fluctuated between 12.65 MW and 85.8 MW, then it stabilizes at the value of 54.5 MW when the fault occurs at F4, F5 and F6. The reactive power variations are indicated in Fig. 11b. It is clear that, when the fault occurs at F1, the reactive power is decreased to − 122.9 MVAR during fault period and fluctuated between − 13.48 MVAR and 2.28 MVAR after fault clearance, then returns to steady state value. When the fault occurs at F2 and F3, the reactive power is decreased to − 128.6 MVAR and fluctuated between − 21.92 MVAR and 9.88 MVAR after fault clearance. Also, it is decreased to − 117.4 MVAR when the fault occurs at the locations of F4, F5, and F6, where it is fluctuated between − 30.72 MVAR and 17.59 MVAR after fault clearance.

4.3 Impacts of fault duration

The impact of fault duration time on the behaviour of the studied wind farm generators is investigated in case of a three-line to ground fault occurs at row 2 for different fault duration times. The fault duration time is varied between 150 ms and 300 ms, where the faults are occurring at the same instant of 150 s and clearing at the different instants of 150.15 s and 150.3 s of simulation time respectively. The variations of active and reactive power for different fault durations are illustrated in Fig. 12. It is obvious that, the wind farm generators return to deliver active power to the grid after fault clearance for different fault duration times.

4.4 Impacts of cascaded faults occurrence

In this subsection, the behaviour of wind farm generators during a composed three-line to ground faults is investigated. The composed fault is a sequence of two faults, the first fault occurs at instant of 150 s and clears at 150.150 s nearest row 1, while the second fault occurs at the instant of 150.075 s and clears at 150.225 s nearest row 2. In this studied case, the active and reactive power variations are monitored at PCC bus of wind farm. During fault period, the active power at PCC bus is fluctuated between 119 MW and 36.86 MW, then returns to steady state value nearly at the instant time of 150.4 s after fault clearance as shown in Fig. 13a. Also, the reactive power is fluctuated between − 122.9 MVAR and 4.05 MVAR, then returns to steady state value nearly at the instant time of 150.4 s after fault clearance as shown in Fig. 13b.

4.5 Impacts of permanent fault occurrence

The behaviour of wind farm generators equipped with the proposed protection technique during permanent three-line to ground fault which it occurs at row 1 is investigated. As shown in Fig. 14a, the value of the exported power to the grid measured at PCC bus is 108.3 MW during steady state condition. It is fluctuated between 49.05 MW and 119 MW during fault period, where it stabilizes to 90.2 MW nearly at the instant of 150.2 s. The reduction of total exported active power to the grid is decreased due to the isolation of wind turbine generators for faulted row 1 by the proposed protection system technique. Also, the measured reactive power at PCC bus is − 5.1 MVAR in steady state condition. It is fluctuated between − 122.9 MVAR and 3.906 MVAR during fault period, where it stabilizes to − 4.25 MVAR nearly at the instant of 150.2 s as shown in Fig. 14b.

4.6 Impacts of external fault occurrence

To study the impacts of external grid fault occurrence, the behaviour of wind farm generators is studied when a three-line to ground fault occurs on the transmission line at distances of 10 km and 30 km from the PCC. Figure 15 shows the impact of different external fault locations on the instants of fault detection and clearance. It is clear that, the time taken by the proposed protection technique to detect the fault occurrence or fault clearance is directly proportional to the distance between the fault location and the wind farm.

On the other hand, the active power and reactive power have more fluctuation when the fault occurs at a distance of 30 km comparing with the case of fault occurrence at a distance of 10 km as shown in Fig. 16.

4.7 Impacts of transition resistance

The impacts of transition resistance Rg on the accuracy of the proposed technique in cases of internal and external fault are studied. It is studied in case of a three-phase to ground fault occurs at row 1 as internal fault, and when the fault occurs on the transmission line at a distance of 10 km from the PCC bus as external fault. The faults are studied at a different transition resistance values such as 0 Ω, 10 Ω, and 100 Ω. The variations of the measured active and reactive power at PCC bus in case of internal and external fault are indicated in Figs. 17 and 18 respectively. It is clear that, the proposed protection technique tends to be more accurate and robust since it is unaffected by the variations of transition resistance values in cases of internal and external faults.

A comparison between the proposed protection technique with others protection techniques [30–32] is provided as follows: Chen et al. [30] have presented superconducting fault current limiter (SFCL) protection system to suppress the fault current of DFIG-Based wind farm. Wang et al. [31] have proposed a flexible fault ride through strategy to allow a few wind turbine generators to trip and maintains the connection to grid of most generators during fault. Noureldeen et al. [32] have proposed an efficient ANFIS crowbar protection for DFIG wind turbines to protect wind turbine generator components during grid fault. The objective of the proposed technique focuses on avoiding trip events of all wind farm during faults and proposes a robust intelligent protection technique for wind farm based on ANFIS approach to trip only the faulted wind turbine. The proposed technique has some advantages comparing with others protection techniques such as classification of fault types, determination of fault locations and isolation of faulted zones. Moreover, the proposed technique is investigated at different fault conditions such as fault duration, cascaded faults, permanent fault, and variations of transition resistance values in cases of internal and external faults.

Finally, the simulation results show that, the proposed protection technique has the ability to detect and classify the fault, hence isolate the faulted zones during fault occurrence, where that lead to support the un-faulted generators to stay in services.