Performance Enhancement of Wind Farms Using Tuned SSSC Based on Artificial Neural Network

control gains of SSSC which can enhance the performance of CWF. On other side, AI represents high technology so that it is storage costly. In last years, Artificial Intelligence (AI) has been used extensively in improving the performance of FACTS and enhancing the performance of wind farms interconnected grid. A genetic algorithm has been implemented to tune different type of FACTS interconnected wind farms and photovoltaic solar plant in [12]. In ref [13] [14] multi-objective genetic algorithm is used to improve the performance of DFIG. Also, multi-objective genetic algorithm is used to find the optimal gains of SSSC in [15]. Adaptive-network-based fuzzy inference system (ANFIS), ANN and genetic algorithm are proposed in [16] to improve the reactive power control of STATCOM. The whale optimization algorithm, genetic algorithm and ANN were used in [17] to determine the optimal parameters of STATCOM integrated with CWF. In Ref [18] particle swarm optimization is used to tune and damp power system oscillation of DFIG wind farms integrated with SSSC. A new control strategy based on ANFIS is proposed in [19] to improve the performance of DFIG wind farm integrated with SSSC.


I. Introduction
R enewable energy is an important source for the power generation. Solar energy, and wind energy are the most famous forms of this technology. Wind energy plays an important role in producing electric power in all the world so that its injection on the grid represents a wide range of studies. This injection depends on the induction generator of the wind turbines. There are two types of induction generator, first type is squirrel cage induction generators (SCIG) which are suitable to fixed speed wind turbines and second type is doubly fed induction generators (DFIG) that are used with variable speed wind turbines. The stability of wind farms is affected by the exchange in the reactive power between the interconnected grid and the wind farms. The compensation devices of the reactive power consider a fundamental element in SCIG wind turbines (SCIG-WT). The flexible alternating current transmission systems (FACTS) were used to damp power oscillation and, enhance power stability. In Ref. [1] a dual STATCOM had been used to damp power oscillations. Tuning parameters of SSSC had been proposed in [2] to damp power oscillations. In Ref [3] a unified power flow controller has been used to damp power oscillations between two areas. The SSSC used to damp power oscillation, enhance power stability and control the power flow of DFIG-WF is studied in [4]- [5].The effect of FACTS such as Static VAR Compensator (SVC), Static Synchronous Compensator (STATCM) and SSSC on the performance of wind farms were studied in [6]- [10]. The impact of SSSC on the performance of different types of wind farms had been discussed in [11].
The main advantage of Artificial intelligence (AI) is solving complex problems in less time and with high precision, such as using optimization methods to solve the complex control problem. Also, AI can easily predict and take the correct decisions with little margin of error. It can be used for predicting the change in wind speed and its impact on stability of power system. In this paper, AI has been used to predict and determine the optimal value of the control gains of SSSC which can enhance the performance of CWF. On other side, AI represents high technology so that it is storage costly. In last years, Artificial Intelligence (AI) has been used extensively in improving the performance of FACTS and enhancing the performance of wind farms interconnected grid. A genetic algorithm has been implemented to tune different type of FACTS interconnected wind farms and photovoltaic solar plant in [12]. In ref [13] [14] multi-objective genetic algorithm is used to improve the performance of DFIG. Also, multi-objective genetic algorithm is used to find the optimal gains of SSSC in [15]. Adaptive-network-based fuzzy inference system (ANFIS), ANN and genetic algorithm are proposed in [16] to improve the reactive power control of STATCOM. The whale optimization algorithm, genetic algorithm and ANN were used in [17] to determine the optimal parameters of STATCOM integrated with CWF. In Ref [18] particle swarm optimization is used to tune and damp power system oscillation of DFIG wind farms integrated with SSSC. A new control strategy based on ANFIS is proposed in [19] to improve the performance of DFIG wind farm integrated with SSSC. This paper aims to improve the performance of CWF which is based on SCIG and DFIG using SSSC controlled by ANN (ANN-SSSC). Also, in this paper the control parameters which had been investigated in [15] are used for implementing ANN. Moreover, a comparison is done between the performances of CWF with ordinary SSSC, CWF associated with SSSC tuned by multi-objective genetic algorithm (SSSC MOGA) investigated in [15] and CWF associated with proposed ANN-SSSC during three phase-faults.
The rest of the paper is organized as follows. Section II presents a brief summary of ANN. Section III presents modelling of wind turbines. Section IV explains the construction, operation and control system of SSSC. Section V introduces the proposed ANN control, which is applied to SSSC. The last two sections present the results and conclusion.

II. Artificial Neural Network (ANN)
The artificial neural network is a modest simulation of the effect, form and content of the neural network found in the human brain. It consists of nodes called neurons and connected together by bonds called weights. Each set of neurons forms a single layer; the ANN is composed of different types of layers. From Fig. 1, it can be observed that it consists of input layer, hidden layer (processing element) and output layer. The hidden layer could be single layer or multi-layers. The input signal is passed from input layer to the output layer through the hidden layer. The input is transferred to the neurons through weight matrix W. The output can be given by [20]: Where Y out represents the output of ANN, x is input signal which starts from 1 to n inputs and w ji represents the synaptic weights between neurons.

III. Modeling of Wind Turbines
The mathematical model of wind turbines was discussed in several articles on wind energy. Fig. 2 shows the equivalent circuit of induction generator. The direct and quadratic (d-q) illustration of IG with respect to the synchronous frame can be illustrated as flows [21] [22]: Equations (2) to (5) and Fig. 2 can be applied to a general modeling of IG. The voltage of rotor of SCIG is equal to zero because the rotor is a short circuit and there is no connection between SCIG's rotor and the grid as shown in Fig. 3 (a). While the rotor in DFIG is a wound rotor so the rotor current and voltage is taken into account. The AC/DC/AC converters are used to connect rotor of DFIG's to the grid through as illustrated in Fig. 3 (b). The equivalent circuits of DFIG and SCIG are illustrated in Fig. 3 (c) and (d). The extracted power from the wind by wind farms is given by: Where, v is the wind speed, ρ is the air density, C p is the power coefficient, A is the area swept by the turbine blades. l b is the blade length or rotor radius, ω r is the rotor speed and it is equal to1.22Kg/m3. C p is a function on the pitch angle β and the tip speed ratio λ.
The electrical torque of SCIG and DFIG is given by: The pitch angle control method is used to control the rotor speed of wind turbine in order to keep the output power inside permissible limits. Fig. 4 illustrates a schematic diagram of pitch angle control system [23].

IV. Static Synchronous Series Compensator (SSSC)
The Static Synchronous Series Compensator (SSSC) belongs to the series devices of FACTS controller [23]. It is a series connected with the transmission line so that it injects a series voltage which is in phase quadrature with the line current. The block diagram of SSSC and its equivalent circuit are shown in Fig. 5. As shown in Fig. 5. The SSSC is connected in series with the transmission line of an electrical grid by a coupling transformer. As shown in Fig. 5 (b) the SSSC injects voltage in series with the line. This voltage injected may be capacitive or inductive, if the injection voltage overrides the voltage drop (VL), the transferred power will be reflected in the direction.

A. SSSC Control System
The main components of SSSC controller are AC voltage regulator and DC voltage regulator. Fig. 6 (a) shows the MATLAB Simulink model of ordinary SSSC control system. Fig. 6

V. Applying the ANN Method
In this work the ANN is based on multi-layer feed-forward network. The multi-layer is divided into three layers input, hidden and output layer. Fig. 7 represents the flowchart of adjusting SSSC's parameters using ANN. As shown in Fig. 7, the neural fitting tool (NFTOOL) and the sample range are based on the value of control parameters of SSSC tuned by multi-objective genetic algorithm (SSSC MOGA) investigated in [15]. Also, Fig. 7 illustrates that the input signal is the change in voltage at the point of connection and the output layer represents the control parameters of SSSC (AC voltage regulator (Kp-vac and Ki-vac) and DC voltage regulator (Kp-vdc and Ki-vdc)). In this work, Levenberg-Marquardt algorithm is used for training the value of control parameters of SSSC tuned by multi-objective genetic algorithm (SSSC MOGA) investigated in [15].
The neural fitting tool (NFTOOL) is composed of sets of processes: training, validation and testing. The application divides input and target into three groups as follows: 70% is applied for training. 15% is applied to validate and 15% is applied to test. Table I shows the parameters of NFTOOL and illustrates the mean square error (MSE) and regression value (R) of training, validation and testing.

Type of algorithm
Levenberg-Marquardt and multi-objective genetic algorithm investigated in [15] Number of hidden neurons 20 neurons In this study NFTOOL is a feed forward network with input, output and multi-layers. Fig. 8 (a) illustrates the neural network size. Fig. 8 (b) illustrates training, validation and testing samples. Fig. 8 (c) illustrates results of training operation. Fig. 9 illustrates best validation performance. Fig. 10 illustrates the ANN controller of DC and AC voltage regulators.

VI. Studied System Description
The studied system contains six wind turbines, each one produces 1.5 MW and 575v. The wind turbines are divided into three SCIG fixedspeed wind turbines and three DFIG variable speed wind turbines. Fig.  11 shows the block diagram of the studied system. A three phase fault is applied at 25 s and removes [e1] the fault after time equal 25.15 s. Fig.11. The block diagram of the system.

VII. Simulation Results
The simulation studied the performance of CWF with ANN-SSSC and CWF with ordinary SSSC (PI-SSSC) during three-phase fault. The voltage, reactive power and active power are measured at the point connection between the interconnected grid and wind farms.
The value of the control parameters of ordinary SSSC, MOGA SSSC and the proposed ANN SSC are shown in Table II.

A. Impact of Three Fault
As illustrated in Fig. 12 the voltage with ANN-SSSC is 0.81 pu while the voltage with ordinary SSSC is 0.73 pu and the voltage with MOGA-SSSC is 0.75 pu. This means that the voltage of CWF has been improved with ANN-SSSC more than with the ordinary SSSC. This enhancement in voltage of CWF with ANN SSSC is due to the enhancement in the performance of SSSC when it is controlled by ANN. This can be observed by monitoring the injected voltage of SSSC in case of ordinary, MOGA and ANN as shown in Fig. 13. As illustrated in Fig. 14, the injected voltage of SSSC has been increased when it is controlled by ANN specially at the begging of fault period. This will decrease the reactive power absorbed by the CWF with ANN SSSC during fault.
As illustrated in Fig. 14 the reactive power of CWF with ANN-SSSC is-2.8 MVAR while the reactive power with ordinary SSSC is -4.75 MVAR and -3.83 MVAR for MOGA SSSC especially at the beginning of fault period. This means that the performance of CWF has been improved with ANN-SSSC more than the ordinary SSSC. As illustrated in Fig. 15 the active power of CWF with ANN-SSSC has the highest value of output power especially at the beginning of fault period. In order to justify the good performance of the proposed method, root mean square Error (RMSE ) is used to measure the impact of ordinary SSSC, MOGA SSSC and ANN SSSC on the performance of CWF during the fault. RMSE is used to measure the error between the reference voltage (Vref = 1 pu) and the actual voltage at the point of common connection during the fault condition. Table III shows the RMSE of the three cases during fault period and after fault clearance.

VIII. Conclusions
This paper has presented the design of a CWF which consists of two types of induction generator, the first one is SCIG fixed speed wind turbines and the second one is DFIG variable speed wind turbines. ANN has been used in order to adjust SSSC's parameters to enhance the performance of combined wind farm (CWF). In addition, this paper includes a comparison among the performances of combined wind farm (CWF) with ANN-SSSC, with performances of CWF with ordinary SSSC and performances of CWF with SSSC tune by Multiobjective genetic algorithm (MOGA SSSC). The performances of CWF with ANN-SSSC, ordinary and SSSC MOGA SSSC have been studied during the three-phase fault. The obtained results showed that the adjusted SSSC using ANN had enhanced the active power, the voltage and the reactive power of CWF, particularly during the threephase fault.

IX. Future Work
This paper opens the door for many future works, for example: 1. Using ANN hybrid with different methods of optimization to determine optimal values for different types of FACTS to improve the performance of wind stations.
2. Use different types of ANN like ANFIS with different methods of optimization to determine optimal values for different types of FACTS to improve wind station performance.