Intelligent maximum power point tracker enhanced by sliding mode control

ABSTRACT


INTRODUCTION
Solar energy advantages which are being clean, inexpensive, and ubiquitous make solar energy a very important source of renewable energy. Photovoltaic (PV) cell is the fundamental part of the PV module, which converts the solar photovoltaic energy to an electricity as a direct current direct current (DC) voltage source. The power quantity is positively proportional to the instantaneous level of the incident light whereas it is negatively proportional with instantaneous level of PV panel temperature. The instantaneous harvested energy is converted to electrical power fluctuating based on the levels of light intensity and ambient temperature [1]- [3]. In the solar PV applications, to gain higher electrical power by solar energy conversion, many maximum power point tracking (MPPT) algorithms are explained in [4], whereas different algorithms and methods have been proposed and analyzed in [5]- [19]. Incremental conductance (IC), short circuit current, constant voltage (CV), and perturb & observe (P&O) algorithm have offered in [5]- [9] to maximize the power point tracking conditions during the work of the PV panel.
The demerits of the above MPPT algorithms are represented by low accuracy, low robustness, slow and oscillated response. The researchers are focusing on mitigating the effects of these demerits by proposing

PHOTOVOLTAIC MODULE
This study selects the PV module type ELDORA 150P [36] of main specifications as shown in Table 1. The level of delivered power from the ELDORA 150P panel is positively proportional to the level of incident light, whereas the delivered power is negatively proportional to ambient temperature of the panel. The output power behaviour of the PV cell is nonlinear with the output voltage variation. The electrical representation of the smaller PV unit is shown in Figure 1.  In (1) shows Ipv which is the output PV current, that equals the generated current Isc minus diode current and minus shunt resistor Rsh current Ish. From (2), the induced current Isc can be calculated. The current value is related to the solar cell area A, the generation rate Gr, and the electron and hole diffusion lengths Ln, and Lp respectively. From (3), diode current Id can be calculated, and from (4), the parallel current Ish of the parallel resistor Rsh can be calculated. Vpv represents the output voltage of PV cell which is calculated from (5) by considering the drop voltage across the series resistor Rs. The total voltage out from the panel Vpv_Module and the current out from the panel Ipv_Module can be calculated from (6), and (7) Figure 2 including current curves in Figure 2(a) and power curves in Figure 2(b) at different levels of incedent light; 200, 400, 600, 800, and 1000 W/m2 at the normal temperature of the room 25 ºC. Figure 3 shows the PV panel current curves in Figure 3(a) and power curves in Figure 3(b) at many ambient temperatures 15, 25, 35, 45, and 55 ºC during a constant level of light intensity 1000 W/m2.

ALGORITHM OF ARTIFICIAL NEURAL NETWORK
The characteristics of maximum power point tracker can be more accurate and fast in response using feed-forward ANN algorithm [17]- [26]. The neural network algorithm estimates the accurate value of the instantaneous MPP reference voltage by evaluating the instantaneous levels of the incedent light and panel ambient temperature. A generated reference voltage enforces the involved boost converter to work in a MPP condition to harvest maximum power from the incident solar energy. Figure 4(a) explains the presented algorithm design of neural network. The ANN algorithm includes one, and two input layer, hidden layers respectively, and four neurons in each hidden layer. The output layer represents the last layer of the proposed ANN. The input layer receives the instantaneous values of light intensity and ambient temperature, whereas the output layer produces the instantaneous value of reference voltage after the processing of the hidden layer. Figure 4(b) shows the neuron structure, in which there are weights of each input to the neuron: Xn1, Xn2, Xn3, and Xn4. After weighting all neuron inputs, all together add to Bias (B) to produce the internal result of Z. One of activation functions (linear, sigmoidal, or hyperbolic transfer function) to produce the output value of yn as showin in (8)-(11) respectively will manipulate the instantaneous produced value Z: Linear bipolar: The maen square error (MSE) is considered for evaluating the effectiveness of the presented NN algorithm. This parameter is demonstrated in (12), which indicates the difference between the target and predicted values. A smaller value of MSE indicates the accurate performance of the designed ANN algorithm.
Where Q is the input vectors number, and err(k) is the error between the target and the estimated values.

SLIDING MODE CONTROLLER DESIGN AND SYSTEM BLOCK DIAGRAM
The presented SMC design is suitable for DC-DC boost converters to remove the overshoot in the output load voltage by evaluating reference base voltage, the design of SMC starts from the dynamic (13) and (14) of the inductor current and output voltage variations with respect to time, as explained in Figure 5, which shows the power electronic circuit arrangement of DC-DC boost converter in Figure 5(a), the equivalent circuit when the insulated gate bipolar transistor (IGBT) is closed in Figure 5 Where the source voltage is represented by Vs, the converter inductance is represented by L, the converter capacitance is represented by C, and the load resistor is represented by R. The controller effectiveness is determined by evaluating the accuracy in determining the instantaneous switching state of u of the converter switch, and determining u formula started by determining the sliding surface S, then equaling S, and Ṡ (the derivative of S) to zero. The sliding surface S is the summation of output voltage error (e = x1 = reference voltage Vref -output voltage Vo) and the derivative of the error (x2 = de/dt): And The formula of ueq can be determined by equaling Ṡ to zero: Whereas the nonlinear component un is: The presented intelligent MPP tracker supported by SMC is demonstrated as a block diagram in Figure 6. The selected PV panel ELDORA 150P is connected to the boost converter. The converter is controlled by the instantaneous value of u through the pulse width modulation (PWM) pluses generator. The ANN algorithm evaluates the variables of light intensity and ambient temperature for accurately producing a reference voltage. Enforcing the converter to be driven by reference voltage guarantees the MPP position to maximize harvesting the electricity through the PV panel. The designed sliding mode controller produces a suitable switching state u by monitoring the instantaneous reference voltage, output voltage and inductor current of the converter.

SIMULATION RESULTS
The boost converter parameters are designed by considering the study in [4], the selected switching frequency is 40 kHz, as shown in Table 2, in which the converter parameters are also shown. The study ANN involves input layer of two neurons, two hidden layers of four neurons each, and output layer of one neuron. Figure 7 shows the algorithm structure in Figure 7(a) and performance in terms of MSE in Figure 7(b) which reflects an accurate ANN performance of MSE (6.0421e-5) at epoch 7.   Figure 8 represents the full simulation program, which includes two simulations. The left part is for simulation of MPP tracker using ANN only, whereas the right position is for MPP tracker using ANN supported by the presented SM controller. The simulation results are collected during 1 second of four equally divisions 0.25 second. The simulation is done in a comparative way using ANN algorithm with and without the SMC inserting to evaluate the effectiveness of the SM controller in smoothing the mitigating the overshoot in the load power. Simulation results are collected during ambient room temperature 25 ºC and different light intensity levels; 600, 800, 1000, and 700 W/m 2 respectively. Figure 9(a) shows the MPP tracker performance based on ANN only. The figure demonstrates the voltage, current, and power of the connected load. Overshoots are clearly noticed in the voltage, current, and power of the load. On the other hand, the positive effect of SM controller in smoothing the shape of load power is noticed in Figure 9(b), which shows the MPP tracker performance when it is supported by sliding mode controller, and all the overshoots are avoided. Figure 10 shows the tracker response in a comparative way before and after inserting the SMC and how the overshoots (red colour) can be smoothed (blue colour) by the inserted SM controller.

CONCLUSION
In this study, an intelligent MPP tracker supported by a robust sliding mode controller has been presented. The designed SM controller is suitable for boost converter. An intelligent tracker is evaluated comparatively with and without inserting SM controller. ANN algorithm has been adopted in this study to guarantee the MPP tracking working conditions through fast predicting an accurate reference voltage, and then this voltage has processed by the SM controller to have a smooth response from the tracker and to avoid the overshot in the load power. Simulink of MATLAB is adopted to simulate the presented intelligent MPP tracker, firstly, and the simulation results are collected without inserting the SM controller into the tracker system.

1045
After that, the designed SM controller has inserted to support and enhance the tracker performance. The results indicate the effectiveness of the presented MPP tracker after inserting the SM controller and promise a highperformance prototype as a future step.