Recursive Bit Assignment With Neural Reference Adaptive Step (RNA) MPPT Algorithm for Photoviolatic System

— A. Nowadays, most of the recent research is directed towards photovoltaic harvesting systems due to their great characteristics. To increase the efﬁciency of a photoviolatic harvesting system, Maximum Power Point Tracking algorithms are utilized to achieve the maximum output power of the PV. This is done by optimizing the duty ratio of the DC-DC boost converter. an efﬁcient MPPT for proposed algorithm First an Second one is a proposed Recursive Bit Assignment present step size of the duty ratio of the DC-DC boost converter. The RBA network consists of N-bit memory. The instantaneous PV power value sets the contents of the memory to generate the variable step size of the duty ratio. design of the give


A. Background
Nowadays, most of the recent research is directed towards photovoltaic harvesting systems due to their great characteristics. To increase the efficiency of a photoviolatic harvesting system, Maximum Power Point Tracking algorithms are utilized to achieve the maximum output power of the PV. This is done by optimizing the duty ratio of the DC-DC boost converter.

B. Results
In this paper, a proposed RNA algorithm is introduced as an efficient MPPT algorithm for the PV system. This proposed RNA algorithm depends on two main parts. The First part is an artificial neural network to produce a reference power. The Second one is a proposed Recursive Bit Assignment network to present the variable step size of the duty ratio of the DC-DC boost converter. The RBA network consists of N-bit memory. The instantaneous PV power value sets the contents of the memory to generate the variable step size of the duty ratio. Moreover, the design of the neural network to give its best performance is explained.

C. conclusions
The performance of the chosen PV module is simulated for a variable solar radiation and a constant temperature. Simulation studies are performed using MATLAB to evaluate the system performance. From simulation results, the proposed RNA can achieve a fast tracking time, a high power efficiency, an actual maximum power point and an acceptable ripple. Additionally, comparisons between the RNA algorithm and other related algorithms such as Perturbe and Observe, Neural Network, and Adaptive Neural Interference System algorithms are executed. The proposed RNA achieves the best performance in all terms.

I. INTRODUCTION
Nowadays, solar PV energy is receiving a great deal of attention in the field of renewable energy sources. Solar PV in harvesting systems has been widely used in many high and low power applications such as wireless sensor networks. The solar PV cells convert the solar energy directly into electrical energy. Several solar PVs are combined to produce the PV module. The main challenge for PV harvesting system is its low efficiency as it is in the range of 10-40% [1].
Furthermore, the maximum output power of the PV changes under changing environmental conditions. In order to maximize the PV efficiency, Maximum Power Point Tracking (MPPT) algorithms are introduced for the PV systems. Its output power depends on the operating ratio of the switch in the DC-DC inductive boost converter which is connected with the PV module. Several MPPT algorithms have been developed to obtain the maximum power point (MPP) from the PV module [2]. The solar PV system is composed of PV module, DC-DC boost converter, MPPT control circuit and load. The DC-DC boost converters are used to raise the PV voltage to the required load voltage. It is worth mentioning that the working ratio of the DC-DC boost converter must be adjusted to obtain the MPP of the PV module. This can be done via the MPPT circuit.
In this paper, a proposed two-stage RNA algorithm combining NN and RBA is presented. In recent years, the neural network has become a more attractive issue due to its ability to solve non-linear and complex problems. Therefore, it is utilized in the proposed RNA algorithm. In the first stage, the NN is introduced for calculating the reference power value. Its front feed is used with two inputs which are irradiance (G) and temperature (T). In the second stage, the RBA circuit is implemented to obtain the value of the variable step size of the DC-DC boost converter.
The proposed RNA algorithm adjusts the working duty ratio of the DC-DC boost converter to achieve the MPP of the solar PV module. The proposed system is studied using MATLAB. Many comparisons are made with other MPPT algorithms such as P&O, NN, and ANFIS algorithms. The proposed algorithm can attain a fast rise time, a high power efficiency, an actual maximum power point which approximate the exact value and an allowable ripple.
The rest of the paper is organized as follows: Section II presents the related work. The description of the whole PV system architecture is shown in Section III. In Section IV, the proposed RNA algorithm is explained in details. The simulation analysis is provided in Section V. Finally, the conclusion remarks are shown in SectionVII .

II. RELATED WORK
Many algorithms are evolved to extract the Maximum Power Point from the solar PV module under varying conditions of solar irradiance and temperature [3]. The traditional techniques of MPPT algorithms which based on hill-climbing method such as Perturbation and Observation (P&O), Incremental Conductance (INC), Fractional Open Circuit Voltage(OCV), and Fractional Short Circuit Current techniques have been discussed in [4]- [10]. These algorithms present a slow response under varying irradiation level. The heuristics based algorithms like particle swarm optimisation (PSO), Firefly Algorithm(FA), and genetic algorithm (GA) have been released to improve the MPPT performance under both partial shading and uniform irradiance climatic conditions as in [11]- [13]. The recent techniques such as artificial neural network [14]- [16], fuzzy logic methods [17]- [18], and artificial neuro-fuzzy inference system [19]- [20]. The artificial neural techniques become the most interested approaches that commonly used in PV systems due to its ability to resolve a significant problems of the traditional methods such as oscillation around the maximum point, failure behavior with rapid changing of the solar irradiance. Several MPPT controllers which based on the artificial neural network have been developed due to the main advantages of the neural network such as it can find complex nonlinear relations between the independent and dependent variables without need an accurate mathematical models. In [21], an artificial neural network has been introduced for fast tracking of maximum power point of the solar PV. In this algorithm, the output of the neural network is the reference voltage of the MPP of the solar PV under different climatic conditions. The authors present an efficient artificial neural network based MPPT scheme for improving the efficiency of the photovoltaic generator. It can be operated accurately and rapidly at MPP without power loss. This can be done through matching impedance between solar PV module and load by using a DC-DC boost converter in which its duty cycle is set by artificial NN. In [22], the authors have been proposed a two artificial neural network based MPPT controller: fixed step and variable step NN. The neural network based MPPT controller is executed in two steps: online and offline. Different set of neural network parameters are required to be trained through the offline step to find the optimal NN. The optimal neural network based MPPT controller can be used in the PV system through the online step. In [23], the authors have been investigated a low complexity MPPT algorithm which based on the neural network model of the solar PV. The expression for the output current of the NN model can explore a gradient, analytical MPPT method which can donate an accurate prediction of the maximum power. The proposed algorithm has a lower computational complexity than the other NNs based MPPT techniques as the position of the MPP is determined by one multiplayer NN or by using two single layer NNs. Ref [24] provided an adaptive neuro-fuzzy system based MPPT algorithm with PI controller. The proposed algorithm donates a maximum power point tracker with a control gain for solar pumping system. Ref [25] improved the efficiency of the solar PV harvesting system by minimizing the energy losses caused by the MPPT controller and dc-dc converter. This can be done by using a successive approximation register based MPPT algorithm. This algorithm has a two main benefits rather than the other MPPT algorithms in terms of energy savings and power consumption. It is considered as an advanced version of hill climbing technique with a fast tracking time. Ref [25] suggested a maximum power point tracking circuit that used analog to digital converter with 4-bit successive approximation register for solar PV harvesting system. This circuit achieves high MPPT efficiency with a low power consumption.

III. SYSTEM DESCRIPTION
The architecture of the solar PV system consists of a PV module, a DC-DC boost converter, a MPPT control circuit, and a load. The main challenge for this system is the maximum power tracking process under different environmental conditions such as G and T. The system architecture is plotted in Fig. 1.
The equivalent circuit of the solar PV module can be modeled as a single diode circuit as shown in Fig. 2. The mathematical expression for the nonlinear I-V characteristics of the ideal solar PV cell is described as follows [21]: where, I P V is the photocurrent (generated current of the incident light). I o is the reverse saturation current. q is charge of T is the temperature of solar cell (300 kelvin) and a is diode ideality constant. The output current of the PV module can be expressed as [23]: where, V is the output voltage. R p is the shunt resistance. R s is the series resistance.
IV. THE PROPOSED RECURSIVE BIT ASSIGNMENT WITH NEURAL REFERENCE ADAPTIVE STEP (RNA) ALGORITHM The proposed RNA algorithm consists of two main steps. First, the feed forward neural network is used. It has two inputs which are the radiation G and the temperature T. Hence, the maximum power point P mpp is the output of the neural network. It can be used as a reference power for the second step. The second step is the recursive bit allocation which is used to obtain the adaptive duty cycle.

A. The Proposed RNA Architecture
The flowchart of the proposed algorithm is shown in Fig. 3. It can be explained as follows: Firstly, the initialization parameters which are the initial value of power (P old =0), the initial value of the duty cycle (D old ) and the fixed step size (d= 0.00001) are set. D old can be donated as follows: where, D max is the maximum duty ratio, D min is the minimum duty ratio, and rand is a random value from 0 to 1.
Then, the irradiance (G) and the temperature (T) are measured. These values are used as inputs for both the PV and the NN modules. The outputs of the PV module are the voltage (V pv ) and current (I pv ) of the PV. Thereafter, the PV power can be calculated by multiplying (V pv ) and (I pv ). Hence, the power difference (dP ) can be estimated by subtracting the P old from the instantaneous PV power value.
After that, RBA algorithm is performed to estimate the variable step size of the duty ratio. It consists of a memory with a length of N bits. After the N bits are assigned, the aggregated value (B) is calculated. The architecture of the whole RBA block is drawn in Fig. 4.
Afterwards, the outputs of both the NN module (P ref ) and the RNA module (B) are used to calculate the variable step size (△D) as indicated in the following equation: Finally, the value of the duty ratio (D) is calculated as where D = (D old + △D) when dP < 0. Otherwise, D = (D old − △D). This condition can be implemented using a  comparator circuit. The hardware architecture of the proposed harvesting system using RBA algorithm is shown in Fig. 5.

B. Recursive Bit Assignment (RBA)
As shown in Fig. 4, the RBA block consists of two comparators, X-NOR, memory with size of N * 1 bits, N multipliers and one adder.
First, dP is compared with zero. Then, the output of the comparator is X-nored with a rand comparator. The X-NOR circuit operation is listed in Table I. Hence, the output of the X-NOR circuit is stored in a sequentially manner in the memory locations.
After all memory contents are assigned (B 0 , ......., B ( N − 1)), the output of RBA block can be determined by summing the weighted values of the assigned bits which can be written as:

C. Neural Network based MPPT
In this research, feed forward neural network with one hidden layer of 10 neurons is used for the MPPT control design. It has two inputs (which are G and T) and one output variable (P mpp ) as shown in Fig. 5. The training process for the neural network is performed using Levenberg-Marquardt back propagation optimization method.

A. Simulation Setup
The whole proposed harvesting system architecture using RNA MPPT algorithm is designed using MATLAB. The proposed system uses a PV module of Soltech 1STH-250-WH type with a maximum power of 250 Watt. The numerical parameters values of the PV type is listed in TABLE II. The nonlinear P-V and I-V characteristics of the PV module for different values of irradince from 200 to 1000 W/m 2 are shown in Fig. 6.

B. Neural Network Performance
The relation between the Mean Square Error (MSE) with number of epochs of the NN performance is shown in Fig. 7. The training, validation, and testing measurements are plotted. From results, the best validation performance is MSE of  9.0106 × 10 −5 at 1000 epochs. Additionally, all the training, the validation, and the testing curves are closed to each others. That is means that the designed NN is reliable and it can efficiently predict its output value.
C. The Proposed RNA Algorithm Performance  results, the proposed RNA algorithm achieves the MPP with fast tracking response and acceptable oscillations around the MPP. Fig. 9 shows a comparison of the output powers of the DC-DC boost converter for the proposed RNA, ANFIS, NN, and P&O algorithms. It is obviously that the proposed RNA system achieves the best performance compared with other algorithms. It achieves the fastest time response with the irradiance variation. Also, it has an acceptable ripple. This is because the utilization of the RBA block that continues fine tuning the duty ratio of the boost converter. Additionally, the performance can be improved by increasing the memory size of the RBA block. This can be done by increasing the number of the bits to properly achieve the maximum power point. At G = 1000 W/m 2 , 800 W/m 2 , and 600 W/m 2 , the proposed algorithm attains maximum output power of 245 W, 195 W, and 146 W, respectively with fast tracking response. While, the NN-based and ANFIS algorithms have a slow responses with less maximum power point values. Whereas, the conventional P&O provides the worst performance in terms of tracking response compared with all other algorithms. At G = 1000 W/m 2 , the proposed system attains the maximum power value within 0.05 sec. While, the ANFIS and NN-based algorithms arrive a maximum power of 237 W and 236 W within 0.17 sec and 0.12 sec, respectively.
The comparison of the PV powers of the DC-DC boost converter for the proposed RNA, ANFIS, NN, and P&O algorithms is shown in Fig. 10. It is cleared that the best performance is occurred with the proposed RNA system in terms of all comparison metrics. At G = 1000 W/m 2 , the proposed algorithm achieves a maximum PV power of 249 W at 0. The output voltages of the DC-DC boost converter for the compared algorithms are shown in Fig. 11. For G = 1000 W/m 2 , the proposed algorithm has the fastest response to arrive the maximum output voltage at 70 V within 0.05 sec. While, the ANFIS requires about 0.14 sec to get a maximum output voltage of 69 V.W Whereas, the NN-based algorithm needs 0.2 sec to obtain the same performance as the ANFIS algorithm. Regarding to the P& O, it has a slow response, as it needs more time of 0.3 sec to follow up the same performance of the two previous algorithms. algorithm. While, the P&O algorithm has the highest delay as it needs a time of 0.57 sec to achieve the maximum value. Also, at G = 600 W/m 2 , the proposed algorithm donates a maximum current of 2.7 A at 0.84 sec. The NN-based algorithm needs 0.03 sec to reach the same performance as the ANFIS algorithm. While, the P&O algorithm requires 1 sec to arrive the maximum output current.
The metric of "MPP achievement" can be measured using "algorithm efficiency" metric. The algorithm efficiency can be defined as: where, M P P algorithm is the obtained MPP by the algorithm. Also, the ripple around the MPP can be measured by "the ripple factor". It can be defined as the difference between the maximum and the minimum power values divided by the minimum power value at the steady state. The summary of the comparisons between the proposed RNA algorithm and the other three MPPT algorithms (ANFIS, NN-based, and P&O) with different irradiance levels (1000, 800, 600 W/m 2 ) are listed in TABLE III. The most important metrics are the rise  time (the fast tacking response), the algorithm efficiency, and the ripple factor.

A. Data Availability
All data sets in this paper are normally available to publishing. Also, the data that support the findings of this paper are available from the corresponding author upon reasonable request. Moreover, the data are not publicly available due privacy.

B. Competing interests
This paper includes a competing interests with other related works.

C. funding
There is no funding source.

D. Acknowlegements
The authors declare no conflict of interest and no financial support. The authors encourage the readers to use all data sets and results in this paper.

VII. CONCLUSIONS
The proposed RNA MPPT controller for solar PV system has been presented. It has been consisted of two main stages which are; the neural stage and the RBA algorithm stage. The neural stage has been used to obtain the reference power. While, the RBA stage has been utilized to determine the variable step of the duty ratio for the boost converter. Comparisons between the proposed algorithm and other techniques such as; ANFIS, NN-based, and P&O have been executed. From the simulation results, the proposed algorithm has achieved the superior performance in terms of both the fast tacking of the MPP and the efficiency. While, it has reaped a slightly higher ripple compared with other algorithms.  Figure 1 The Solar PV harvesting system architecture. The equivalent circuit model of the PV module. The owchart of the proposed RNA algorithm.

Figure 4
The Recursive Bit Assignment (RBA) Block.

Figure 5
The proposed harvesting system architecture using RNA MPPT algorithm.

Figure 6
The V-I and V-P characteristic of the PV module.

Figure 7
The performance of the Neural Network.

Figure 8
The Input irradiance pattern.

Figure 9
The output power of the proposed RNA algorithm.

Figure 10
The PV power of the proposed RNA algorithm.

Figure 11
The output voltage of the proposed RNA algorithm.

Figure 12
The output current of the proposed RNA algorithm.