Microgrid Integration Based on Deep Learning NARMA-L2 Controller for Maximum Power Point Tracking

T his paper presents a hybrid energy resources (HER) system consisting of solar PV, storage, and utility grid. It is a challenge in real time to extract maximum power point (MPP) from the PV solar under variations of the irradiance strength. This work addresses challenges in identifying global MPP, dynamic algorithm behavior, tracking speed, adaptability to changing conditions, and accuracy. Shallow Neural Networks using the deep learning NARMA-L2 controller have been proposed. It is modeled to predict the reference voltage under different irradiance. The dynamic PV solar and nonlinearity have been trained to track the maximum power drawn from the PV solar systems in real time. Moreover, the proposed controller is tested under static and dynamic load conditions. The simulation and models are done by using MATLAB/Simulink. The simulation results from the proposed NARMA-L2 controller have been compared with existing Perturb and observe PO-MPPT and Incremental Conductance INC -MPPT methods.


INTRODUCTION
Renewable energy sources such as solar PV have increased significantly, especially recently (Wen et al., 2019;Singh et al., 2022).However, increasing power penetration on a microgrid may cause problems such as voltage deviation, surges, and frequency fluctuations.Predicting the maximum power drawn from the PV system may enable better voltage and current responses (Abdul Hussain and Habbi, 2018; Kulkarni and Deshmukh, 2019).Therefore, extracting maximum power with good transient performance is a challenge.The exciting MPPT methods have disadvantages with partial shading in PV and solar temperature and load conditions (Chen et al., 2022;Jamil et al., 2022;Singh et al., 2022).MPPT algorithms had many difficulties in partial shading conditions due to obstructions such as buildings, cloudy weather, trees, etc.These difficulties might lead to more points of local maximum power, and consequently, the conventional MPPT algorithms, such as PO and INC, may struggle to identify accurately the global MPP.In verse, in these situations, it could stick in local MPPT.Another matter, partial shading causes rapid oscillations in the output voltage and currents of the solar PV panel.These oscillations may lead to misleading toward the incorrect tracking of MPP.The voltage and current oscillations influence the microgrid efficiency and reduce the lifetime of the power electronic switches.Besides that, the sudden and dynamic behavior of the MMPT algorithms during shading conditions made the conventional methods respond immediately.As some of the MPPT methods take significantly longer than the global MPP, this will affect reducing the energy extracted during the variation of the irradiance Afterward, two loads (static and dynamic) were used to test the proposed controller.In addition, the proposed controller is tested with constant and variable solar irradiances.This paper aims to control the maximum power point of an integrated microgrid during partial shading conditions (i.e., variable solar irradiance).The proposed controller is Shallow Neural Networks using the deep learning NARMA-L2 controller.The modeling uses the prediction of the reference voltage under different irradiance.However, dynamic PV solar has been trained to track the maximum power drawn from the PV solar systems in real time.

THE PROPOSED METHODOLOGY
It's a challenge to extract the controller for obtaining the maximum power from a PV solar with variations of solar irradiance and temperature (Jamil et al., 2022).The plant identification process lets training a neural network to model the grid.It should identify the grid before training the controller whenever it is unsatisfied.The controller needs to be identified (Wen et al., 2019;Saadaoui et al., 2023).The proposed microgrid consists of solar PV, a utility grid, an energy storage system, and a load, as shown in Fig. 1.

THE SYSTEM MODELING
The system is implemented with MATLAB/Simulink, as shown in Fig. 2.

Load Model
The loads consist of static and dynamic, as given in Table 1.

Load type
Parameters RL load 10 kW active power and reactive power of 5 kVar Induction motor 3-phase,5HP, 400V,50Hz

PV Solar System Model
The equivalent circuit of the PV cell is shown in The I/V and P/V characteristics are shown in Fig. 4 under different irradiance and specified temperatures of 25 o C. The solar PV parameters for the simulation are presented as follows: MPP is 10100W at 25Co, a current of 42A, and a voltage of 390V.The DC voltage will decrease when the temperature increases when the irradiance is 1000W/m 2 .Fig. 5 gives the PV specification.
where u(t) is the system input, and y(t) is the system output.The neural network is trained in the nonlinear function F.Then, the system output is The nonlinear controller is developed as follows:

Battery Storage Model
The battery storage system is concerned with the microgrid to compensate for the fluctuations in the renewable energy penetration.This system depends on the microgrid's charging and discharging requirements (Asadi et al., 2023;He et al., 2023).
The constraints of the state of charge equation are given as follows: The battery's SOC is limited to 30% and 70% of its power in ampere-hour capacity.It prevents the battery life from undercharging or overcharging.The limits of the charging battery power are The limits of the discharging battery power are The proposed bidirectional controller for the battery storage is modeled with Matlab/Simulink, as shown in Fig. 10.It is clear that the battery will charge or /discharge depending on the degradation between the power generation and the load demands.The system is tested under different types of loads (static and dynamic loads) to verify the proposed system's effectiveness and performance.Fig. 18 shows the performance of the load voltage and current with the static load.Fig. 19 shows the load voltage and load current under dynamic load.It is observed that the proposed system behavior does not influence the load type.The proposed MPPT algorithm optimizes the generation and demand powers.

INC
Complex computations, unstable under partial shading

NARMA-L2
It is adaptive to track MPP, handling complex and nonlinearities, robust to weather variations and shading profiles, resilient to load dynamics, and fast response.
However, an optimal balance between the generated power from the PV solar and the demand power impacts the overall system efficiency and energy utilization.However, the dynamic load adds challenges to energy utilization.The simulation results illustrated the fast response, and the NARAM-L2 had more agile adjustments to track the varying MPP and verify the dynamic load requirements.It can figure out the implementation of deep learning NARAM-L2 controller MPPT for different loads, as given in Table 3.  Type of load Performance static constant power consumption, stable and efficient in partial shading conditions dynamic Unpredictable power changes, Deep learning MPPT adapts the power extraction, system efficiency, and energy utilization.

CONCLUSIONS
presents MPPT algorithms under partial shading.Although the conventional MPPT methods, such as PO-MPPT and INC-MPPT.These methods are simple and easy to implement, but they have limitations in addressing challenging conditions such as irradiance variations and dynamic load.Therefore, a deep learning toolbox NARMA-L2 controller is proposed and used to track PV solar MPP.This controller can effectively handle the dynamic behavior of PV solar in real-time.As well as it has an advantage in adapting the changing irradiance to enhance the system performance and is capable of harvesting the energy while balancing the generated and consumed powers.Moreover, the NARMA-L2 controller is tested for dynamic load.The simulation results demonstrate the effectiveness of the NARMA-L2 in tracking the global MPP under different irradiance profiles, and it achieves a good response of load voltage and current for static and dynamic loads.

(
Alcaide et al., 2022; Dagal et al., 2022; Isknan et al., 2023; Pervez et al., 2022).Some researchers studied the impact of measurement errors caused by shading conditions (Jane et al., 2022; Millah et al., 2022).The PV solar and nonuniform configuration design may induce an imbalance in power losses (Habbi and Alhamadani, 2018).Therefore, using an effective MPPT under partial shading limits the microgrid performance (Burlacu and Navrapescu, 2022; Rao and Sundaramoorthy, 2022).Maximum power point tracking has been used with the boost converter since the MPPT generates the switching pulses for the dc-to-dc converter(Zhou et al., 2016).The DC voltage of the inverter is independent of (Millah et al., 2021; Gadiraju, 2022)the output voltage of the solar PV.PO-MPPT had a problem with steady state oscillations near to maximum point and in local tracking instead of globally in partial shading conditions (Kulkarni and Deshmukh, 2019; Vanti et al., 2022).On the other hand, the INC-MPPT has higher accuracy than the PO-MPPT, but it is more complex due to its dynamic operation and selecting the step size, which is becoming smaller, reaching the steady-state error(Paduani et al., 2022).That will lead to concluding that the conventional MPPT algorithms had the demerits of inaccurate output power setting and oscillation because of the slow tracking of the calculation (Kulkarni and Deshmukh, 2019; Awan et al., 2022; Yang et al., 2023).A recurrent neural network has been used for load demand forecasting in smart grids.This scenario is accomplished based on the consumer demand pattern (Kaushal and Basak, 2020; Nhung et al., 2022).Hybrid Fuzzy-particle swarm optimization-based grid system has been used to predict the model to improve the accuracy, but it has limited performance when the data are huge (Azeem et al., 2022).Addressing the challenges mentioned above and problems during the operation of solar PV under partial shading.This paper proposed a NARMA-L2 controller to use and obtain the maximum power in the microgrid-connected system during partial shading.This controller depends on real-time information and tracks the global MPP accurately, even in partial shading conditions.The simulation results are obtained and compared with conventional methods such as Perturb and observe (PO-MPPT) and Incremental Conductance (INC -MPPT) (Mohamed and Habbi, 2020; Jamil et al., 2022).

Fig. 6
shows the block diagram for the solar PV model based on the MPPT boost converter.The two conventional MPPT, P&O and INC -MPPT algorithms, have been used.The codes are given in Appendix A for both methods (Abdul Hussain and Habbi, 2018; Kulkarni and Deshmukh, 2019; Ahmed et al., 2022).

Figure 4 .Figure 5 .
Figure 4. I/V and P/V characteristics of the PV cell

Figure 10 .
Figure 10.Bidirectional converter for battery storage system

Table 3 .
MPPT based on NARMA-L2 controller with different loads Comparison.