Cascaded Feed Forward Multilayer Neural Network based MPPT Controller for Improving the Performance of Photovoltaic System

Mr. M Rupesh, Dr. T S Vishwanath, Dr. M Venkatesh Kumar Research Scholar, EE VTU, Belagam, Karnataka, India, and Assistant Professor, Department of EEE, BVRIT HYDERABAD College of Engineering for Women, Hyderabad, India. Professor, Department of ECE, Bheemanna Khandre Institute of Technology, Bhalki, Karnataka, India, Assistant Professor (Sr. G), Amrita School of Engineering, Amrita Vishwa Vidyapeetham, Chennai


I. INTRODUCTION
The quick growth in energy consumption, CO2 emissions and the global inadequacy of demand and supply is due to the increasing rate of population growth and levels of urbanization [1] [2]. Under the environmental concern like shortage of energy and pollution, RES such as Solar and Wind are the most suitable replaced energy sources which are presence the foremost energy of recent power systems. Micro grid (MG) integrates flexible DGs, such as wind power, solar power and fuel cells, with manageable storage and loads to form a low-voltage distribution system [3][4][5][6]. They improve the reliability of the network and provide durability and quality electrical energy. Managing an MG with an extensive selection of Distributing Generations, fluctuating loads and ESA is a difficult task, particularly below the high level of penetration of renewable energy (RG) generation. The RG is typically organized using maximum power tracking algorithms to emphasize the high use of efficiency energy [7][8][9][10]. It is therefore considered a generation that cannot be modulated because of changing and uncontrollable weather conditions [11][12][13][14][15]. Maximum power monitoring technology will play a major role in producing maximum energy from a photovoltaic cell in various weather conditions. A major challenge to regulate the PV generated non-linear current characteristics and voltage during periods of low sunshine or partially shaded situations. Many MPPT algorithms have now been proposed by the researchers to achieve maximum energy production from a photovoltaic system. Among many little MPPT techniques are very popular like P&O, Incremental Conditions, Feedback voltage and current, voltage and frequency method, feedback power methods. The above conventional methods fail to achieve the speed of operation and maximum power production due to lack of self-regulation capability. In this document proposed various intelligent controller based on MPPT techniques to achieve maximum power output as well as operating speed (auto-adjustment). MPPT techniques based on intelligence are modelled and analyzed in the MATLAB environment. Section 2 provides the mathematical model for the photovoltaic system and design of boost converter. The various Smart Controllers are addressed in Section 3. The proposed intelligent controllers are modelled in Section 4 intelligent Controllers are simulated in MATLAB and analyzed the performances of PV system with various weather conditions. Finally, the hardware and comparative study is presented in section 5. Conclusion of the proposed research is delivered in Section 6.

II. MPPT SYSTEM
RGs plays a vibrant role in reaching consumer's energy requirement because of their ampleness and lower environmental influence. The key obstacle to expanding PV power is the capital cost of implementing the PV system [8]. The production of energy through PV is not constant throughout the day because of climate change. The productivity of electricity production is very low (the productivity is in between 10-17% in low radiation regions). Consequently, MPPT technologies play a vibrant role in the production of PV energy for best energy production in various environmental circumstances. Various MPPT methodologies are discussed below.

II.I Mathematical Model of Photovoltaic System
The photovoltaic cell is depicted as a Dependent current source parallel to a diode as shown in the Figure 1 including supplementary series and parallel resistors. It should be noted that there will be no power generation in the absence of solar light and that the PV cell will act as a diode. True current from the current source (PV cell) depends on the daylight effectively falling on the PV cell (photo-current) (Figure 1) Cascaded Feed Forward Multilayer Neural Network based MPPT Controller for Improving the Performance of Photovoltaic System ________________________________________________________________________________________________ 7010

II.II. Boost Converter Design.
Boost Converter is working as a DC-DC step up converter, which is used to convert fluctuating DC voltage with respect to change in weather conditions to constant stepped up voltage to connect to the inverter for grid integration and domestic applications, here the boost converter is designed with diode, MOSFET as switch and load element to capture the output voltage. The output voltage is varied according to triggering duty cycle. Fig.2 shows the basic structure of boost converter.

III.I. Fuzzy Logic (FL) Controller Algorithm
The FL controller is the most popular expert systems and proven intelligent algorithm. The basic fuzzy control block diagram is shown in fig.3.a the fuzzy logic supervisor has two inputs such as a modification in voltage and modification in power as shown in

III.II. Genetic Algorithm
GA is an algorithm for stochastic optimization through natural genetic selection. The process takes OC Voltage and Isaac as contributions and it results the best maximum power current (Imp) using the modelling of the cell, without knowledge of the ordinance and temperature [16][17][18]. The process is developed using MATLAB Code in the embedded Simulink block by the essential functions, with the Genetic Algorithm constraints and PV constraints are respectively the parameters Genetic Algorithm and PV panel. The fig.4 shows the flowchart of Genetic algorithm.

III.III. KGMO Algorithm
The basic principle of gas molecules optimization algorithm (KGMO) was suggested with the concept of laws of gas molecules. It develops laws based on experiential explanations to convey the macroscopic performance of gas particles. The nuclear theory of gases tells that all materials consist of a great number of molecules or atoms. The properties of gases like pressure, volume and temperature are the outcomes of the achievement of the particles that comprise the gas. There are five theories that define how molecules perform in a gas. The molecular kinetic concept of 4. There are no alluring or repulsive forces exist among the particles. 5. The avg KE of a particle is 3, it is=2, when T is the out-and-out temp, and k is Boltz const.
This segment outlines the kinetic optimization development with gas molecules. The gas particles moves with in the vessel till they meet into the part of vessel which has the least temp and KE. Gas particles interest each other on the basis of low intermolecular electric forces, where the electric force is the outcome of progressive and adverse loads in the particles. In this method, each gas particle and the agent, has four features: position, KE, speed & mass. The kinetics of every gas particle determine the speed and location of the gas molecule. In this method, the gas particles discover the whole research space and achieve the lowest temp. The fig.5. Shows the flow chart of KGMO algorithm

III.IV. Cascaded Feed Forward Neural Network (CFNN) (Machine Learning based) MPPT Algorithm
The CFNN is a feed-forward (FF) neural network, s but it includes a link from the contribution layer and each prior layer to the following layers. In a three-layer network, the production layer is also associated openly to the contribution layer adjacent to the concealed layer. As with FF networks, a cascading network with 2 or more layers ________________________________________________________________________________________________ 7013 may learn any arbitrarily finite I-O relationship with enough hidden neurons [19][20][21][22]. The CFNN can be used for any type of contribution to the production cartography. The benefit of this technique is that it takes into account the non-linear association among entry and exit without eradicating the linear affiliation between the two.
In the perceptron assembly which is designed among the contribution and the production is a form of undeviating association, whereas on FFNN the assembly formed among the contribution and the production is a secondary association. The linking is non-linear by an activating function in the concealed layer. If the linking form on perceptron and multilayer grid is combined, then the grid with a direct linking among the contribution layer and the production layer is molded, moreover the linking incidentally. The network formed from this connecting model is known as the CFNN.
The network molded by this linking model is named a cascading neural network (CFNN). The equations can be as follows: Where fi is the activation function and wii is weight from the contribution layer to the production layer. If a bias is introduced to the contribution layer and the activation function of each neuron in the concealed layer is fh then equation becomes In this investigation, the CFNN model is applied in time series data. Thus, the neurons in the contribution layer are the lags of time series data Xt-1, Xt-2, ..., Xt-p, whereas the production is the current data Xt.
The proposed multi-layered cascade neural network model has been developed for an algorithm to trace maximum power points for the PV system as shown in fig 6.a.. This network is equipped with two contributions such as PV voltage and PV current. The production of this network is the duty cycle of DC-DC converter. There are 4 concealed layers used from the contribution layer to the production layer.

IV. SIMULATION RESULTS
In this paper that the finding Maximum Power Point using different algorithms has been implemented and the simulation results are as shown below.  fig. 9.c.

V. HARDWARE RESULTS
The proposed system was developed as a functional model of 10W PV system and power converter as shown in Figure  12.a. The proposed CNFF algorithm has been interfaced with Arduino mega 2560 controller, which is helpful to develop duty cycle based on input changes. Arduino's Mega 2560 interfaces directly with MATLAB using a MATLAB simulation library. Design data can be found in Table 2. The switching pulses are generated by the proposed algorithm and its run cycle will change with respect to climate change. Figure 12.b accounts for 50% of the utilization cycle generated by the proposed algorithm.

VI. CONCLUSION
This research has focused on the maximum energy production of photovoltaic systems in a variety of meteorological conditions. The mathematical model of the photovoltaic cell has been developed and analyses its performance in different weather conditions. According to the simulation results, the MPPT algorithm was inevitable to generate the maximum power of the PV system. In this study, a number of MPPT algorithms were tested in a variety of meteorological conditions. The following algorithms were analyzed, namely  Fuzzy, GA, KGMO and CNFF. Based on simulation results and comparative analyses, the CNFF produces better results relative to other MPPT algorithms. Comparative analyses can be found in Table 3 and Figure  13. Finally, the prototype work model was developed and checked by the proposed MPPT algorithm.