Smart Control Solution for Single-Stage Solar PV Systems

Solar photovoltaic (PV) systems unpredictable characteristics and tight grid-codes demand power electronic-based energy conversion devices. Hence, as the power levels generated by the solar PV systems rise, multi-level voltage source converters (VSC) and their control mechanisms become more necessary for effective energy conversion. Continuous control set model predictive control (CCS-MPC) is a class of predictive control approach that has emerged recently for the applications of power converters and energy conversion systems. In this paper, an artificial neural network (ANN) based controller for single-stage grid-connected PV is implemented. The CCS-MPC is used as an expert / a teacher to generate the data required for off-line training of the neural network controller. After the off-line training, the trained ANN can fully control the inverter’s output voltage and track the maximum power point (MPP) without the need for MPC during testing. The proposed control technique is assessed under various operating conditions. Based on the results obtained, it is observed that the proposed techniques offer improved objective tracking and comparative dynamic response with respect to the classical approaches.

Introduction

Nowadays, Photovoltaic power has become very relevant due to the environmental impact and shortage of fossil fuels. The amount of electric power generated by solar arrays varies with weather conditions, and electric power generation conversion efficiency is low. As a result, even at maximum power, the energy conversion rate is low. Solar panels provide energy to the system by extracting it from the sun, and excess incident solar rays are either converted to heat or reflected into the atmosphere. As a result, we use maximum power point tracking (MPPT) techniques to get the most power out of the PV system [1]. With the usage of DC-DC converter topologies (buck converter, boost converter, buck-boost), the MPPT is utilized to maximize the solar array power output regardless of meteorological conditions (such temperature and radiations) and electrical characteristics. The power curve of a typical PV system is depicted in Fig. 1. The power curve is critical because it shows the panel’s predicted power generation depending on the voltage (“V”) and current (“I”) generated by the panel.

Fig. 1. Solar PV power curve.

Recently, there has been an increasing deployment of Micro grids in electricity networks, from the power plants all the way to the consumers of electricity in homes and businesses. The grid refers to the networks that transport electricity from power plants to homes and businesses. Wires, substations, transformers, and switches are all part of the grid [2]. The need to adopt solar PV generation arises since it’s a cheaper and better option to adopt with this recent development. The main advantages are considerable increases in energy efficiency on the power grid as well as in the homes and workplaces of energy consumers.

The most difficult challenge that practically all researchers confront when integrating a solar system into the utility grid is poor power quality, voltage instability, an insufficient grid system, and stability. Such circumstances have caused a variety of different types of equipment damage. Because these consequences affect not only an individual but also a location or sector, the appropriate action and control strategy for minimizing the magnitude of this challenge must be proposed, which is explored in this paper.

Furthermore, as the power levels of solar PV systems rise, multi-level voltage source converters (VSC) and their control mechanisms become more necessary for effective energy conversion. A current research issue is the design and development of innovative control methods for grid-connected solar PV systems. The major control requirements for a grid-connected PV system include maximum power extraction, DC-link capacitor voltage balancing, leakage current reduction, ideal current/power tracking, fast dynamic response, lower current THD, and lower switching losses. However, several classical control schemes are available in the literature, out of which voltage-oriented control (VOC) with space vector modulation (SVM) is the most widely used control scheme for the solar PV system. However, the difficulty in the design procedure for these control schemes to integrate multiple objectives of a solar PV system has inspired the research of advanced control schemes for the single-stage solar PV system. Single-stage PV configuration is shown in Fig. 2.

Fig. 2. Single-stage solar PV configuration.

This paper proposed a new control technique for a single-stage grid-tied PV inverter that combines CCS-MPC with feed-forward ANN with the goal of tracking the MPP under partial shading and different irradiance conditions, lowering THD and increasing the system’s steady and dynamic performance. CCS-MPC is first utilized in the training phase as an expert to create data for training the proposed neural network.

Single-Stage 3-Phase Grid-Tied Solar PV with Artificial Neural Network

Model predictive control (MPC) has become one of the well-established modern-day control approaches for single-stage grid-tied PV inverters with an output AC filter, where a high-power quality with low harmonic injection is required, and to track the MPP under partial shading condition and different irradiations. Although it is simple to learn and use controller, it has a significant drawback of necessitating large number of online computations to solve the optimization issue [3]. In the field of power electronics and drives, on the other hand, the use of artificial intelligence techniques, such as artificial neural networks (ANN), is fast increasing.

The neural network may be successfully utilized online for maximum power point tracking without the requirement for CCS-MPC once it has been tuned. MATLAB/Simulink environment was used to simulate and evaluate the proposed ANN-based control approach under different environmental conditions. The first phase setup of the proposed neural network is presented in Fig. 3. Furthermore, the performance of the ANN-based controller is assessed and compared to that of CCS-MPC under different environmental conditions.

Fig. 3. Proposed artificial neural network training phase.

Overview of Artificial Neural Network

Artificial intelligence (AI) is fast advancing and has become one of the most important research topics in recent years. The goal of AI is to provide systems with intelligence capable of learning and thinking in the same way that humans do [4], [5]. It offers a lot of benefits and has been used in a lot of different industries, including image classification, voice recognition, driverless automobiles, computer vision, and so on. The advancement of AI benefits power electronics, which has enormous potential. Some of the applications of IA include maximum power point tracking (MPPT) control for solar and wind energy conversion systems [6], [7], design optimization of power module heatsinks [8], remaining useful life (RUL) prediction for supercapacitors [9], intelligent controller for multicolour light-emitting diode (LED) [10], anomaly detection for inverters [11], [12], etc.

Artificial Neural Networks (ANN) are artificial intelligence neural network models. A learning algorithm or learning rule is coupled with a model. An ANN model is a type of function that may be created by changing parameters, connection weights, or design features like the number of neurons or their connections. A neural network model may be thought of as a representation of current knowledge about how neurons work and interact. Since feed-forward networks do not contain loops, they may be arranged into layers and utilized to provide memoryless, or dynamic, input-to-output mappings. A mathematical example of a neuron is shown in Fig. 4.

Fig. 4. Mathematical example of a neuron [13].

Where, (xq−xn) is input vector, (w1−wn) is weights for each input, f is activation function and b is bias (also a weight). Advantage of feed-forward network is its simplicity, it is possible to differentiate one input layer, hidden layers, and one output layer that connect the input to the output. The ANN training goal is to optimize the cost function by finding optimal values for wi and b. Although recent research has focused on larger scale challenges such as deep learning, also, new strategies to increase the dependability of smaller networks have also been proposed in [14], [15]. Hardware vendors have begun to implement reduced precision floating point [16], [17] and integer arithmetic [16], as well as small scale, specialised architectures [18].

A multilayer perception (MLP) is a type of feedforward neural network in which each layer is completely linked, and the number of nodes in each layer is the same in certain configurations. Furthermore, the activation function across hidden layers is often the same in several configurations [19]. This is shown in Fig. 5. Each layer that feeds into the next links to all nodes in the following layer, as can be seen.

Fig. 5. Multilayer network [13].

This approach is used when there is not much information about the structure of the problem [13]. Using just completely linked layers, as defined by an MLP, allows structure to be learned rather than imposed. While the depth and width of the network may be increased, this just adds to the flexibility of function approximation [20].

In this paper, the control model was implemented using a feed-forward neural network because of its functions as a universal approximator. The number of input and output units are bound by the number of input and output variables, respectively, and a grid search tuning technique permitted the selection of a configuration with 15 units in the hidden layer. The Levenberg-Marquardt approach was used for training since it has a strong optimization convergence characteristic [3].

Steps for ANN Training Phase

Training an ANN is an iterative process in which the network is fed with training data examples one by one, and the weights are modified each time. The measured input variables in dq coordinates supplied to the ANN are the filter current, the output voltage, the DC link voltage, the reference current, and angle from PLL. The neural network takes both the real and imaginary parts of these variables, increasing the total number of input characteristics to eight. The optimal modulation signal to be applied to PWM at each sampling moment is the output of ANN. If only a few neurons are picked for each layer during the training phase, the network will be extremely simple and will not be able to capture the training data properly [21]. However, when a network is constructed with too many neurons, the trained network matches the training samples well but fails to uncover the natural structure of the input/output data connections. After fine-tuning the ANN, the trained ANN can be utilized to operate the system shown in Fig. 6.

Fig. 6. Proposed trained ANN controller.

Simulation Results and Discussion

The proposed single-stage ANN controller for grid-tied PV has been implemented in MATLAB/Simulink as shown in Fig. 6. The simulation was carried out using the parameters presented in Table I. The training was carried out multiple times using Levenberg-Marquardt algorithm. The network accepts eight inputs, and the hidden neurons are fifteen with an output modulating signal of three. The results will be analysed in subsequent sections to evaluate the capability of ANN controller for solar PV.

Table I. System Parameters

Parameter	Value
Nominal power (Ppv)	100 kW
Grid voltage (Utility)	25 kV
Open-circuit voltage (VOC)	685.768 V
Grid frequency (f)	50 Hz
Filter inductance per phase (Lf)	500 µH
Filter resistance (Rf)	3.77 mΩ
Capacitor bank	10 kVar
DC link capacitor	3 mF
Switching frequency fsw	10 kHz
PV model	SunPower-305E

Simulation Result for Linear Irradiance Change

The system has been simulated using the irradiance profile Fig. 7 to evaluate the system performance. The simulation time starts at t=0 s, the MPPT was tuned on at t=0.1 s, and the maximum power point was tracked. Fig. 8 shows the results, output power, dc-link voltage, voltage at the PCC, and AC output current. It can be observed from Fig. 8 that the proposed ANN controller was able to track the maximum power point within few milliseconds and send power to the grid according to the PV irradiance profile in Fig. 7.

Fig. 7. Irradiance profile applied for verification.

Fig. 8. Simulation results with the ANN control under linear irradiance change: (a) DC-AC power, (b) DC voltage, (c) output AC voltage and (d) AC output current.

The proposed controller has been able to ensure high power quality extraction within the range of the irradiance and even with the step changes in irradiation. The accuracy of the ANN controller can also be measured from the results on how the Vdc and the output current were able to follow the MPP tracking profile. The current and voltage THD values recorded in Table II, shows that the system performance satisfied the 5% grid interconnection standard.

Table II. Comparison for Current and Voltage THD for the Irradiance Profile

Time (S)	ANN Controller		CCS-MPC
	I THD (%)	V THD (%)	I THD (%)	V THD (%)
0.2	4.02	0.06	3.14	0.03
0.4	2.95	0.03	4.29	0.06
0.8	5.9	0.10	5.59	0.04
1.7	3.66	0.04	3.74	0.04
1.8	2.87	0.04	3.66	0.04

Simulation under Partial Shaded Condition

The system shown in Fig. 9 has been simulated using the irradiance profile of ideal PV depicted in Fig. 10 to assess the system performance under partial shading effect. The simulation time starts at t=0 s, the MPPT was tuned on at t=0.1 s to track the PV curve, and the MPP was tracked. Different irradiance values, [1000 700 400] to [1000 500 700] W/m2 and [800 1000 600] to [1000 400 800] W/m2 were used to simulate the system, and the power and voltage recorded are shown in Table III.

Fig. 9. PV connection under partial shaded condition.

Fig. 10. Irradiance of [500 800 1000] for ideal PV curve.

Table III. Difference Irradiance Values with Their Power and Voltage Values

S/N	Irradiance (W/m2)	P (kW)	V (V)
1	[1000 700 400]	38.16	278.1
2	[1000 500 700]	35.2	253.36
3	[800 1000 600]	36.8	261.07
4	[1000 400 800]	38	269.79

The simulation results are shown in Figs. 11 and 12.

Fig. 11. Simulation results with the ANN with an irradiance change [1000 700 400] W/m2: (a) Output power, (b) PV voltages, (c) output AC voltage and (d) AC output current.

Fig. 12. Simulation results with the ANN with an irradiance change [1000 400 800] W/m2: (a) Output power, (b) PV voltages, (c) output AC voltage and (d) AC output current.

From the simulation results presented in Figs. 11 and 12, it can be observed that the proposed ANN works perfectly under partial shaded conditions as it was able to track the current reference according to the applied irradiance profile, and the MPP and the DC voltage were able to track their references.

Simulation Results under Fault conditions

The setup implemented for the simulation is presented in Fig. 13. The system has been simulated using the irradiance profile in Fig. 14 to evaluate the system performance under fault condition. The simulation time starts at t=0 s, the MPPT is tuned on at t=0.1 s. The results gotten from the analysis are presented in Fig. 15. The plots for the output power, DC-link voltage, voltage at the PCC, and AC output current were all taken into consideration.

Fig. 13. PV system under fault condition.

Fig. 14. Step irradiance change profile.

Fig. 15. Simulation result for ANN when fault was introduced to the system: (a) Output power (kW), (b) DC current, (c) Output AC voltage, and (d) Output AC current.

Symmetrical fault was introduced into the system during the operation between 0.02 to 0.15 s. The behaviour of the system under symmetrical fault can be observed in Fig. 13. The grid voltage is zero throughout the fault period while the grid current spike very high at the same period. When the fault caused instability on the system during the time the fault lasted. Similarly, when the fault period elapsed, the system becomes stable again and was able to track the maximum power when compared to the irradiance curve in Fig. 14. The instability introduced in the system by the fault can also be seen in the DC current reference tracking in Fig. 15.

The THD observed at the middle of the fault at 0.05 s is 1.44% for voltage and 36.82% for current. At the fault time, voltage in each phase falls to zero while the current increases extremely high. At the fault clearing time the PV system returns to normal operation and the system stability was restored.

Conclusion

In this paper, a smart control system for single-stage grid-connected solar PV has been implemented. Artificial neural network (ANN) is the advanced controller implemented in MATLAB/SIMULINK environment. Maximum power point tracking (MPPT) technique has been utilized for the proposed controller. The harmonic component in the output current has little effect on the system, and thus high-quality current is fed to the AC grid. The cost function is designed to reduce the current tracking error and the average switching frequency. The tracking of current reference, and THD below 5% was achieved with a tracking error of about 0.33%. Accurate current tracking will improve the current THD, and a lower average switching frequency will reduce the switching loss. The quality of current injected into the grid was observed in terms of current THD. The THD for voltage and current met the IEEE standard for grid interconnection. Fault analysis was conducted to verify the fault handling capability of the proposed system. The controller has an improved capability to track the reference current during transients and a decoupled current control function. The proposed model can be an excellent option to build a greener environment for the future by improving the current power generating PV system by minimizing current THD and power loss.