A Photovoltaic (PV)-Wind Hybrid Energy System Using an Improved Deep Neural Network (IDNN)-Based Voltage Source Controller for a Microgrid Environment ^†

Abstract

Presently, there has been a huge rise in the demand for power owing to increases in population and commercial organizations. Traditional power plants are not able to keep up with the increasing needs of customers. Finding a different way to meet consumers’ needs is the main problem in the current situation. Most RESs (renewable energy sources) like wind, solar, hydro/water sources and fuel cells are environmentally beneficial. The number of available resources has no bearing on how much electricity can be produced using RESs. Due to differences in natural resources, there are constant fluctuations in the availability of RESs. In this technical study, two significant RE (Renewable Energy) power sources—PV (photovoltaic) cells and WES (wind energy systems)—are studied in various weather scenarios. First, a cutting-edge intelligent controller system was created, which aids in tracking the peak power point. Due to the unpredictable nature of weather, a MPPT (maximum power point tracking) controller is required for RES. This work aims to present IDNN- (improved deep neural network) and MPPT-based unique methods for power generation using solar and winds. When a hybrid PV/WES system is integrated into MG s(microgrids), power quality may be improved and THD values can be reduced. It was confirmed from the results of the simulation that the proposed IDNN system yields better performance in different operating situations by means of lower MSE (mean square error) rates, lower THD (total harmonic distortion) and lower computational complexity than the existing method.

1. Introduction

In the past decade, global warming has become a critical challenge, and consequentially, it is a huge threat to global life. Due to their simplicity, accessibility, and non-offensive nature, PV and WP (wind powers) have important roles in electric power generation [1,2]. MGs and smart grids can be created by integrating RESs with primary electrical power grids.

Applying rapidly developing RESs is facilitated by using improved WP and PV reproductive technologies, aiming primarily at decreasing program costs. Electrical energy from RESs, including WP and PV, will see a rapid drop in prices compared to electricity, as mentioned in [3]. The size, characteristics, and features of WT and PV technology may significantly influence system effectiveness in terms of financial indicators such as the cost of electrical energy in thermal power plants, the quality of the electricity and electricity loss.

Due to their simultaneous use of a number of alternative energy sources, hybrid RESs are becoming one of the most often-used technologies to produce electricity. Load disparities, voltage fluctuations, poor loads and qualities, frequency fluctuations and reliability are a few of the challenges faced [4]. High power flows like OPF refer to a power system’s general consistency in performances, which is made possible by the control system’s appropriate flexibility. Mathematically speaking, this is a significant nonlinear static issue as well as a challenging linear transformation problem.

Due to the OPF’s problem of transient energy aspects, multiple WP integrations and the PV system’s power supply modifications exacerbate problems. To explain and address OPFs caused by wind or PV technologies, the following factors need to be integrated into singular models: (i) significances and underlying causes of problems; (ii) calculations of WP and PV power productions based on consistent wind speeds and solar radiation attributes; (iii) establishment of objective functions; (iii) adoption of comprehensive views of technical challenges, adaptive controls and reliable adaptabilities; and (iv) determination of handling of OPF problems. Recent studies on OPF issues by various scholars have focused on a few of the aforementioned functions. To improve the role of RESs in the electrical grid, which is helpful in minimizing these problems, new and innovative concepts and technologies are required.

For a PV/Wind grid-integrated system, a control mechanism for power flow management has been used, which depends on batteries that are an ESS (energy storage system). The BESS (battery-based energy storage system) is useful for regulating the BESS active power transfer and includes both a conventional droop control and an inertia simulation function. Moreover, if there is a degradation in the MG power quality, the BESS separates and continues to function in isolated mode. However, because the BESS is a high-rated component and has a direct connection to the dc-link, it may negatively impact battery performance and shorten its lifespan. Battery module performances can be improved by integrating dynamic- and high power densities-based ESDs (energy storage devices), including ultracapacitors, with current systems. HESS (hybrid energy storage system), a combination of ESDs that includes batteries and ultracapacitors, is being used in MGs. Therefore, using ultracapacitors may increase long-term performances of battery modules. The main advantages of PV/Wind hybrid systems are given as follows:

• BESS serves the purpose of regulating active power transfer, incorporating both conventional droop control and an inertia simulation function.
• Enhancements in battery module performance can be achieved through the integration of dynamic- and high-power-density-based ESDs, such as ultracapacitors, with current systems. HESS, a combination of ESDs comprising batteries and ultracapacitors, are employed in Microgrids. Consequently, the utilization of ultracapacitors has the potential to enhance the long-term performance of battery modules.is useful for regulating BESS active power transfer and includes both a conventional droop control and an inertia simulation function.
• This work aims to examine PV cells and WESs under different weather scenarios in a grid environment. There are several research techniques presented, however better MSE and THD has not been considerably ensured [5]. To solve the above stated problems, in this technical work, IDNN is used to yield performance improvements in the overall MG. The primary contributions made by this technical work involve modelling of a MG, use of PV batteries for grids, managing power and frequency deviations in the MG and controlling load frequencies of the MG. The proposed technique is helpful in achieving increasingly accurate results using efficient algorithms for the provided setup.

This work’s remaining sections are as follows: Section 2 includes literature reviews for power and frequency deviation in the MG and load frequency control in the MG. Section 3 studies the proposed power and frequency deviation in the MG and load frequency control in the MG in detail. Section 4 introduces the results of the experiments. This work concludes in Section 5.

2. Related Work

In Ref. [1], Adefarati et al. (2019) studied allowing the load to be shared between the elements of a MG system in the form of a condition to make the best use of the monetary advantages of RE technologies. MATLAB2021b- was used to implement this optimization technique, and it is shown from the results of the simulation that the approach is inexpensive. The genetic algorithm is quite simple in terms of implementation and the computational time for solving issues in power systems was reduced. The suggested algorithm had the capability to reduce overall costs associated with MGs.

Al-Quraan and Al-Qaisi (2021) [2] constructed and managed a stand-alone micro-grid system that uses PV and PMSGs (permanent magnet synchronous generators) based on WECS (wind energy conversion system). When the energy of PV panels was greater than that of the demanded loads, batteries were charged, while in times of inadequacies, powers were drawn from previously charged batteries. Furthermore, controllers contributed to the security of battery storages (maintaining added charges and discharges). When the SOC (state of charge) values of batteries were between 20% and 80%, controllers operate normally, but when the values of SOC exceeded 80%, wind turbines and PV panels were turned off and stayed in this state till the SOC value fell below 75% in controllers. Inverters were cut off and loads eliminated when the SOC decreased by less than 20%. The electricity to the inverter was switched off till there was adequate charging of the batteries.

In Ref. [6], Qadir et al. (2021) proposed machine learning model categorization that might be beneficial and effective for calculating power and energy. The study found relevant patterns for learning and subsequently selected features with linear regressions where the results showed better performances than other models in evaluations with MSE (mean squared error) values of 0.000000104, MAE (mean absolute error) values of 0.00083 and AE (absolute error) values of 0.00083. It can be inferred from the results that the proposed sustainable computational approach has great prospects for improving the efficacy of smart grids through the prediction of energy generated by RESs.

In Ref. [7], Refaai et al. (2022) discussed grids based on LR to predict distributed energy resource powers. An RES with solar arrays, based on their presented design, had battery banks working as energy storage devices. The financial aim was to incur the least amount of power bills, while the technical goal was to achieve max charge states for batteries and the environmental goals were to reduce emissions. Sliding window predictions, which use information from PV output energies and load demands, were effective in assessing the LR model’s performances in over 1000-run time frames. Future studies may attempt to use other deep learning algorithms to increase this technique’s performance, accuracy and other metrics.

In Ref. [8], Tom and Edward (2021) suggested a solar/wind-battery-based HRES (hybrid RES) for grids, allowing sole management of these systems and meaningful usage of the RES. When LUO converters were used, PV outputs were improved and closed loop systems based on the crow search algorithm were used to regulate them. In wind energy conversions, DFIG (doubly-fed-induction generator) outputs were converted to DC using PWM rectifiers operated by PI controllers. Bidirectional buck-boost converters were utilised for batteries, and ANNs (artificial neural networks) were employed to maintain the battery’s SOC. This work aimed to estimate LVRT- utilising signal processing techniques based on DNNs (deep neural networks) [9]. Notch filters were used to pre-process data for suppressing noises, while Hilbert transformations segmented the data and SIFT extracted features [10]. A DNN classifier, using training and test data, was used in classifications, and subsequently LVRT (Low-Voltage-Ride-through) estimates were undertaken. The system was implemented with the aid of MATLAB, and the outcomes were examined. The estimated LVRT was 2.6 s, while grid’s current THD was determined to be 4.72%.

In Ref. [11], Zia et al. (2022) introduced an island-based direct current MG that combined Li-ion battery storages in addition to RESs like solar, wind, and tidal energies and was an excellent example of island tidal energy’s potentials. The first step executed the appropriate energy sharing schedule from each energy source, enhancing functional performance. In order to ensure optimal utilisation, the secondary phase assisted in revising the decision mechanisms through the review of aggregated scheduled production and demand profiles.

3. Proposed Methodology

In this work, an IDNN algorithm is proposed as a power management strategy. The proposed work includes an MG model, a grid PV-battery system, power and frequency deviations in the MG and load frequency control in the MG, an IDNN for hybrid wind/PV/battery-based MG environment and the evaluation of results.

3.1. Microgrid Model

Common standalone energy sources, storage devices, loads, power converters and control systems are all included in a MG [12]. In a typical standalone DC MG, the recommended secondary storages maintained the DC bus voltages within threshold limits and were their main backup power sources. Measurements (denoted by M) included sensors and filter circuits used in the generation of measurement signals [13]. Controllers utilised these signals to generate their signals. Figure 1 depicts hybrid energy systems based on PV/wind/batteries.

The overall system of the standalone DC MG functions towards sustaining power balances in systems. Conditions for net power deficits or availabilities in MG adheres to Equation (1)

$$\Delta i_{diff}=i_s-i_L$$

(1)

where $\Delta i_{diff}$ refers to systems’ net instantaneous current deficits or availabilities—$i_s$ represent sums of currents provided by all energy sources to DC buses—and $i_L$ represents the amount of currents drawn from DC buses by all MG loads. Equations (2) and (3) are used to compute currents and $i_L$ as follows:

$$\left\{i_s={\displaystyle \sum }_{k=1}^P{i}_{ESK}\right.$$

(2)

$$\left\{i_L={\displaystyle \sum }_{k=1}^Q{i}_{LK}\right.$$

(3)

where iESk represents currents provided by the kth energy sources to DC buses and k = 1, 2,…, P, and iLk represents currents taken from DC buses by kth loads and k = 1, 2,…, Q. When $\Delta i_{diff}$ is not positive, it implies a MG power deficiency where ESDs discharge power for balance [14]. A positive $\Delta i_{diff}$, on the other hand, shows the availability of extra power in an MG, which is utilised to charge ESDs.

3.2. Grid PV-Battery System

In this study, a robust RES for residential constructions with PV panels and battery banks (lead-acids) for storing energies was examined [15]. The system’s connection to electric networks allowed energy to travel across systems. PEMS (predictive EMS) controllers were used to link PV-producing systems to battery banks. It is possible to manually switch between utility grids and PEMS controllers, which are crucial safety features, using circuit breakers (safety devices). The highest level SOC for PV systems was also managed by this chip along with charging/discharging in battery banks.

Wind and battery mode: This mode was initiated when the point of common coupling (PCC) voltage surpassed the maximum threshold, and the wind converter reached its maximum reactive power potential [16]. To provide dynamic reactive power assistance, the battery voltage management mode was activated. To get the best performance in voltage regulation and active power control, the controllers’ parameters were optimized using a genetic algorithm.

Power and frequency deviations in MGs: In addition to an uncontrolled RES, additional sources such as DEGs (diesel engine generators), FCs (fuel cells), and aqua electrolysers can be totally controlled. Hydrogen is the fuel that powers FCs and is produced by AE through oscillations in wind energies. The DEGs and FCs were used as actuators in the MG vehicles. Control signals sent by the PI controllers regulated power injunctions between the diesel generators and fuel cells in order to maintain varying frequencies within the allowed bounds. The MG’s total power, expressed as Equation (4), included the primary sources of power, including winds/solar PVs, and secondary sources like fuel cells, diesel generators, and storage sources like FESS and BESS batteries. Excessive powers created by solar/winds were delivered to AE, which used it to produce the needed hydrogens for the FCG. By sucking power oscillations in the WTG and PVs, electrolysers separated water into hydrogen. Electrolysers then consumed power, as shown by the assumptions of negative signs. Power drawn from storage sources exhibits a dual nature, with these sources charging during regular scenarios to balance demands. Additionally, they discharge power to address sudden increases in load demands, supplying power to grids for brief periods. Following this, they recharge and return to the charging state. Therefore, they commonly serve as a load to the grid and consequently, the negative sign exists in the expression. At times, if they deliver power to the MG, a positive sign will be seen in the expression:

$$P_{total}=P_{WTG}+P_{PV}-P_{AE}+P_{FC}+P_{DEG}+P_{BESS}+P_{FESS}$$

(4)

Equation (5) defines net power fluctuations in sources and has to be zero for ensuring stability of MG power systems:

$$\Delta P_{total}=\Delta P_{WTG}+\Delta P_{PV}-\Delta P_{AE}+\Delta P_{FC}+\Delta P_{DEG}+\Delta P_{BESS}+\Delta P_{FESS}$$

(5)

Inputs to MG power systems are given by power differences ($\Delta P_{error}$) between variations in demands ($\Delta P_{Demand}$) and total powers generated ($\Delta P_{total}$). Equation (6) is helpful in calculating the power mismatch when put through a MG.

$$\Delta P_{error}=\Delta P_{Demand}-\Delta P_{total}$$

(6)

Equivalent inertia coefficient “M” and load damping constant “D”, with intrinsic temporal delays between frequencies and power differentials, can be utilised to model hybrid power systems. The delay is ignored in this case, resulting in the assumption of a linear approximation. As seen in Equation (6), input (i.e., power) errors and outputs (i.e., variations in frequencies) may be described as transfer functions. It connects the MG system’s fluctuating power to the frequency differential.

3.3. Load Frequency Control in MG

Using the first-order transfer function model to describe all the power sources allows the examining of tiny signal frequency shifts and their impacts on regulated sources. Estimated wind profiles were fed to wind turbine generators, which then produced the electrical power that the WTG offered. Similar to this, the solar PV model generated solar electricity using an assumed irradiance profile [17].

Once this has started in the MG, the distributed energy resources are used organically to achieve primary control, which has a direct link with system inertia [18]. Principal control actions are carried out promptly after disturbances to restrict high rates of oscillations in frequencies. Secondary controls, in addition to primary controls, are used to restore frequencies by regulating powers from extra sources. In order to evaluate the controller response, the MG is subjected to a step disruption to analyse the load frequency behaviour and the reactions of other regulated sources.

3.4. IDNN Approach for Hybrid Wind/PV/Battery-Based MG Environment

As shown in Figure 2, the described MPPT algorithm for WES was developed using DNN based controllers. A total of 66,000 data were used for training and refining the MPPT algorithm in this DNN learning method.

Simulation models of the suggested IDNN-based systems were built in MATLAB using 30 kW WES and boost converters for step-up voltages. The MPPT networks were trained using the DNN layers and inputs (wind voltage and current), and the outputs were the converter’s duty cycles.

IDNN controllers were trained using the aforementioned technique, and Figure 3 shows the validation performances. Figure 4 depicts the IDNN controller errors (target outputs), and the histogram data and construction of the IDNN-based MPPT algorithm is proved with the best possible regressions in the training, tests, validations and overall performances of data inputs. The regressions made use of multiple input factors, such as currents and wind voltages, to estimate the output variables (duty cycles). A 30 kW WES was utilised to simulate the planned and constructed IDNN-based MPPT algorithm. The suggested MPPT system was assessed taking into account various wind speeds.

As shown in Figure 3, a DNN controller was used for a wind energy system in accordance with the MPPT algorithm. A total of 66,000 data points were taken into consideration for training in this DNN learning algorithm in order to create the MPPT algorithm. It was undertaken to simulate the planned IDNN-based system. A 30 kW WES was used in the suggested model’s simulation, and a boost converter was also taken into account to step up the voltage. The IDNN controller’s flow chart is depicted. Using input data (wind voltages and currents), the DNN layers were used to train the MPPT networks, and the outputs obtained were the converter’s duty cycles.

Figure 4 displays the suggested IDNN controller errors (target outputs) and associated histogram data. Finally, the best regression for the suggested IDNN-based MPPT algorithm’s training, test, validation and overall performance data has been generated and presented. The regression makes predictions about an output variable (duty cycle) that depends on two input variables (current and wind voltage). The planned and created IDNN-based MPPT algorithm was applied and simulated on a 30 kW WES. Analyses of the MPPT systems were done in relation to different wind speeds.

4. Results and Discussion

The system in this study was simulated on MATLAB utilising various operating scenarios, and the results of the simulations were exhibited. The MG’s consumer load consumed around 220 kVR of reactive electricity, which depicts the total power generated by the hybrid PV/WES system. The investigation was conducted by largely keeping the voltage and current on the MG grid from fluctuating in terms of MG voltages and current waveforms in connection grid-integrated hybrid PV/WES nonlinear power sources. Although the hybrid PV/wind generators in this recommended concept produced more than 40 kW of electricity, the key challenge is regulating power at power-generation stations based on consumer demands. An essential goal of this study was to enhance power qualities and minimize THD values at PCC.

The MSE is used to estimate unobserved variables and statistically measure the averages of squares of errors or average squared differences between estimated and actual values, as shown in Equation (7).

$$MSE=1/n{\displaystyle \sum }_{i=1}^n{(Y_i-\widehat{Y_i})}^2$$

(7)

where $n$ stands for sample counts, $Y_i$ implies observed values and $\widehat{Y_i}$ signifies predicted values.

The below Figure 5 shows a comparison analysis between popular techniques and the proposed approach in terms of MSEs. Along the x-axis, the techniques are listed and along the y-axis, the MSE values are listed. The method including the DNN algorithm provides higher MSEs, whereas the proposed IDNN algorithm yields reduced MSEs for this particular setup. Hence, it can be concluded from the suggested IDNN results that it improves hybrid energy system performances.

The below Figure 6 depicts the comparison analysis carried out between the prevalent and proposed techniques in terms of THD. Along the x-axis, the techniques are plotted and along the y-axis, the THD values are listed. A method like the DNN algorithm provides higher THD values, whereas the proposed IDNN algorithm provides lower THD values for the given setup of the PV current system. It can be proven from the results that the proposed IDNN can improve the hybrid energy system’s performance.

Figure 7 demonstrates the comparison analysis between the available and proposed techniques in terms of THD values. Along the x-axis, the techniques are plotted, and along the y-axis, the THD value are listed. A prevalent method like the DNN algorithm provides higher THD values, whereas the proposed IDNN algorithm provides lower THD values for the given setup in terms of wind voltage. Hence it can be deduced from the result that the proposed IDNN can improve the hybrid energy system’s performance.

Execution time: The system is efficient if the execution of the proposed algorithm is finished in a reduced amount of time.

The below Figure 8 shows the comparison analysis between the popular techniques and proposed approach in terms of execution times. Techniques are on the x-axis, and execution times are on the y-axis. The existing DNN technique had increased execution time, while the proposed IDNN algorithm had reduced execution time for the provided PV/wind/battery-based setup. Therefore, it can be concluded from the result that the proposed IDNN can improve performance.

5. Conclusions

This study proposes a MG-based PV-wind hybrid energy system with an upgraded deep neural network-based voltage source controller. The IDNN method is used to create WES and hybrid PV/wind systems. The simulation results of the tests are performed in a range of weather circumstances and were used for evaluating the efficiency of the proposed controllers. Following modelling, the proposed WES and hybrid PV/wind system were built, and their integration with an MG was completed. In this work, simulation models for MGs based on smart inverters were built, and they were integrated with WES and hybrid PV/WES that evaluated the suggested IDNN algorithm in MATLAB simulations. The primary purpose of this technological endeavour was to visualise the THD values of winds, PV, the distributed grid voltages and current profiles. They all indicated that when the RES and MG integration is effective, power quality improves. The data illustrate the efficacy of the proposed technique, making it appealing for hybrid wind/PV and grid-connected RESs. Findings of the simulations imply that the suggested IDNN systems outperform traditional techniques in a range of operating situations, with lower MSE rates, lower THD values and lower computational complexity. The proposed IDNN system can be used in real time as a protective measure for microgrids.