Energy Reports 11 (2024) 1079–1087

Contents lists available at ScienceDirect

Energy Reports
journal homepage: www.elsevier.com/locate/egyr

Research paper

Power control of hybrid grid-connected renewable energy system using

machine learning
M. Karthikeyan a, *, D. Manimegalai b, karthikeyan RajaGopal c
a
Centre for Advanced Wireless Integrated Technology, Chennai Institute of Technology, Chennai, India
b
Department of Electrical and Electronics Engineering, Rajalakshmi Engineering College, Chennai, India
c
Centre for Nonlinear system, Chennai Institute of Technology, Chennai, India

A R T I C L E I N F O A B S T R A C T

Keywords: This article addresses the crucial challenge of maintaining a reliable power supply in integrated electric systems
Hybrid grid-connected system that combine solar power and energy storage. It focuses on optimizing key parameters for remote photovoltaic,
Machine learning wind, and battery energy storage within the grid, with a primary goal of sustaining 24-hour load demand. The
Battery storage
study introduces a comprehensive approach that integrates fuel price dynamics and battery depletion costs into
Nested learning algorithm
the optimization model for hybrid grids. This approach seeks a delicate balance between sustainable energy
usage and cost-effectiveness. To achieve this balance, the research employs genetic algorithms, a robust
computational technique. These algorithms play a vital role in calculating discharge-charge cycles and assessing
battery health, addressing concerns about battery longevity and performance. Additionally, the study explores a
machine learning framework, specifically nested learning, to further enhance the optimization process. Nested
learning allows for the development of a sophisticated target function that considers various grid operation
parameters and constraints. This machine learning approach enhances the system’s ability to dynamically
monitor power consumption and generation. Within this machine learning framework, the TD Lambda learning
algorithm takes a prominent role in identifying the optimal function. This algorithm is well-known for its effi­
ciency, particularly in non-Markovian scenarios. Its quick convergence enhances the overall performance of the
hybrid grid, ensuring efficiency in changing conditions. The research also evaluates the impact of weight factors
on various aspects, including battery charging status, algorithmic decisions, and their consequences on optimal
energy pricing within the hybrid grid. This comprehensive assessment deepens the understanding of how these
weight factors influence grid performance and economic viability.

that these integrated electric systems operate optimally, they must be

finely tuned to sustain the load demand for a full 24-hour cycle. This
1. Introduction critical requirement sets the stage for a multifaceted problem that re­
quires a holistic approach for resolution. One of the fundamental issues
In an era characterized by a growing emphasis on sustainable energy to be addressed is the optimization of key parameters governing these
sources and the relentless pursuit of efficient power supply solutions, hybrid grids. Achieving the delicate balance between energy generation,
integrated electric systems stand as a critical frontier in the quest for storage capacity, and load requirements is paramount. Moreover, the
reliable, eco-friendly electricity generation and distribution. These sys­ dynamic nature of fuel prices and the costs associated with battery
tems seamlessly combine solar power and energy storage, alongside depletion add layers of complexity to the optimization model. Inte­
other renewable sources like wind energy, to create a potential para­ grating these ever-changing factors effectively is an ongoing challenge.
digm shift in the way we meet our energy needs. However, within this Efficient battery management is another significant facet of this prob­
promising landscape, a series of complex challenges emerge, demanding lem. Energy storage is a linchpin in these systems, and ensuring the long-
innovative solutions. At the heart of this endeavor is the formidable term health and performance of batteries is essential. Calculating
challenge of maintaining a consistent and uninterrupted power supply. discharge-charge cycles, assessing battery health, and addressing con­
This is not a trivial task, especially when considering the inherent cerns regarding battery longevity are vital aspects that require
variability of renewable energy sources like solar and wind. To ensure

* Corresponding author.
E-mail address: [email protected] (M. Karthikeyan).

https://doi.org/10.1016/j.egyr.2023.12.060
Received 18 October 2023; Received in revised form 9 December 2023; Accepted 28 December 2023
Available online 5 January 2024
2352-4847/© 2024 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-
nc-nd/4.0/).
M. Karthikeyan et al. Energy Reports 11 (2024) 1079–1087

Nomenclature Hhub Height of the hub in meters.

α Exponent of power.
GPV Solar irradiation. PG Grid power.
IrrPV Solar irradiation received by the PV panel (W/m2). PL Load power.
AreaPV Area of the PV panels in m2. PB Battery power.
eff PV Overall efficiency of the PV panels. PR Battery rated power.
EPV Total energy output of the PV system. PBC Power consumed by the battery controller.
NPV Number of PV modules. PFDC Line frequency-dependent component for the battery.
EWES Total energy output from the wind energy system. PSOC Offset component for battery SOC compensation.
EWES(T) Total energy output of the wind energy system. SOCset value Desired SOC value.
NWT Number of wind turbines. G Gain used to control the speed of offset power
ηWES Efficiency of the wind energy system. compensation.
ρair Air density in kilograms per cubic meter (Kg/m3). ΔE Change in energy in the battery.
CP Coefficient of power for the wind energy system. ΔEfc Change in battery energy for frequency control.
A Rotor area. ΔESOC Change in battery energy for SOC compensation.
Ng Speed of the wind generator.
Ng(ref ) Reference speed for one hour.

meticulous attention. frequently utilized forms of renewable energy. However, due sources of
To meet these challenges head-on, emerging technologies in machine renewable energy are sporadic in nature, it is advised that they be
learning, such as nested learning, offer a promising avenue for supported by suitable storage units and best incorporated into the
enhancing the real-time monitoring of power consumption and gener­ hybrid system. Several studies were conducted for the best delivery of
ation within these grids. Developing sophisticated target functions that hybrid grid generation and storage devices in order to make the most of
consider various operational parameters and constraints can enable the battery storage in the hybrid system. Numerous cutting-edge opti­
better decision-making and resource allocation. Furthermore, as these mization techniques were consequently published. In the reference (Xu
integrated systems often operate in non-Markovian environments, et al., 2023), author provides some guidelines for incorporating the data
achieving efficient convergence in optimization algorithms becomes into nearly every aspect of the battery process for next-generation bat­
increasingly challenging. The introduction of advanced techniques like teries. Specifically, important issues related to the cooperation between
the TD Lambda learning algorithm offers potential solutions to this data-driven approaches and various theoretical strategies, experimental
issue, enhancing the overall performance and adaptability of these methods, models, characterization tools, and electrochemical perfor­
hybrid grids. mance tests in batteries are discussed.
Additionally, comprehensively evaluating the impact of weight fac­ Multi-target integral linear programming is used to solve a multi­
tors on various dimensions, such as battery charging status, algorithmic media problem involving a battery source and a hybrid energy source
decisions, and their consequences for energy pricing, is crucial. This (Yang et al., 2022). The amount that the battery is charged and dis­
evaluation seeks to provide insights into how these factors influence the charged is decided by fuzzy logic control (FLC). The target functions to
overall performance and economic viability of integrated electric sys­ be minimised are electricity costs, maintenance and capital costs, and
tems. In essence, this multifaceted problem statement encapsulates the fuel costs. A genetic algorithm is used to combine the local generation,
intricate challenges surrounding the optimization and efficient opera­ battery storage, and the power management system solution for
tion of integrated electric systems, aiming to ensure a reliable, cost- grid-connected hybrid systems (Madhuri et al., 2022). The amount of
effective, and sustainable power supply while navigating dynamic var­ grid electricity that will be introduced to the energy storage system and
iables and enhancing grid monitoring and adaptability. Solving this the amount that is modified on the primary grid will be determined by
problem holds the key to unlocking the full potential of renewable en­ the proposed power management system. The main goal is to increase
ergy integration in our quest for a cleaner and more resilient energy power grid trade profits as much as possible.
future. In (El-Bahay et al., 2023), author investigates the planning of grid
An electrical distribution system known as a hybrid grid enables the generation plants that utilize battery and hybrid power systems. To
integration of different domestically made sources, either in addition to address the power management problem within the system, a deep
the use of storage devices (Mastoi et al. 2023). Renewable energy Q-learning algorithm is employed, which operates on a stochastic
sources (RES), conventional generators, electricity expenses, storage framework. In (Kasis et al., 2021) a dynamic dispatch in a
areas, and hybrid grid technology can all be used to increase reliability battery-powered hybrid grid is discussed. The problem’s primary goal is
and reduce costs. It is possible for a hybrid grid to operate both con­ to maximise the battery’s operational benefit. A reinforcement study
nected and isolated. When connected to the primary grid in power mode and a Monte-Carlo tree search are suggested as a solution to the battery
through a familiar grid point, the hybrid grid can take part in electricity management problem.
markets as a buyer or seller via power system invoice or transmission The research conducted by authors (Schweiger et al., 2020), a fuzzy
systems. logic controller (FLC) was proposed to regulate the state of charge (SOC)
A power management system (PMS) is also required in order to make of the battery storage within the specified minimum and maximum
effective usage hybrid grid modules. PMS techniques can supervise limits. Furthermore, a backtracking search algorithm was incorporated
hybrid grid system voltage and frequency, as well as maximise the to optimize the battery SOC control, ensuring uninterrupted and effi­
transfer of energy production outputs from hybrid systems and cient operation of the battery system while minimizing overload and
achieving economic power demands. The most recent hybrid grid with maintaining capacity stability. In (Al-Hinai et al., 2021), an optimized
power management system architecture, various heat and power sour­ power dispatch algorithm was introduced for a grid-connected hybrid
ces, and electrical vehicles were described in (Hassan et al., 2022), along system comprising of PV units and battery deactivators. The objectives
with its main features and practical applications of optimization tech­ of the algorithm were to maximize the utilization of PV capacity,
nology. In hybrid grid-connected systems, wind and solar are the most minimize operational costs, and facilitate electricity exchange with the

1080
M. Karthikeyan et al. Energy Reports 11 (2024) 1079–1087

grid. Additionally, a comprehensive power management solution for a conceivable objective functions and apparatus; Lambda TD Application
full-day plan was tested, aiming to achieve an optimal balance between is concerned with the use of machine learning in power management;
energy generation, consumption, and grid interaction (Wang et al., The simulation and discussion are covered in Results and Discussions,
2020). and the article is concluded with a Conclusion.
The optimization challenge in hybrid grid generation units involves
modeling to maximize the utilization of renewable energy sources while 2. Modeling of hybrid grid-connected system
minimizing operating costs. To address this challenge, in (Aduda et al.,
2018) employed convex programming and a rolling horizon-predictive Modeling a hybrid grid comprising a photovoltaic (PV) system, wind
controller, coupled with a predictive model, to determine the optimal energy system, and battery storage using the TD Lambda algorithm in
battery storage settings. Additionally, the study utilized a cuckoo-search machine learning involves several key steps.
algorithm, a renewable energy system, and battery storage to focus on Firstly, it’s important to define the various elements within the sys­
power management in an isolated hybrid grid setting. The reduction of tem: the PV system, wind turbines, battery storage, and the grid itself.
overall costs and gas emissions is the problem feature that is being tar­ Each of these components interacts with one another and is influenced
geted. The utilization of a genetic algorithm to implement renewable by external factors like weather conditions and energy demand.
energy systems offers economic benefits (Vivian et al., 2022). The TD Lambda algorithm, a form of reinforcement learning, re­
The device under study comprises an isolated grid, battery, and quires the identification and representation of a state space, action
renewable energy sources. The objective of this study is to minimize space, and reward function. The state space comprises factors like solar
hybrid grid unit costs, fuel costs, and pollution costs. The grid-connected irradiance, wind speed, battery state of charge, grid demand, and more,
floral pollination algorithm (Lu et al., 2021) considers integrated capturing the current system conditions. The action space involves the
renewable energy sources, micro turbines, fuel cells, and battery storage possible decisions or actions that the system can take at each state, such
to develop a power management system solution. The aim is to minimize as adjusting energy generation or storage. The reward function is a
the costs associated with energy production, battery operation, energy critical aspect as it evaluates the system’s performance based on defined
exchange with the grid, and demand response payments. In reference objectives. For instance, the function might aim to reward the system for
(Bünning et al., 2022), a modified algorithm based on particle swarm maximizing the use of renewable energy sources, minimizing reliance on
optimization (PSO) was proposed to optimize the battery’s operating the grid, or reducing overall costs.
condition in a population hybrid grid. The algorithm also includes an Implementing the TD Lambda algorithm involves training a machine
efficient feature for battery charging and discharging. learning model using historical data or simulations. The model learns
In (Talaat et al., 2023), authors discusses the various optimization the optimal policy for decision-making by predicting future rewards and
techniques used to reduce the total cost of integrated energy sources. In iteratively updating value functions based on observed outcomes. This
addition, it examines the use of up-to-date methods to improve the learning process allows the system to improve its decision-making
performance of the electrical grid. A case study is conducted to analyze abilities over time.
the impact of using artificial intelligence when integrating RESs. The Validation and testing are crucial steps to ensure that the trained
results of the case study prove that the use of artificial intelligence helps model performs effectively in real-world scenarios. This validation
to improve the accuracy of operation to provide effective and accurate might involve testing the model with real-time data or through addi­
prediction control of the integrated system. A hybrid approach is pro­ tional simulations to verify its ability to make decisions that optimize the
posed in reference (Apparao and Chandra Sekhar, 2023) to achieve the hybrid grid’s operation.
load power requirement for grid connected hybrid photovoltaic wind Once validated, the learned policies from the TD Lambda algorithm
system. The proposed approach is the combined execution of both the can be integrated into the control system of the hybrid grid. This inte­
Modified Dragonfly Algorithm (MDA) and Adaptive Neuro-Fuzzy gration allows the system to autonomously manage and optimize energy
Interference System (ANFIS), hence it is called MDA-ANFIS. ANFIS generation, storage, and distribution in real-time, aiming to achieve the
approach is improved by the MDA approach to minimize the error defined objectives, such as maximizing renewable energy utilization or
functions. minimizing operational costs.
Most of the commercially available optimization programs adopt an Continuous monitoring and adjustments are essential as the system
algorithm that assumes repeated weather conditions, which is becoming operates, ensuring that the learned policies remain effective and
more unrealistic considering the recent erratic behavior of weather adaptable to changing environmental conditions and energy demands.
patterns. To address this issue, a data-driven framework is proposed in This process helps in maintaining the efficiency and stability of the
(Abdullah et al., 2023) that combine machine learning and hybrid hybrid grid system over time.
metaheuristics to predict weather patterns over the lifespan of a hybrid
renewable energy system in optimizing its size. The framework uses
machine learning tree ensemble methods such as the cat boost regressor, 2.1. PV system
light gradient boosting machine and extreme gradient boosting to pre­
dict the hourly solar radiation and load demand. In a simplified model, the performance of the photovoltaic module is
Most of these research studies overlook the practical objective of directly proportional to the solar irradiation and can be computed using
extending battery lifespan. It has been demonstrated that the charging Eq. (1):
and discharging processes have a significant influence on battery
GPV = IrrPV × AreaPV × eff PV (1)
durability. Lifecycle deficiency costs are one of the objective functions
that should be cut in this article. The machine learning algorithm is used Here, GPV represents the solar irradiation, IrrPV denotes the solar
to quantify the cost of battery deficiency and to calculate charging and irradiation received by the PV panel (W/m2), AreaPV is the area of the PV
discharging cycles. The efficiency of the Temporal difference (TD) panels in square meters, and eff PV represents the overall efficiency of the
lambda algorithms based on machine learning has been shown to be PV panels.
effective in resolving the unit engagement of a grid-connected hybrid To determine the total energy output for a given number of PV
grid. In this study, a 48-hour programming problem for remote rural modules, the following Eq. (2) can be used:
hybrid grid units is solved using the TD lambda algorithm.
EPV = GPV × NPV (2)
An overview of the article is given below: Discussed in modelling of
hybrid grid-connected system are hybrid grid devices. Power manage­ In Eq. (2), EPV represents the total energy output of the PV system,
ment system formulation addresses the proposed system, including and NPV denotes the number of PV modules.

1081
M. Karthikeyan et al. Energy Reports 11 (2024) 1079–1087

PB = PFDC + PSOC (7)

Here, PFDC represents the line frequency-dependent component for
the battery, while PSOC represents the offset component for battery SOC
compensation. Therefore, the Eq. (6) becomes
PG = PL + (PFDC + PSOC ) + PBC (8)
The offset power PSOC compensates for the energy wasted in battery
regulation and is calculated using Eq. (9):
PSOC = (SOCset value − SOC) × G × PR (9)
In Eq. (9), SOCset value refers to the desired SOC value, PR is the battery
rated power, and G is the gain used to control the speed of offset power
compensation.
Fig. 1 illustrates the relationship between regular battery life loss and
depth of discharge, showing that cycle loss increases as the cycle depth
decreases. In this study, we propose a technique for quantifying the
amount of battery energy lost during the power factor control process.
To calculate the change in energy, we utilize Eq. (10):
Fig. 1. Relation between batteries DOD with its life loss. ΔE = ΔEfc + ΔESOC (10)

2.2. Wind energy system In Eq. (10), ΔE represents the change in energy in the battery, ΔEfc
corresponds to the change in battery energy for frequency control, and Δ
The total energy output of a wind energy system can be determined ESOC denotes the change in battery energy for SOC compensation.
using Eq. (3):
3. Machine learning with TD Lambda algorithm
EWES = 0.5 × ηWES × ρair × CP × A × Ng 3 (3)
When modeling a hybrid grid consisting of a photovoltaic (PV) sys­
In Eq. (3), EWES represents the total energy output from the wind
tem, a wind energy system, and a battery storage system using the TD
energy system. ηWES is the efficiency generated by the wind energy
Lambda algorithm in machine learning, a Reinforcement Learning (RL)
system. ρair represents the air density in kilograms per cubic meter (Kg/
approach would be suitable. Reinforcement Learning involves training
m3). CP is the coefficient of power for the wind energy system. A rep­
an agent (in this case, the hybrid grid controller) to make sequences of
resents the rotor area, and Ng is the speed of the wind generator. The
decisions to achieve a long-term goal (minimizing costs, optimizing
wind speed for a given hour can be calculated using Eq. (4):
battery SoC, maximizing renewable energy utilization) in an uncertain
( )α
Ng Hhub environment.
= (4)
Ng(ref ) Href Within Reinforcement Learning, using the TD Lambda algorithm is
apt for this scenario. TD Lambda is a temporal-difference learning
In Eq. (4), Ng represents the speed of the wind generator, Ng(ref ) is the method that combines elements of both Monte Carlo methods and
reference speed for one hour, Hhub is the height of the hub in meters, and Temporal Difference methods. It allows the model to learn by predicting
α is the exponent of power. future rewards and iteratively updating value functions based on
The total energy output of the wind energy system, taking into ac­ observed outcomes.
count the number of wind turbines, can be calculated using Eq. (5): The steps to apply TD Lambda within a Reinforcement Learning
EWES(T) = EWES × NWT (5) framework for modeling the hybrid grid:

Here, EWES(T) represents the total energy output of the wind energy • Define State Space: Determine the variables and parameters that
system, and NWT is the number of wind turbines. describe the state of the hybrid grid system (solar irradiance, wind
speed, battery SoC, energy demand, etc.).
2.3. Battery storage system • Action Space: Establish the possible actions the agent can take in
response to each state (adjusting power output of PV and wind sys­
The grid power in the power system can be determined using Eq. (6): tems, controlling battery charging/discharging, supplying excess
PG = PL + PB + PBC (6) energy to the grid, etc.).
• Reward Function: Develop a reward function that quantifies the sys­
In Eq. (6), PG represents the grid power, PL is the load power, PB is tem’s performance based on the objectives (minimizing costs,
the battery power, and PBC is the power consumed by the battery maintaining adequate battery SoC, maximizing renewable energy
controller. utilization).
The battery storage system’s controller is responsible for power • Training the Agent: Train the agent (using the TD Lambda algorithm)
conversion and control. It can be assumed that it governs the power using historical data or simulations representing the behavior of the
flow. hybrid grid system. The agent learns by interacting with the envi­
Formulation of Power Management System: ronment and optimizing its actions to maximize cumulative rewards
In this study, load-side power factor control is primarily achieved over time.
through battery power control (PFC). Additionally, we aim to maintain • Validation and Testing: Validate the trained agent’s performance
the battery’s state of charge (SOC) at a specific level, such as 90 %. using real-world data or further simulations to ensure it can effec­
Consequently, the battery power PB is divided into two components: the tively optimize the hybrid grid’s operation while minimizing costs
line frequency-dependent component, denoted as PFDC , and the offset and managing battery SoC insufficiencies.
component for battery SOC compensation, denoted as PSOC . This divi­
sion is shown in Eq. (7):

1082
M. Karthikeyan et al. Energy Reports 11 (2024) 1079–1087

Fig. 2. Flow chart of proposed machine learning based TD Lambda algorithm.

Fig. 3. Load demand of a hybrid grid for one day.

storage system. This approach facilitates dynamic decision-making to

Table 1
achieve predefined goals while considering uncertainties in the envi­
Specifications of battery.
ronment and system complexities.
Parameters Specifications The TD Lambda algorithm is employed in this study to determine the
Energy Capacity (KWhr) 15 objective functions. Known for its high performance and convergence
Maximum Charging power (KWhr) 12 rate in non-Markovian settings, this algorithm incorporates the lambda
Maximum discharging power (KWhr) 12
parameter as the trace decay parameter, which ranges between 0 and 1.
SOC (Maximum) in % 80
SOC (Minimum) in % 15
In this context, SOC represents the state of charge for each state-action
Cost of Battery in Rs. 55,000/- pair (a, b). The charge level for state "a" is denoted as "c" and the
Maintenance cost in Rs. 18,000/- change in battery energy for frequency control is denoted as "d". For SOC
Rate of Interest (%) 6 compensation, the parameter "b" also represents a change in battery
Life span 15 years
energy, with its corresponding "E" representing the discharged battery
energy. The TD Lambda algorithm consists of the following steps:
• Integration and Deployment: Integrate the learned policies into the Step 1: Initially, an arbitrary value is assigned to SOC(a, b).
hybrid grid’s control system to enable autonomous decision-making Step 2: Based on the SOC(a, b) values, an action "b" is selected for
based on learned strategies in real-time. each step "t".
Step 3: The exploration-exploitation policy is employed to reduce the
The TD Lambda algorithm, within a Reinforcement Learning likelihood of getting trapped in an optimal solution. This means that the
framework, enables the hybrid grid controller to learn adaptive and action "b" is not selected solely based on the maximum SOC(a, b) for the
optimal policies for managing the PV system, wind system, and battery current state.

1083
M. Karthikeyan et al. Energy Reports 11 (2024) 1079–1087

Fig. 4. Effect of weight factor on SOC of battery with (a) epochs= 0, (b) epochs= 300 and (c) epochs= 1000.

1084
M. Karthikeyan et al. Energy Reports 11 (2024) 1079–1087

Table 2
Cost of batteries with different weight factor.
Cost Epochs= 0 Epochs= 200 Epochs= 300 Epochs= 500 Epochs= 700 Epochs= 900 Epochs= 1000

High (Lakhs Per day) 10.4 10.34 104 257 548 1028 752
Medium 215 488 507 738 602 438 338
(Lakhs Per day)
Low 90 195 233 176 267 252 225
(Lakhs Per day)

Table 3
Optimum Power generated by Battery for Epochs= 200.
Time in hour 0 2 4 6 8 10 12 14 16 18 20 22 24

Battery power in MW 0 125 250 312 2210 0.32 0.15 0.37 0.36 0.04 3.2 0.25 0.12

Table 4
Optimum powers generated by PV and Wind generator.
Time in hour 0 2 4 6 8 10 12 14 16 18 20 22 24

PV Power in MW 0 0 0 0 1.62 0.51 3.84 3.56 0.11 0.74 0 0 0

Wind generator power in W 0 7.43 4.71 5.22 13.84 16.91 294.02 747.89 702.93 548.40 277.64 359.41 372.66

Step 4: A new state is determined, and a reward is assigned based on 4. Results and discussions
the chosen action.
Step 5: The SOC(a, b) values are updated for various pairs of a and b To evaluate the proposed objective feature using machine learning, a
within SOC(a, b). SOC(a, b) represents the eligibility of each previously standalone hybrid grid was utilized, comprising PV panels, a wind
used state-action pair, which depends on the reward and state. generator, and a battery charger. The following variables were consid­
Step 6: The value of the lambda (λ), a new constant ranging from 0 to ered in this study: PPV (t), PWG (t). The simulation covered a 48-hour
1, is utilized. scheduling horizon, running for a period of two hours. The hourly
Step 7: When considering the eligibility of the state-action pair, the load demand for a typical day is illustrated in Fig. 3, with a consistent
SOC value for each state action is updated. load demand of 15 KW. Notably, between 18 and 22 h, there was an
Step 8: Consequently, all other pairs are disregarded, and only the increase in power demand on the load side. The wind generator, when
most recent state-action pair "PB " is recorded. subjected to a wind speed of 25 m/s, can generate a maximum power
In this study, the costs associated with insufficient batteries and the output of 80 KW. The wind generator’s parameters for swept area, air
fuel prices for conventional generators are taken into consideration. To density, and coefficient efficiency were measured as 20, 1.5, and 0.35,
address issues in the power management system, a proposed TD Lambda respectively. Additionally, the solar photovoltaic system employed in
algorithm is employed. The objective function is optimized using this this study has a capacity to produce 20 kW of power under rated envi­
machine learning-based algorithm, specifically tailored to determine the ronmental conditions.
battery depth of discharge. Fig. 2 depicts the proposed TD Lambda al­ In this study, the battery constraints are defined as shown in Table 1,
gorithm for resolving the power management system problem. assuming an initial SOC battery of 50 %. By utilizing a population size of
During abnormal environmental conditions, optimizing the objective 100 and conducting a total of 200 iterations, the TD Lambda algorithm
function in a hybrid grid system, which comprises a photovoltaic (PV) successfully resolves the energy distribution problem. To evaluate the
system, wind energy system, and battery storage, becomes pivotal for effectiveness of the proposed objective feature, the assessment includes
sustained efficiency and stability. A dynamic objective function must better costs, worse cost, and mean cost, employing various weighting
adapt to changing circumstances. For instance, in situations like severe criteria.
weather affecting solar irradiance or wind patterns, the objective func­ The battery’s state of charge (SOC) serves as a limit in addressing the
tion could shift priorities, perhaps emphasizing grid stability or mini­ economic dispatch issue to prevent overcharging and over discharging.
mizing reliance on external sources rather than solely maximizing Fig. 4 illustrates the influence of the weight factor on the battery’s SOC
renewable energy utilization. Real-time adaptation mechanisms based behavior. The objective feature assesses weight factor ranges of 0, 300,
on environmental sensor data could trigger shifts in the objective and 1000. The presence of numerous charge and discharge cycles,
function when abnormalities arise, redirecting system goals to maintain clearly visible when the battery is not fully discharged, indicates a sig­
stability or mitigate disruptions caused by environmental variations. nificant contribution of the battery to the hybrid grid’s energy supply.
Hierarchical objectives or weighted functions may dynamically adjust However, by avoiding additional charging cycles and aiming for battery
priorities, focusing on critical aspects such as maintaining battery state life extension, the two objectives intersect. This graph demonstrates that
of charge within safety margins or ensuring essential loads are powered load durations are reduced, maintaining a more stable SOC condition.
during irregular conditions, even at increased costs. Incorporating ma­ Moreover, hybrid grids require fewer battery replacements, resulting in
chine learning algorithms enables the system to learn from and adapt to higher cost functions compared to those in Table 2.
abnormal events, continuously refining the objective function for To maintain the trade-offs between the two objectives, a simulation
improved responses. This adaptive approach ensures the hybrid grid of the ED problem for HG units is conducted with Epochs= 200. The
system can respond effectively and efficiently during abnormal envi­ optimal power exchange between the battery and the hybrid grid, ach­
ronmental scenarios, maintaining reliability and performance. ieved through the TD Lambda algorithm, is presented in Table 3.
Additionally, Table 4 showcases the top power generation outcomes
obtained using the TD Lambda algorithm. As the PV and wind generators
initiate electricity production, they contribute renewable energy,

1085
M. Karthikeyan et al. Energy Reports 11 (2024) 1079–1087

Fig. 5. Economic Dispatch for (a) PV power in MW (b) Wind Power in W and (c) Battery Power in MW.

1086
M. Karthikeyan et al. Energy Reports 11 (2024) 1079–1087

thereby supporting power generation stability. Acknowledgements

Based on the information provided in Tables 3 and 4, Fig. 5 illustrates
the optimal solution for economical dispatch in hybrid grid devices M. karthikeyan and karthikeyan Rajagopal are gratefully acknowl­
throughout a 24-hour period. Both hybrid grid units actively contribute edge this work is funded by the Centre for Advanced Wireless Integrated
to fulfilling the load demand. As the load demand decreases, the battery Technology, Chennai Institute of Technology (CIT), India, vide funding
charging is minimized, and other generators are also scaled down until number CIT/CAWIT/2023/RP-010.
the PV panels and wind generator commence electricity generation. As
mentioned in the article, the battery serves the dual purpose of sup­ References
plying power to the hybrid grid while being charged. In cases where
renewable energy sources provide insufficient energy, the battery can Abdullah, H.M., Park, S., Seong, K., Lee, S., 2023. Hybrid renewable energy system
design: a machine learning approach for optimal sizing with net-metering costs.
compensate for the shortfall. During night time when renewable energy Sustainability 15, 8538.
supply is limited and load energy demands are higher, the battery col­ Aduda, K.O., Labeodan, T., Zeiler, W., 2018. Towards critical performance
laborates to meet the load demand. considerations for using office buildings as a power flexibility resource-a survey.
Energy Build. 159, 164–178.
Al-Hinai, Amer, Alyammahi, Hajer, Haes Alhelou, Hassan, 2021. Coordinated intelligent
5. Conclusion frequency control incorporating battery energy storage system, minimum variable
contribution of demand response, and variable load damping coefficient in isolated
power systems. Energy Rep. 7, 8030–8041.
This study provides valuable insights into optimizing energy trans­ Apparao, A., Chandra Sekhar, G., 2023. Hybrid MDA-ANFIS approach based control of
mission within isolated hybrid grids using a machine learning-based TD grid connected solar system with nine level inverter wind energy conversion. Int.
Lambda algorithm. The results underscore the importance of consid­ Rev. Appl. Sci. Eng. 14 (2), 183–200.
Bünning, Felix, Warrington, Joseph, Heer, Philipp, Smith, Roy S., Lygeros, John, 2022.
ering both fuel costs and battery depletion expenses in the objective
Robust MPC with data-driven demand forecasting for frequency regulation with heat
function for effective energy supply management. The identified sig­ pumps. Control Eng. Pract. 122, 105101.
nificance of selecting an appropriate weight factor to achieve an optimal El-Bahay, M.H., Lotfy, M.E., El-Hameed, M.A., 2023. Computational methods to mitigate
cost function offers practical guidance for real-world implementations of the effect of high penetration of renewable energy sources on power system
frequency regulation: a comprehensive review. Arch. Comput. Methods Eng. 30,
similar hybrid grid systems. The study’s findings suggest that a weight 703–726.
factor of 200 not only minimizes overall costs but also contributes to Hassan, C.Au, Iqbal, J., Ayub, N., Hussain, S., Alroobaea, R., Ullah, S.S., 2022. Smart grid
enhancing the durability of the battery’s state of charge (SOC) while energy optimization and scheduling appliances priority for residential buildings
through meta-heuristic hybrid approaches. Energies 15, 1752.
reducing the frequency of charging cycles required. Kasis, Andreas, Monshizadeh, Nima, Lestas, Ioannis, 2021. Primary frequency regulation
Moreover, the research sheds light on addressing the trade-off be­ in power grids with on–off loads: Chattering, limit cycles and convergence to
tween battery deficiency and fuel costs, revealing the feasibility of optimality. Automatica 131, 109736.
Lu, Xi, Xia, Shiwei, Sun, Guangzeng, Hu, Junjie, Zou, Weiwei, Zhou, Quan,
solving the economic dispatch problem within the hybrid grid gener­ Shahidehpour, Mohammad, Chan, Ka. Wing, 2021. Hierarchical distributed control
ating units through 200 epochs. The successful utilization of machine approach for multiple on-site DERs coordinated operation in microgrid. Int. J. Electr.
learning techniques to combine various units within the hybrid grid Power Energy Syst. 129, 106864.
Madhuri, N.S., Shailaja, K., Saha, Debasmita, P, Revathy, Glory, K.B., Sumithra, M.,
demonstrates its efficacy in efficiently fulfilling load dispatch re­ 2022. IOT integrated smart grid management system for effective energy
quirements. These findings have broader implications for the design and management. Meas.: Sens. 24, 100488.
operation of hybrid energy systems, emphasizing the need for a Mastoi, Muhammad Shahid, Zhuang, Shengxian, Munir, Hafiz Mudassir, Haris, Malik,
Hassan, Mannan, Alqarni, Mohammed, Alamri, Basem, 2023. A study of charging-
comprehensive approach that considers various cost factors and opti­
dispatch strategies and vehicle-to-grid technologies for electric vehicles in
mizes energy dispatch strategies. distribution networks. Energy Rep. 9, 1777–1806.
The study’s outcomes offer valuable guidance for energy policy­ Schweiger, Gerald, Eckerstorfer, Lisa V., Hafner, Irene, Fleischhacker, Andreas,
makers, grid operators, and stakeholders involved in designing and Radl, Johannes, Glock, Barbara, Wastian, Matthias, Rößler, Matthias, Lettner, Georg,
Popper, Niki, Corcoran, Katja, 2020. Active consumer participation in smart energy
managing isolated hybrid grids. Implementing similar approaches and systems. Energy Build. 227, 110359.
weight factor considerations could lead to improved cost-effectiveness, Talaat, M., Elkholy, M.H., Alblawi, A., 2023. Artificial intelligence applications for
enhanced battery performance, and optimized energy utilization in microgrids integration and management of hybrid renewable energy sources. Artif.
Intell. Rev. 56, 10557–10611.
various hybrid grid configurations. Additionally, these findings pave the Vivian, Jacopo, Croci, Lorenzo, Zarrella, Angelo, 2022. Experimental tests on the
way for further research avenues, encouraging exploration into refining performance of an economic model predictive control system in a lightweight
optimization techniques and machine learning algorithms for broader building. Appl. Therm. Eng. 213, 118693.
Wang, Huilong, Wang, Shengwei, Shan, Kui, 2020. Experimental study on the dynamics,
applications in renewable energy integration and grid management, quality and impacts of using variable-speed pumps in buildings for frequency
thereby contributing to sustainable and efficient energy systems. regulation of smart power grids. Energy 199, 117406.
Xu, Liqianyun, Wu, Feng, Chen, Renjie, Li, Li, 2023. Data-driven-aided strategies in
battery lifecycle management: prediction, monitoring, and optimization. Energy
Declaration of Competing Interest Storage Mater. 59, 102785.
Yang, Yulong, Yan, Gangui, Mu, Gang, 2022. Bi-level decentralized control of electric
The authors declare that they have no known competing financial heating loads considering wind power accommodation in real-time electricity
market. Int. J. Electr. Power Energy Syst. 135, 107536.
interests or personal relationships that could have appeared to influence
the work reported in this paper.

Data Availability

No data was used for the research described in the article.

1087