A CNNbased energy management strategy for a Hybrid energy storage system in electric vehicles

Abstract

When approaching a large-scale performance, the choice, size, and administration of an Energy Storage System (ESS) for an Electric Vehicle (EV) are crucial. As the peak-to-average power demand ratio is relatively large, particularly for an urban ride that is frequently marked by rapid deceleration as well as acceleration, the complementary characteristics of the battery and Ultra-Capacitor (UC)render this arrangement a viable Hybrid energy storage system (HESS) for EV. Enhanced dynamic responsiveness, increased miles per charge, and extended battery life provided by the HESS increase the Electric Vehicle (EV’s) effectiveness. The primary objective of the study that has been suggested is to create a smart Energy Management Strategy (EMS) for EVs. A battery package and a properly sized Ultra-Capacitor (UC) together give the required high power along with energy density. This research proposes a CNN-based power management technique to aid in efficient EMS. Additionally, by adjusting the Convolution Neural Network (CNN) classifier’s weights through Improved Honey Badger Optimization (IHBO), the adaptive approach of the Standard HBO Algorithm, the performance of the classifier is improved. In MATLAB, the suggested CNN-based model for BESS EM is simulated and experimental assessment and analysis are done in terms of converter current and battery SoC. By contrasting the suggested approach against several standardized models, the performance analysis of the proposed work is assessed to validate its performance.

Keywords

Electric vehicle energy management convolutional neural network IHBO battery UC

1. Introduction

Electric Vehicles (EVs) are increasingly being studied and developed around the globe to preserve energy, lowering consumption, and resolve the issue of polluting the surroundings. Nevertheless due to their fundamental chemical properties, batteries possess little power density, which will certainly impact the EV function. A supplementary ESS must be taken into consideration for EVs to address the frequent power transient demands, and also to satisfy the automobile’s peak power needs as it may compensate for the battery’s limitations. Expanding the range of driving an EV by employing a Hybrid energy storage system (HESS) is a good idea, however managing numerous energy storage devices requires a complex Power Management System (PMS) A Hybrid EV (HEV) with an Internal Combustion Engine (ICE) and a HESS was explored in [1]. A couple of novel energy management techniques, that consider the shelf life of the battery, have been suggested for the HESS that links the batteries with the super-capacitor, which includes a Model Predictive Control (MPC) system as well as a Dynamic Programming (DP) technique.

The Equivalent Consumption Minimization Strategy (ECMS) was utilized between ICE and HESS. The equivalent HEV arrangement is the study that has also been utilized in [2]. A DP methodology was recommended for the EM of the HESS, which enhances the lifespan of the batteries, and a fuzzy controller was utilized for the EM among the ICE and the HESS. An FCHV that consists of an FCS and a HESS was the subject of an earlier study [3]. To lessen the use of hydrogen, a multi-dimensional DP coding was investigated the EM among the FCS/BESS/SC. For a Hybrid EV (HEV) to attain the best fuel efficiency and shelf life of batteries, earlier studies [4] applied the PMP to the energy management that includes the ICE, the battery, as well as the super-capacitor. To increase the range along with the efficiency of the EV, a rule-based energy management technique for the HESS was suggested in [5]. However, a DP technique for the HESS was published in [6] to minimize the loss of the battery in EVs. In EVs constructed with HESSs, rule-based energy management algorithms are extensively implemented [7, 8]. Although reliable and computationally effective, these methods do not ensure energy efficiency.

Developing the appropriate standards also requires technical knowledge as well as manual customization. From the solutions obtained through alternative optimization approaches, such as DP and genetic algorithm-based optimization [9, 10, 11], several governing laws can be derived. To manage the energy consumption of hybrid as well as long-range EVs, fuzzy logic controls are also used [12, 13]. For optimal results in onboard electricity management, fuzzy logic controllers can either supplement or work in conjunction with various additional instruments [14]. To increase efficiency and dependability during energy storage and discharge cycles, make sure that various energy storage technologies such as ultracapacitors and lithium-ion batteries integrate and synchronize seamlessly. To preserve performance and extend the life of batteries and other components, efficient thermal management systems must be put in place inside the energy storage system. To reduce the dangers of energy storage systems, such as overcharging, short circuits, and thermal runaway events, strengthen safety procedures and dependability measures. Maximizing the potential of hybrid energy storage systems in electric vehicles can result in enhanced performance, efficiency, and user satisfaction by tackling these obstacles with creative technologies, strong management approaches, and industry-wide cooperation.

The following is an outline of this study:

–
Establishes an adaptive model of Standard Algorithm Honey Badger Optimization (HBO) named Improved HBO Algorithm (IHBO) entrenched on Convolutional Neural Network (CNN) for energy storage systems in Electric EVs.
–
Introduces a novel optimization methodology, named IHBO for the enhancement of CNN performance by tuning the weights of the CNN model.

The organization of this research is provided below: Section 2 describes the remarks about traditional methods along with the features as well as challenges of various HESS for the Power management of EVs, Section 3 narrates the proposed Energy Management strategy in EVs with HESS, Section 4 explains the process of Tuning of weights by Improved Honey Badger Optimization Algorithm whereas Section 5 provides the results and discussion and Section 6 summaries the research.
2. Literature survey

In 2020, Bindu and Sushil [15] offered a fuzzy logic control-based power management technique that employs Ultra-Capacitor (UC) to relieve the battery of peak discharge currents and quick charging currents. By controlling the battery current’s size as well as the rate of change, this suggested Work seeks to prolong the longevity of the battery. By keeping the SoC of the UC between the bounds of 0.85 to 0.96, the Ultra-Capacitor (UC) can be rendered for sharing the peak load. This is accomplished by charging the UC at a time of low load and discharging it during a time of high load, as determined by the Fuzzy Logic Controller (FLC) As the reference battery current set by the FLC is gradually variable and the UC, which is voltage regulated, will distribute the leftover power, this also aids in balancing the battery current. Simulations were performed and the outcomes were examined. Various operational methods are investigated using a prototype experimental configuration.

In 2021, Prasanthi et al. [16] developed an ideal hybrid energy source sizing technique for Hybrid EVs (HEVs) that combines UC and Fuel Cells (FC) with battery Units (BU). Dynamic-source frameworks are used to design a multifaceted issue that evaluates the system’s establishing expenses, size, operating expense, as well as supply deterioration costs. Additionally, a brand-new Adaptive EMS (AEMS) that emphasizes driving cycle power requirement and dynamic-source features is put forth as a solution to the optimization issue. Finally, the BOA is enhanced by using the quantum waveform notion for better exploring the search space for solving the hybrid energy source optimization issue. The effectiveness of the developed technique is assessed in Matlab/simulation, and the findings indicate that the suggested AEMS operates more effectively and might lower the system’s relative expense and weight for the Battery unit ultra-capacitor Fuel Cell (BU-UC-FC) configuration by 16% and 10%, respectively.

In 2020, Ye et al. [17] suggested an Energy Management Strategy (EMS) along with a hybrid energy storage system optimization approach for the energy management as well as control strategy of an EV for conventional driving cycles. The Fuel Cell System (FCS) suggests optimized methods, such as Genetic Algorithm (GA) and Particle Swarm Optimization (PSO) that employ consumption of energy as their ideal goal activity. Two control techniques are discovered after analyzing the fuzzy control strategy in MATLAB/Simulation depending on the enhanced EV concept. The ideal control method was discovered and contrasted with the simulation results relying on the Urban Dynamometer Driving Schedule (UDDS) as well as the New European Driving Cycle (NEDC). In numerous simulated scenarios, this study demonstrates that, in comparison to the PSO, the output peak current and current variation of the BESS enhanced by the GA algorithm are less, and are more stable, as well as the overall consumption of energy is decreased by 3–9%.

In 2020, Bai et al. [18] presented a multi-dimensional size optimization framework and HEMS to maximize the element size as well as power of a Plug-in Hybrid Electric Vehicle (PHEV) with the HESS. A PHEV with a BESS is utilized as a comparative guidance, and the DP technique is established as a standard for opposition, to assess how well size optimization and power optimization execute. The size optimization approach investigates the ideal configuration of the system, by considering the maximal power and capacity of the battery and the SC as well as the system’s maximum power. The battery degrading rate was decreased by 48.9%, as well as the automobile economy has risen by 21.2%, according to the results of size optimization and HEMS.

Table 1
Features and challenges of various HESS for power management of EVs.

Author [citation] Adopted methodology Features Challenges

Bindu and Sushil [15] FLC based PMS
–
Increased effectiveness.
–
Enhances the voltage level.

–
Because of a cell balancing difficulty, the battery as well as UC voltage is constrained.

Prasanthi et al. [16] BOA strategy
–
The expense is less by sixteen percent.
–
Superior function

–
The formulation of the optimization issue has mainly disregarded the source dynamic properties.

Ye et al. [17] Fuzzy control $+$ PSO $+$ GA
–
Rapid optimization velocity.
–
Improved performance

–
Real-time optimization cannot be done using this time-consuming methodology.

Bai et al. [18] DP algorithm
–
Battery deterioration costs are decreased by 38.4 percent.
–
Greater fuel efficiency.

–
It can be challenging to maximize the battery’s longevity.

Yi et al. [19] PMP
–
Safeguard the battery.
–
Accumulate energy.

–
EV energy management is a common example of an ideal regulation issue in a system with nonlinear dynamics that is time-varying and subject to limitations.

Komarappa and Lakshmi Kantha [20] ANN controller-based HESS
–
Batteries SOC decreases under 60%.
–
Increased efficiency

–
Aids in resolving issues brought on only by a single ESS.

Sanusi et al. [21] Reinforcement learning and adaptive DP
–
Boost system efficiency.
–
Lowest possible fuel usage.

–
DP’s rely on precise mechanism models.

Gonsrang and Kasper [22] CQP based PMS
–
Requires optimum power.
–
The time required for computation is less.

–
Ignores the temperature dependence of the variables for simplicity, which is one of the main drawbacks.

Author [citation]	Adopted methodology	Features	Challenges
Bindu and Sushil [15]	FLC based PMS	– Increased effectiveness. – Enhances the voltage level.	– Because of a cell balancing difficulty, the battery as well as UC voltage is constrained.
Prasanthi et al. [16]	BOA strategy	– The expense is less by sixteen percent. – Superior function	– The formulation of the optimization issue has mainly disregarded the source dynamic properties.
Ye et al. [17]	Fuzzy control $+$ PSO $+$ GA	– Rapid optimization velocity. – Improved performance	– Real-time optimization cannot be done using this time-consuming methodology.
Bai et al. [18]	DP algorithm	– Battery deterioration costs are decreased by 38.4 percent. – Greater fuel efficiency.	– It can be challenging to maximize the battery’s longevity.
Yi et al. [19]	PMP	– Safeguard the battery. – Accumulate energy.	– EV energy management is a common example of an ideal regulation issue in a system with nonlinear dynamics that is time-varying and subject to limitations.
Komarappa and Lakshmi Kantha [20]	ANN controller-based HESS	– Batteries SOC decreases under 60%. – Increased efficiency	– Aids in resolving issues brought on only by a single ESS.
Sanusi et al. [21]	Reinforcement learning and adaptive DP	– Boost system efficiency. – Lowest possible fuel usage.	– DP’s rely on precise mechanism models.
Gonsrang and Kasper [22]	CQP based PMS	– Requires optimum power. – The time required for computation is less.	– Ignores the temperature dependence of the variables for simplicity, which is one of the main drawbacks.

In 2021, Yi et al. [19] created an EMS for HEVs that considers battery degradation and relied on PMP. The hybrid energy storage EV model is initially built to validate the EMS. To create a battery deterioration framework, the battery life cycle tests have been performed. After that, an RB control strategy and Pontryagin’s Minimum Principle (PMP) optimization technique are employed to rationally distribute electricity in a HESS. The created EMS is verified by a simulation experiment with UDDS settings, and the results show that the recommended EMS offers a lesser rate of consumption of energy along with battery degradation than the RB technique.

In 2018, Komarappa and Lakshmi Kantha [20] presented an Artificial Neural Network (ANN) controller-based HESS architecture that is constructed by employing both NiMH battery and ultra-capacitor in Hybrid EV (HEV). The HEV had two sources, such as the battery and the UC. The battery was the main supply in that situation, and UC was a redundant supply. In this case, the HEV load demand is satisfied using the UC. The UC’s capabilities are charged and discharged by the ANN controller. Once the battery’s SOC falls below 60%, the UC starts to discharge. To determine when the UC should be turned on or off, the battery SOC was tracked throughout each iteration. Based on this, each iteration of the HEV received the necessary power. The conclusion drawn from the results is that the HEV with the ANN controller performs better than the HEV without the ANN controller.

In 2019, Sanusi et al. [21] developed an online learning system that utilizes reinforcement learning along with adaptive dynamic programming for the power control of hybrid electric systems. The approaches used for managing electricity currently are conventional and unable to properly account for system fluctuations brought on by shifting functional as well as health situations. These conservative plans make less efficient use of the power sources that are already accessible, raising the entire system cost and escalating the likelihood of failure because of variances. As the reinforcement signals are nonlinear, a Neural Network (NN) is used to execute the system. When contrasted to a traditional offline dynamic programming strategy, simulation findings reveal enhanced system performance utilizing the suggested approach.

In 2018, Gonsrang and Kasper [22] presented a Power Management System (PMS) for an EV equipped with a battery pack, UC bank, and range extender. Without breaking any physical restrictions, the suggested Constrained Quadratic Program (CQP) – based PMS distributed the best power loads to the automobile’s elements. Weight considerations were effective instruments for directing the optimization program to accomplish specific objectives, such as Super Capacitor (SC) bank charging solely with regenerative power. Similar to the Nonlinear Model Predictive Control algorithm (NMPC)-based program, the CQP may accomplish the highest possible energy usage. Integer parameters as well as quadratic conditions took the CQP solver a fair amount of time to resolve. The quadratic restriction can be linearized, which will speed up calculation and make it easier to use this program for online applications. Table 1 provides the features and challenges of various HESS for the power management of EVs. The FLC-based PMS approach is introduced in [15], which enhances the voltage level with increased effectiveness. However, the batteries as well as UC voltages are constrained because of cell balancing difficulty. ABOA strategy was introduced in [16], which decreased the expense by sixteen percent with superior functionality. Here, the formulation of the optimization issue has mainly disregarded the source dynamic properties. Moreover, Fuzzy control with PSO and GA was designed in [17], which provides improved performance with rapid optimization velocity. However, Real-time optimizing cannot be done using the time-consuming methodology. A DP algorithm has been introduced in [18], which offers greater fuel efficiency in addition to a decrease in battery degradation by 38.4 percent. Yet, it can be challenging to maximize the battery’s longevity. PMP was described in [19], which was used to accumulate energy as well as safeguard the battery. However, EV energy management is a common example of an ideal regulation issue in a system with nonlinear dynamics that is time-varying and subject to limitations. Nevertheless, ANN controller-based HESS was introduced in [20], which decreases the battery SoC by under 60% and offers increased efficiency. Here, it aids in resolving issues brought on by a single ESS. Reinforcement learning and adaptive dynamic programming were utilized in [21], which boosts system efficiency with the lowest possible fuel usage. However, DP’s rely on a precise mechanism model. CQP-based PMS was developed in [22] that requires only optimum power and the time required for computation is less. However, it ignores the temperature dependence of the variables for simplicity.

3. Energy management strategy in electric vehicles with hybrid energy storage system

The HESS of the proposed model consists of an induction motor, a battery, two DC-DC converters, a UC, and a three-phase inverter.

3.1. Battery

The equivalent circuit of a battery may possibly be contemplated as a basic electrical design of a battery cell, where the electromotive force $ε$ is connected in series with the internal resistance $R_{B}$ of the battery. In fact, by adopting Kirchhoff’s law, the resultant formulation is obtained.

\begin{aligned} U_{B} = N_{S} \times V_{S} = N_{S} \times R_{B} \times I_{B} \end{aligned}

(1)

Here, $V_{S}$ indicates the voltage of the source, $N_{S}$ indicates the cell numeral. The capacity of the battery $C_{B}$ is known through Eq. (2)

\begin{aligned} C_{B} = C_{n_{10}} \frac{1.67 \times (1 + 0.005 \times t_{a c})}{1 + 0.67 \times {(\frac{I_{n}}{I_{n_{10}}})}^{0.9}} \end{aligned}

(2)

Such that, $t_{a c}$ denotes the accumulator temperature, $C_{n_{10}}$ denotes the minimal capacity of a nominal current $I_{n_{10}}$ . Additionally, SoC is stated as in Eq. (3)

\begin{aligned} S o C = 1 - \frac{I_{B} \times T}{C_{B}} \end{aligned}

(3)

Whereas, $T$ refers to the discharge time at $I_{B}$ current. Alternatively, the charge $(V_{q})$ along with discharge $(V_{Q})$ voltages are stated by Eqs (4) and (5) correspondingly [23].

\begin{aligned} V_{q} & = N_{S} \times (2 + 0.16 \times S o C) + N_{S} \times \frac{I_{B}}{C_{n_{10}}} (\frac{6}{1 + I_{B}^{1.3}} + \frac{0.27}{S o C^{1.5}} + 0.002) \times (1 - 0.007 \times t_{a c}) \end{aligned}

(4)

\begin{aligned} V_{q} & = N_{S} \times (2 + 0.16 \times S o C) - N_{S} \times \frac{I_{B}}{C_{n_{10}}} (\frac{6}{1 + I_{B}^{1.3}} + \frac{0.27}{S o C^{1.5}} + 0.002) \times (1 - 0.007 \times t_{a c}) \end{aligned}

(5)

The SOC of a battery is the most important component of its energy status. The latter mostly depends on how the battery is charged and discharged. When contrasted with the maximal overall battery energy, the SoC is the energy that is quite obtainable. We thus have:

\begin{aligned} \frac{d e_{B}}{d T} = P_{B} \end{aligned}

(6)

where,

e_{B}

specifies the energy of the battery through the energy power

P_{B}

. In the present scenario, the amount of power on the DC bus affects how the energy in the system varies. It is zero, either positive or negative, indicating that the battery is not working, depending on whether it is charged or discharged. The variance in battery energy is then calculated using Eq. (7). The system uses Eq. (8) to express the variance in battery energy while considering the battery’s two charging and discharging states.

\begin{aligned} e_{B} (T) & = e_{B} (T - 1) + Δ e_{B} \end{aligned}

(7)

\begin{aligned} Δ_{e} B & = {\begin{matrix} P_{b} η_{q} η_{d c - d c} & i f P_{b} > 0 \\ \frac{P_{b}}{η_{d c - d c} η_{Q}} & i f P_{b} < 0 \end{matrix} \end{aligned}

(8)

Taking into consideration that $Δ T$ is equivalent to 1 hour, the deviation of energy is the power existing on the bus. The energy deposited in the battery is comparatively smaller than the energy available in the bus, due to the losses caused by battery charging $P_{b} (1 - η_{q})$ and by DC-DC conversion because of its efficiency $η_{d c - d c}$ . Or else, the battery must aid the discharging losses and those acquired by dc-dc converter at the same time as sustaining the demand in the DC bus. i.e. $P_{b} < 0$ [24].

3.2. Ultracapacitors

Positive and negative charges are actually separated by the UC to store energy. Two parallel plates separated by an insulator are used for storing the charges. UCs have an extended cycle life yet possess a low energy density as they do not undergo chemical modification. Like batteries, UC units in vehicles are made up of several cells connected in series and perhaps even in parallel. For ease of building the UC pack for the automobile, several cells are typically integrated into units. However, the functionality of the UC unit is primarily determined by the cell properties. Resistance R (Ohm) and capacitance C (Farad) are the two main performance factors for UCs [25]. Figure 1 depicts the simple UC model.

Figure 1.

Simple UC model.

The UC terminal voltage is exactly proportional to the SOC, which is another aspect of the device. UCs can be used to help HEVs with their energy storage needs. Urban driving frequently involves stop-and-go circumstances and the overall power needed is little. Due to its high charge and discharge rates, UCs are ideal for absorbing energy from regenerative braking and swiftly supplying power for acceleration.

Usually, UCs have a larger power density than BESS. Cost savings come from a long lifespan and minimal maintenance. Batteries and UCs could be coupled in HEV applications to maximize the advantages of both components [26].

3.3. DC-DC converter

UCs have a substantially higher density power than batteries, which have a reasonably high energy density. The UCs should be directly linked to the DC link to insulate the batteries from supplying peak power and from being directly recharged, which will increase battery life. The HESS’s bidirectional DC-DC converter comprises buck and buck-boost modes, and the efficiency of conversion in each mode is consistently regarded as the most crucial element. To increase the effectiveness of the entire system, the UCs and batteries should deliver the power directly without the use of a DC-DC converter, as energy loss cannot be completely avoided in either buck mode or buck-boost mode [27].

The goal of the DC-to-DC power converter is to create a bidirectional power flow between two numerous voltage levels under both normal and abnormal conditions. This is possible if the DC-DC converter has an appropriate topology. To regulate energy between the UC and the battery, the control laws for DC/DC converters must be established. The two converters with parallel topologies used in this study are the buck and buck-boost converters. The DC-DC converter connected to the battery is referred to as the buck converter, while the DC-DC converter coupled to the UC is the buck-boost converter. The HEV energy management establishes the converter dynamic control strategy [28, 29].

3.4. Three-phase inverter

Each switching component of a simple three-phase bridge inverter is made up of a controlled rectifier and a diode coupled in a pair back-to-back. A switch can be used to symbolize this pair. Emergency supplies and AC motor drives typically employ these inverters. For loads with different power factors, these applications call for control of the output voltage and output frequency [30]. The DC link voltage and pulse width modulation can be adjusted to control the three-phase inverters’ output voltage. For many applications, the pulse width modulation is appealing because it doesn’t call for a second stage of the regulated DC link.

3.5. Proposed methodology

The ultimate aim of the suggested research is to develop an intelligent EMS for EVs. The battery has been included in this research for the persistence of alleviating the power density shortage of ESSs in EVs. A battery, two DC-DC converters, a UC, a three-phase inverter, and an induction motor make up the HESS. The battery is considered to be the primary energy source in charge of sustaining the vehicle’s range. The power distribution among batteries in both charged and discharged modes is managed by the bidirectional DC-DC converter. The proposed method’s schematic diagram is shown in Fig. 2.

Figure 2.

Proposed model of Energy Management System for BESS using EVs.

In this proposed work, the converter current, battery SOC, and UC SOC are given as input to the CNN [31], which predicts the power to be delivered by the battery and the UC. Additionally, by fine-tuning the CNN classifier’s weights using the IHBO, the performance of the classifier is improved. Battery SoC and UC SoC are provided as the input to the CNN classifier from the Simulink model by modifying the acceleration pattern and the accompanying converter current, which results in the achievement of battery power as well as UC power. The IHBO fine-tunes the weights of the CNN for maximum performance. The IHBO is the Standard HBO Algorithm’s adaptive model [32]. The honey badger’s sophisticated foraging technique served as the inspiration for HBA, a mathematical technique for creating an effective search method for resolving optimization issues. The HBA’s explored and exploited stages are derived from the honey badger’s dynamic searching behavior, which employs tactics like digging and honey searching. Also, by employing regulated randomization techniques, HBA keeps an adequate population range even after the searching stage.

3.5.1. Data normalization

Data normalization, which is the conversion of characteristics into a common array, is a crucial pre-processing step that prohibits significant numerical aspect values from predominating small numerical feature values. The fundamental objective is to reduce the amplification of those sequence category characteristics that quantitatively provide beyond options. The importance of data normalization for developing accurate predictive models has been looked into by CNN. Using the min-max [0,1] normalization method [33], the data was normalized and is provided by,

\begin{aligned} z^{'} = \frac{z - z_{min}}{z_{max} - z_{min}} \end{aligned}

(9)

where,

z^{'}

is the normalized signal and

z

is a set of signals for all charging cycles.

3.5.2. CNN architecture

CNN comprises multiple convolutions as well as sub-sampling layers. Each of the layers involves numerous feature maps that are linked to preceding layer maps using a set of minute receptive fields. The CNN architecture is shown in Fig. 3.

Figure 3.

CNN architecture.

Several feature maps are created from the given data as needed. The location of the appropriate receptive field reflects the feature occurrence on the feature map, defining the features map. The subsampling layer is applied after every convolution layer. By halving each map dimension four times, reduces data for the next analysis. The convolution layer is placed after each subsampling layer. In this instance, the connectivity matrix is defined by links between layers. This procedure is repeated until the feature maps are too minute. Following these feature maps, there is a fully interconnected MLP, the output of which is a vector of classifiers [34]. Acceleration, battery current, battery power, convergence, UC current, UC power, and UC SoC are the input characteristics that are used to estimate the efficiency of EMS in the automobile. The outputs of the CNN are battery power and UC power. The weights of the CNN strategy are tweaked ideally utilizing the established optimization method to contribute to better EM. The obtained input parameters for acceleration to time are shown in Fig. 4 as Battery power, Battery current, convergence, UC current, UC power, and UC SoC.

Figure 4.

Collected input parameters.

4. Tuning of weights by improved honey badger optimization algorithm

The recommended IHBO methodology involves the optimal tuning of the weights of the CNN model.

4.1. Objective function ( $O b j_{f n}$ ) and problem formulation

The suggested alternatives in the EMS aim to lower the CNN classifier’s intended output, which includes predictions about the vehicle’s speed and driving style. Thus, Eq. (10) provides the perceived $O b j_{f n}$ of the generated approach.

\begin{aligned} O b j_{f n} = E r_{min} \end{aligned}

(10)

4.2. Proposed IHBA algorithm

The HBA mimics the honey badger’s foraging style, and its underlying bionic concept is as follows: During the global exploring phase, the honey badger utilizes the mice’s smell abilities to find the hive’s location and, once there, it chooses the best spot to collect honey. Honeyguide birds can locate honeycombs and are expected to collect them [35]. Honey badgers are led by honeyguide birds to directly identify honeycombs and subsequently gather honey throughout the local development stage. To provide better convergence capability with a maximal convergence rate, HBA enhancement is also necessary.

The prey’s scent intensity $I_{s c}$ handled by the honey badger at its existing position will certainly influence its progress speed. The scent intensity equation is given below:

\begin{aligned} I_{s c} & = a_{1} \times \frac{I_{c o n}}{4 π D_{s c}} \end{aligned}

(11)

\begin{aligned} I_{c o n} & = (w_{s c} - w_{s c + 1})^{2} \end{aligned}

(12)

\begin{aligned} D_{s c} & = w_{p r e y} - w_{s c} \end{aligned}

(13)

Where, $I_{c o n}$ indicates concentrated intensity, found by the honey badger’s current spot $w_{s c}$ as well as the next position $w_{s c + 1}$ , $D_{s c}$ specifies the gap between the prey's spot $w_{p r e y}$ and the present honey badger spot $w_{s c}$ , $a_{1}$ is an arbitrary limit within [0, 1]. The diminishing factor $β$ safeguards a smooth transition of HBA from the global exploring phase to local expansion and is mathematically represented as:

\begin{aligned} β = X \times \exp ({\frac{- h}{h}}_{M A X}) \end{aligned}

(14)

Here, $h$ denotes the current iteration number, $h_{M A X}$ denotes the maximal iterated count, also $X$ be a constant that is greater than or equal to 1. Throughout the HBA’s global exploring stage, the populace searches the searching space using the trajectory movement of the heart-shaped line. However, in the initial stages of iteration, the heart-shaped line’s traversal range is inhibited and its ability for global exploration is insufficient. The spiral search mechanism is presented in this study as a means of avoiding the aforementioned condition. The population will adjust its position based on the spiral’s trajectory. The updated position is as follows:

\begin{aligned} w_{n e w} = w_{p r e y} + f_{i n} * α * I * w_{p r e y} + f_{i n} * α_{3} * β * D_{s c} * δ \end{aligned}

(15)

Figure 5.

Flowchart of proposed IHBO model.

Here, $w_{n e w}$ indicates the new honey badger position, $w_{p r e y}$ indicates the prey’s spot and the global ideal position found so far, $α ⩾ 1$ is the capacity of the honey badger to obtain money, $I_{c o n}$ and $D_{s c}$ are estimated by Eqs (11) and (12), which $β$ is found by Eq. (13) and

\begin{aligned} δ = \cos (2 π k) * l * \exp (k * m) \end{aligned}

(16)

Here,

k

is an arbitrary number between [

-

1, 1],

a_{3}

is an arbitrary number between (0,1),

l =

1, and

m =

1.2. The Interference flag

f_{i n}

offers a possibility of altering the honey badger’s search orientation, providing the user with additional opportunities to carefully review the entire search field, which is provided as,

\begin{aligned} f_{i n} = {\begin{matrix} 1 & i f α_{5} \leq 0.5 \\ - 1 & e l s e \end{matrix} \end{aligned}

(17)

During the local development stage, the Schaffer function can be used to model the population movement trajectory and hence this adaptive process is referred to as the IHBO algorithm.

\begin{aligned} w_{n e w} = w_{p r e y} + f_{i n} * a_{6} * β * D_{s c} * f_{i n_{1}} \end{aligned}

(18)

Where, $a_{6}$ is an arbitrary number and

\begin{aligned} f_{i n_{1}} & = 0.5 - (\frac{k_{1}}{k_{2}}) \end{aligned}

\begin{aligned} k_{1} & = S i n^{2} \sqrt{w_{1}^{2} + w_{2}^{2}} - 0.5 \end{aligned}

where,

k_{2} = [1 + 0.001 (w_{1}^{2} + w_{2}^{2})]^{2}

such that

w_{1} = 0

and

w_{2} =

0.1001. The Pseudocode of the proposed IHBO model is given in Algorithm 1, and the flow chart is given in Fig. 5.

Table 2

Convergence analysis for learning rate of 60%.

Iterations	PSO	HBA	PROP
1	0.042768	0.04269	0.042732
2	0.042685	0.042636	0.042707
3	0.042685	0.042636	0.042707
4	0.042685	0.042636	0.04269
5	0.042685	0.042636	0.04262
6	0.042657	0.042636	0.04262
7	0.042657	0.042636	0.04262
8	0.042657	0.042636	0.04262
9	0.042657	0.042636	0.04262
10	0.042657	0.042636	0.04262
11	0.042657	0.042636	0.04262
12	0.042657	0.042636	0.04262
13	0.042657	0.042636	0.04262
14	0.042657	0.042636	0.04262
15	0.042657	0.042636	0.04262
16	0.042643	0.042636	0.04262
17	0.042643	0.042636	0.04262
18	0.042643	0.042636	0.04262
19	0.042643	0.042636	0.04262
20	0.042643	0.042636	0.04262

Table 3

Convergence analysis for learning rate of 70%.

Iterations	PSO	HBA	PROP
1	0.043919	0.043938	0.043901
2	0.043919	0.043894	0.043901
3	0.043919	0.043889	0.043901
4	0.043919	0.043889	0.043901
5	0.043919	0.043889	0.043901
6	0.043919	0.043889	0.043901
7	0.043919	0.043889	0.043901
8	0.04391	0.043889	0.043901
9	0.04391	0.043872	0.043901
10	0.04391	0.043872	0.043901
11	0.04391	0.043872	0.043901
12	0.04391	0.043872	0.043861
13	0.04391	0.043872	0.043861
14	0.04391	0.04386	0.043861
15	0.04391	0.04386	0.043861
16	0.04391	0.04386	0.043852
17	0.04391	0.04386	0.043852
18	0.04391	0.04386	0.043852
19	0.04391	0.04386	0.043852
20	0.04391	0.04386	0.043852

Table 4

Convergence analysis for learning rate of 80%.

Iterations	PSO	HBA	PROP
1	0.044169	0.044142	0.044161
2	0.044169	0.044142	0.044159
3	0.044164	0.044137	0.044103
4	0.044115	0.044125	0.044103
5	0.044115	0.044125	0.044103
6	0.044115	0.044125	0.044103
7	0.044115	0.044125	0.044103
8	0.044115	0.044125	0.044103
9	0.044115	0.044125	0.044103
10	0.044115	0.044125	0.044103
11	0.044115	0.044125	0.044103
12	0.044115	0.044125	0.044103
13	0.044115	0.044125	0.044103
14	0.044115	0.044125	0.044103
15	0.044115	0.044106	0.044088
16	0.044115	0.044106	0.044088
17	0.044115	0.044106	0.044088
18	0.044115	0.044106	0.044088
19	0.044115	0.044106	0.044088
20	0.044115	0.044106	0.044088

Table 5

Error Measures of created models compared to typical techniques.

Condition	Metrics	NN	CNN	PSO $+$ CNN	HBA $+$ CNN	PROP $+$ CNN
Lp $=$ 60	MAE	0.9811	0.1417	0.1416	0.14092	0.1313
	MSE	0.9411	0.0491	0.0490	0.04905	0.0391
	RMSE	0.9701	0.2217	0.2214	0.22148	0.2117
	RMSPE	3579.3	862.47	108.52	111.530	107.28
	MAPE	3579.3	196.46	67.755	69.516	66.733
	MARE	1.8681	1.9646	0.6775	0.67516	0.6573
	MSRE	3.4899	74.873	1.1877	1.17790	1.1509
	RMSRE	1.8681	8.7377	1.1852	1.08530	1.0628
Lp $=$ 70	MAE	0.7823	0.1405	0.1404	0.14053	0.1210
	MSE	0.6120	0.0482	0.0486	0.04837	0.0313
	RMSE	0.7823	0.2196	0.2196	0.21994	0.2100
	RMSPE	190.52	773.54	107.89	108.51	101.21
	MAPE	190.52	190.64	66.697	67.3260	62.943
	MARE	0.9052	1.9964	0.6669	0.67326	0.6294
	MSRE	2.8194	59.397	1.1857	1.17750	1.1268
	RMSRE	1.9052	7.6254	1.0889	1.08510	1.0121
Lp $=$ 80	MAE	0.6523	0.1421	0.1390	0.14010	0.1116
	MSE	0.6003	0.0464	0.0483	0.04665	0.0295
	RMSE	0.9701	0.2023	0.2192	0.20283	0.1926
	RMSPE	186.35	378.44	105.49	106.27	100.24
	MAPE	186.35	180.45	65.939	65.5710	60.028
	MARE	0.7930	1.8045	0.6613	0.66571	0.5802
	MSRE	1.1255	14.321	1.1770	1.13810	1.1198
	RMSRE	1.7285	3.7844	1.0849	1.01270	1.0024

5. Results and discussion

5.1. Simulation setup

The established CNN technique with IHBO methodology-entrenched EMS in EVs was implemented using MATLAB 2021b for experimental assessment of the proposed method as well as the comparative analysis. By varying the Acceleration pattern of the vehicle the corresponding converter current, battery soc, ultracapacitor soc, battery power, and uc power values are collected. The simulation graphs are obtained for acceleration, battery power, battery current, convergence, UC current, UC power, and UC SoC, respectively. In this instance, the efficacy of existing methodologies, such as NN, CNN, PSO $+$ CNN, and HBA $+$ CNN was assessed. The function of the suggested method is compared to that of traditional procedures by employing error metrics to demonstrate its efficacy in effective EM.

5.2. Convergence analysis

Figure 6.

Convergence analysis over different learning rates.

The proposed approach and traditional models in terms of cost function are contrasted in Fig. 6. Different iterations, including 5, 10, 15, and 20, are taken in this particular instance. As shown by the convergence study’s results, which are presented below in the order of learning rates of 60, 70, and 80 using the appropriate Tables 2, 3, and 4, the suggested IHBO scheme dependably provided less cost values when contrasted with the traditional schemes.

Figure 7.

Error indices of created approach versus conventional strategy.

The suggested system achieves a lower cost function of 0.04262 in the tenth iteration with a learning rate of 60%, which is less than the values acquired by well-known approaches, like PSO and HBA, the values of which are 0.042657 and 0.042636, respectively. The implemented IHBO also produced the lowest cost function of 0.043901 on 70% of the learning percentage, which is lower than the values of the PSO and HBA algorithms in the tenth iteration, which are 0.04391 and 0.043872, respectively. The proposed technique’s cost function is 0.044103 in 80% learning rates throughout the same 10th iteration as that of previously described learning rates, which is lower than suggested approaches, like PSO and HBA by values of 0.044115 and 0.044125, correspondingly. As a result, the suggested IHBO technique provides lower cost measures than the conventional methodologies under these various learning rate factors. Thus, the suggested approach can be altered to be capable of satisfying the simulation analysis’s demand for a low-cost function.

5.3. Error measure analysis

Figure 7 shows the error assessment of the established classification over the conventional system. The suggested IHBO model with the CNN classifier yields the lowest MAE error value. Assuming a 60% learning rate, the generated model achieves an MAE error value of 0.1313, which is lesser than that of the conventional approaches like NN (0.98113), CNN (0.14174), PSO $+$ CNN (0.14161), and HBA $+$ CNN (0.14092), and reflects the suggested method’s best prediction. According to MSE, the adopted model achieves 0.03915, while the other methods achieved, respectively, 0.94117 for NN, 0.049152 for CNN, 0.049024 for PSO $+$ CNN, and 0.049052 for HBA $+$ CNN. The established IHBO $+$ CNN approach’s error metrics such as “RMSE, RMSPE, MAPE, MARE, MSRE, and RMSRE” possess minimal values of 0.21, 107.28, 66.733, 0.65733, 1.1509, and 1.0628, respectively. Additionally, the applied IHBO $+$ CNN model achieved a lesser MSE value of 0.031399 on observing 70% of the learning percentage, which is lower than NN, CNN, PSO $+$ CNN, and HBA $+$ CNN schemes. The executed scheme achieved MAE of 0.78232, RMSE of 0.78232, RMSPE of 190.52, MAPE of 190.52, MARE of 0.90522, MSRE of 2.8194, and RMSRE of 1.9052, constantly. Additionally, for 80% learning, the RMSE values for NN, CNN, PSO $+$ CNN, and HBA $+$ CNN are 0.97014, 0.2023, and 0.20283 respectively, while the recommended technique has attained a lower RMSE value of 0.19263, which is slower than that of the existing models. This method, which concentrated on the RMSPE and RMSRE measures, achieved the lowest possible error rates for all learning percentages. To achieve lower error rates, the study demonstrated the superiority of the developed IHBO $+$ CNN model. The outcomes of this experiment show how much more effective the adopted technique is at spotting battery issues With LP of 60, 70, and 80, Table 5 shows the entire error indices of the proposed strategy in comparison with the prevailing strategy.

6. Conclusion

Based on CNN and the IHBO Technique, an EMS approach for EVs has been established and executed in this research. The recommended configuration relies on managing the energy between the battery, three-phase inverter, DC-DC converter, and ultra-capacitor as needed. The CNN model employing the UC, battery, and EV characteristics supports the EM, and such information serves as the input to the suggested CNN classifier, which is involved in the predicted EV speed. The weights of the CNN approach are set optimally utilizing the proposed IHBO approach to contribute effective EM. The error measure is used as the basis for the performance analysis. The simulation graphs for acceleration, battery current, power, convergence, UC current, power, and UC SoC are taken separately. When compared to established methods, the proposed method achieves the minimum value in the simulation of error metrics. This offers much-needed energy storage to support the electrification of society, the switch to renewable energy sources, and energy security. The suggested method obtained an RMSPE of 190.52, MAPE of 190.52, MARE of 0.90522, MSRE of 2.8194, RMSRE of 1.9052, MAE of 0.78232, and RMSE of 0.78232. The examination of system efficiency in high-voltage circumstances will be the main focus of this paper’s future research. But to keep the advantages of the suggested system intact and reduce total system costs, it is necessary to weigh the DC/DC converter’s size against the choice of UC. To confirm that the suggested HESS is electrically viable, a scaled-down experimental setup was constructed. The topology is electrically viable, and the hysteresis control is a straightforward but efficient control method, as demonstrated by the experimental findings. The suggested IHBO technique will be implemented in a closed-loop real-time vehicle control mode prototype as a future development of this research.

Footnotes

Abbreviation

References

Santucci

Sorniotti

Lekakou

. Power split strategies for hybrid energy storage systems for vehicular applications. Journal of Power Sources. 2014; 258: 395-407.

Masih-Tehrani

Ha’iri-Yazdi

Esfahanian

Safaei

. Optimum sizing and optimum energy management of a hybrid energy storage system for lithium battery life improvement. Journal of Power Sources. 2013; 244: 2-10.

Ansarey

Panahi

Ziarati

Mahjoob

. Optimal energy management in a dual-storage fuel-cell hybrid vehicle using multi-dimensional dynamic programming. Journal of Power Sources. 2014; 250: 359-71.

Vinot

Trigui

. Optimal energy management of HEVs with hybrid storage system. Energy Conversion and Management. 2013; 76: 437-52.

Armenta

Núñez

Visairo

Lázaro

. An advanced energy management system for controlling the ultracapacitor discharge and improving the electric vehicle range. Journal of Power Sources. 2015; 284: 452-8.

Song

Hofmann

Han

Ouyang

. Optimization for a hybrid energy storage system in electric vehicles using dynamic programing approach. Applied Energy. 2015; 139: 151-62.

Sandoval

Alvarado

Carmona

Lopez

Gomez-Aguilar

. Energy management control strategy to improve the FC/SC dynamic behavior on hybrid electric vehicles: A frequency based distribution. Renewable Energy. 2017; 105: 407-18.

Hwang

Chen

Kuo

. The study on the power management system in a fuel cell hybrid vehicle. International Journal of Hydrogen Energy. 2012; 37(5): 4476-89.

Chen

Tsai

. Design and analysis of power management strategy for range extended electric vehicle using dynamic programming. Applied Energy. 2014; 113: 1764-74.

10.

Zhang

Xiong

. Adaptive energy management of a plug-in hybrid electric vehicle based on driving pattern recognition and dynamic programming. Applied Energy. 2015; 155: 68-78.

11.

Sorrentino

Rizzo

Arsie

. Analysis of a rule-based control strategy for on-board energy management of series hybrid vehicles. Control Engineering Practice. 2011; 19(12): 1433-41.

12.

Chen

Liu

Huang

. Energy management strategy for fuel cell/battery/ultracapacitor hybrid vehicle based on fuzzy logic. International Journal of Electrical Power & Energy Systems. 2012; 43(1): 514-25.

13.

Chen

Guan

. Adaptive power management control of range extended electric vehicle. Energy Procedia. 2014; 61: 67-70.

14.

Chen

Xiong

Wang

Cao

. An on-line predictive energy management strategy for plug-in hybrid electric vehicles to counter the uncertain prediction of the driving cycle. Applied Energy. 2017; 185: 1663-72.

15.

Bindu

Thale

. Power management strategy for an electric vehicle driven by hybrid energy storage system. IETE Journal of Research. 2022; 68(4): 2801-11.

16.

Prasanthi

Shareef

Asna

Ibrahim

Errouissi

. Optimization of hybrid energy systems and adaptive energy management for hybrid electric vehicles. Energy Conversion and Management. 2021; 243: 114357.

17.

. Optimization of hybrid energy storage system control strategy for pure electric vehicle based on typical driving cycle. Mathematical Problems in Engineering. 2020; 2020(1): 1365195.

18.

Bai

Dos Santos

Yang

. Optimal design of a hybrid energy storage system in a plug-in hybrid electric vehicle for battery lifetime improvement. IEEE Access. 2020; 8: 142148-58.

19.

Wang

Pan

Tao

Zhou

Zhao

. Energy management strategy for hybrid energy storage electric vehicles based on pontryagin’s minimum principle considering battery degradation. Sustainability. 2022; 14(3): 1214.

20.

Komarappa

LakshmiKantha

. Power management strategy for hybrid energy storage system in hybrid electric vehicles. System. 2018; 13.

21.

Sanusi

Mills

Konstantopoulos

Dodd

. Power management optimisation for hybrid electric systems using reinforcement learning and adaptive dynamic programming. In 2019 American Control Conference (ACC) 2019 (pp. 2608-2613). IEEE.

22.

Gonsrang

Kasper

. Optimisation-based power management system for an electric vehicle with a hybrid energy storage system. International Journal of Automotive and Mechanical Engineering. 2018; 15(4): 5729-47.

23.

Mokrani

Rekioua

. Modeling, control and power management of hybrid photovoltaic fuel cells with battery bank supplying electric vehicle. International Journal of Hydrogen Energy. 2014; 39(27): 15178-87.

24.

Chakir

Tabaa

Moutaouakkil

Medromi

Julien-Salame

Dandache

Alami

. Optimal energy management for a grid connected PV-battery system. Energy Reports. 2020; 6: 218-31.

25.

Burke

. Ultracapacitor technologies and application in hybrid and electric vehicles. International Journal of Energy Research. 2010; 34(2): 133-51.

26.

Khaligh

. Battery, ultracapacitor, fuel cell, and hybrid energy storage systems for electric, hybrid electric, fuel cell, and plug-in hybrid electric vehicles: State of the art. IEEE transactions on Vehicular Technology. 2010; 59(6): 2806-14.

27.

Wang

Cao

Zhou

. A novel multimode hybrid energy storage system and its energy management strategy for electric vehicles. Journal of Power Sources. 2015; 281: 432-43.

28.

Pramanik

Pati

. Modelling and control of a non-isolated half-bridge bidirectional DC-DC converter with an energy management topology applicable with EV/HEV. Journal of King Saud University-Engineering Sciences. 2023; 35(2): 116-22.

29.

Camara

Gualous

Gustin

Berthon

Dakyo

. DC/DC converter design for supercapacitor and battery power management in hybrid vehicle applications – Polynomial control strategy. IEEE Transactions on Industrial Electronics. 2009; 57(2): 587-97.

30.

Jardan

Dewan

Slemon

. General analysis of three-phase inverters. IEEE Transactions on Industry and General Applications. 1969; (6): 672-9.

31.

Monikandan

Chellaswamy

Geetha

Sivaraju

. Optimized Convolutional Neural Network-Based Capacity Expansion Framework for Electric Vehicle Charging Station. International Transactions on Electrical Energy Systems. 2022; 2022(1): 2915910.

32.

Hashim

Houssein

Hussain

Mabrouk

Al-Atabany

. Honey Badger Algorithm: New metaheuristic algorithm for solving optimization problems. Mathematics and Computers in Simulation. 2022; 192: 84-110.

33.

Singh

. Investigating the impact of data normalization on classification performance. Applied Soft Computing. 2020; 97: 105524.

34.

Levinskis

. Convolutional neural network feature reduction using wavelet transform. ElektronikairElektrotechnika. 2013; 19(3): 61-4.

35.

Lei

Yang

Jiao

. Solar photovoltaic cell parameter identification based on improved honey badger algorithm. Sustainability. 2022; 14(14): 8897.

A CNNbased energy management strategy for a Hybrid energy storage system in electric vehicles

Abstract

Keywords

1. Introduction

3.1. Battery

3.4. Three-phase inverter

3.5. Proposed methodology

4.1. Objective function ( O b j f n ) and problem formulation

5.1. Simulation setup

5.2. Convergence analysis

6. Conclusion

Footnotes

Abbreviation

References

4.1. Objective function ( $O b j_{f n}$ ) and problem formulation