A Dynamic Mechanism Based on the Neural Network and Fuzzy Logic for Accurate Speed Control for an Electric Vehicle

Earlier research in Speed control in EV shows it mainly focus on basic control strategies such as PID control which regulates speed on predefined setpoints and feedback signals. The methods discussed here provide a foundation for subsequent fuzzy logic control and are emerging as a promising approach for EV speed regulation due to its ability to handle non-linearities and uncertainties in the control system. The adaptive and robust speed control mechanism can be achieved when fuzzy logic, which uses linguistic variables and fuzzy rules to model human-like decision-making processes are integrated.

The literature related to EV speed control can be organized into three major methodological categories: conventional PID-based control, fuzzy and fuzzy-PID control, and neural network/neuro-fuzzy-based control.

2.1 Conventional PID-based EV speed control

Conventional PID and PI-based control methods provide the foundation for speed regulation in EVs and electric motor drives. These approaches use feedback information and controller parameters to regulate the desired speed. However, the performance of conventional controllers can be affected by non-linearities, changing loads and variations in motor parameters.

Wang and Zhao [1] integrating a conventional PID control which uses fuzzy logic and addressing non-linear, time-varying challenges and response lag. While it does not explicitly mention neuro-fuzzy methods, this integration of fuzzy logic enhances the PID controller's adaptability and robustness, improves stability and accuracy in speed control. Thus, represents an advanced approach within the broader category of speed control management methods.

The limitation of this approach in the context of the present study is that the method integrates fuzzy logic with PID control, but it does not explicitly investigate neural-network-based adaptation of the fuzzy control parameters. Therefore, the learning capability of a neuro-fuzzy controller remains outside the scope of this work.

Xu et al. [2] explored the application of adaptive fuzzy PID controllers in pure EVs, it showcases excellent adaptability to varying speed requirements.

Although this approach demonstrates adaptive speed control for pure EVs, the controller is based on an adaptive fuzzy-PID structure. The material reviewed here does not indicate the use of neural learning to optimize the fuzzy parameters, which leaves scope for investigating a neuro-fuzzy approach.

Ghafarian and Yazdi [3] used fuzzy PID controllers for optimization of EV speed control, and they had got a great improvement in the energy efficiency and responsiveness of the system.

The limitation relevant to the present study is that the controller is based on fuzzy-PID optimization and does not establish how neural learning can be used to adapt the fuzzy parameters under changing EV operating conditions.

2.2 Fuzzy logic and fuzzy-PID speed control

Fuzzy logic control provides an extension to conventional control methods by using linguistic variables and fuzzy rules. The approach is useful when the mathematical relationship between system variables is difficult to represent precisely and when non-linearities and uncertainties are present.

Zadeh [4] laid the foundation for fuzzy logic systems. It introduced a framework that handles uncertainty and imprecision in control systems.

The work provides the theoretical foundation for fuzzy logic but is not specifically developed for EV speed control. Therefore, the direct application of the framework to EV propulsion-speed regulation requires additional controller design and validation.

Lee [5] investigated the application of fuzzy logic in control systems and provides knowledge about the robustness of fuzzy logic systems in the presence of system disturbances.

The limitation in relation to the present research is that the work establishes the general principles of fuzzy control rather than providing a neuro-fuzzy EV speed-control mechanism. It therefore does not address neural adaptation of fuzzy parameters for EV operating conditions.

Wang [6] gave a comprehensive course on fuzzy systems and control, focusing on the flexibility of fuzzy systems in dynamic environments.

This work provides a general foundation for fuzzy systems and control; however, it does not specifically address the integration of neural learning with fuzzy control for EV speed regulation.

Sivanandam et al. [7] combined fuzzy logic with MATLAB tools in order to improve practical application in the field.

The limitation for the present research is that the work provides a practical foundation for fuzzy logic implementation but does not specifically address adaptive neuro-fuzzy speed control for an EV.

A PID-type fuzzy logic controller (FLC) with tuning using particle swarm optimization was proven to be more robust and converging faster by Bouallègue et al. [8].

The approach improves fuzzy-PID parameter tuning using particle swarm optimization (PSO), but the tuning process is based on an optimization technique rather than a neural learning mechanism. Therefore, the capability of continuous neural adaptation of fuzzy parameters is not investigated.

It is a novel fuzzy controller for a charging station applying Agrawal and Shrivastava [9]. The method enhances the speed control of BLDC motor using fuzzy logic with the help of PSO which helps in minimizing the error and increasing the torque of the BLDC motor.

The limitation relevant to the proposed work is that the approach combines fuzzy logic with PSO for BLDC motor control but does not implement neural-network-based adaptation. Hence, it does not directly establish the benefit of combining fuzzy reasoning with neural learning.

In the study [10], the performance of fuzzy-PI controllers for BLDC motors was investigated and their dynamic and steady-state behavior was shown to be improved.

The study focuses on fuzzy-PI control of a BLDC motor, while the integration of neural learning for automatic adaptation of fuzzy parameters is not addressed in the material reviewed here. This leaves a gap for investigating neuro-fuzzy control for adaptive EV speed regulation.

Han et al. [11] proposed model for vehicle active suspension mechanism, an adaptive fuzzy PID control strategy that demonstrates effective control performance under diverse road conditions.

Although the study demonstrates the ability of adaptive fuzzy PID control to respond to changing road conditions, its application is focused on vehicle active suspension rather than propulsion-speed control. Therefore, the findings cannot directly establish the effectiveness of the approach for EV motor-speed regulation.

Xu et al. [2] implemented an adaptive fuzzy PID controller for real-time speed control in pure EVs. The method involves applying fuzzy logic and PID controller to control the armature voltage of the motor. The speed control system model in MATLAB with Simulink Toolbox is utilized to test the performance of the adaptive speed control method. The performance validation involves testing the system with acceleration and deceleration processes, each divided into stages with desired accelerations and decelerations around 2 ms².

The system model in MATLAB with Simulink Toolbox validated the adaptive speed control method, showing a good role in adapting speeds of the vehicle to targets within desired time frames. The adaptive speed controller had a great performance on both acceleration and speed control, particularly when the vehicle was expected to accelerate or decelerate with a significant change in speed.

The limitation is that the evaluation is based on the specified MATLAB/Simulink operating conditions and an adaptive fuzzy-PID structure. The material does not establish neural learning of the controller parameters or practical hardware validation, leaving scope for further investigation.

2.3 Neural network and neuro-fuzzy control

Neural networks provide learning capabilities that can be combined with fuzzy logic to improve adaptability. Neuro-fuzzy systems aim to combine the rule-based reasoning of fuzzy logic with the learning capability of neural networks.

Breesam et al. [12] discussed speed control management methods, specifically highlighting traditional PID controllers and the proposed fuzzy neural network (FNN) control method. The FNN approach discussed here utilizes a backpropagation algorithm, offers improved speed response and synchronization in multi-motor systems, while PID controllers face challenges such as load disturbance and sensitivity to motor parameters.

The study demonstrates the potential of a fuzzy neural network and backpropagation for speed control; however, the work focuses on a multi-motor system. The direct application and validation of the approach for the proposed EV speed-control architecture therefore requires further investigation.

Ding et al. [13] extended the same research and incorporated fuzzy logic with neural networks as a result developing a neuro-fuzzy controller for an electric motor drive that displayed better adaptability in complicated situations.

The limitation relevant to the present study is that the work addresses an electric motor drive, while the integration of the neuro-fuzzy controller with complete EV operating conditions and vehicle-level dynamics requires further investigation.

Reddy and Sharma [14] developed an adaptive neuro-fuzzy controller (ANFC) for EVs powered by fuel cells.

Although the work addresses an ANFC for an EV, its application is associated with fuel-cell-powered EVs. The controller therefore does not directly establish the performance of the proposed approach for the BLDC-based battery-operated EV configuration considered in this study.

Pamuji et al. [15] compared fuzzy logic and neural network-based BLDC motor controllers, and found that the neural network-based BLDC motor controller provides better speed regulation and energy efficiency.

The comparison between fuzzy logic and neural-network-based BLDC controllers is relevant to intelligent motor control; however, the study compares the two approaches separately rather than establishing a unified ANFC. Thus, it does not directly demonstrate how neural learning can optimize a fuzzy controller.

George et al. [16] introduced an ANFIS-based fractional-order PID controller, optimized using ant colony optimization, and achieved robust speed tracking performance.

The controller combines ANFIS, fractional-order PID, and ant-colony optimization, resulting in a comparatively complex control structure. The rule base formulation for the fuzzy logic scheme was also identified as time-consuming. This indicates that controller design complexity and parameter formulation remain important considerations for practical implementation.

Reddy et al. [17] highlighted the optimization of the proposed controller using neural networks and compare its performance with existing conventional methods, including PID controllers. A neural network-based fuzzy plus PD controller specifically designed for speed control in EVs, particularly for BLDC motors.

The study is directly relevant to intelligent EV speed control; however, the fuzzy-plus-PD structure differs from the proposed neuro-fuzzy speed-control framework. Further investigation is therefore required to establish the effectiveness of adaptive fuzzy control and neural learning under the operating conditions considered in the present study.

2.4 Other intelligent and hybrid control approaches relevant to EV control

Comparative summary of existing neuro-fuzzy-based speed management techniques along with key findings and specific limitation gaps is presented in Table 1.

Qin and Badgwell [18] presented a predictive control model of industrial applications that stresses the importance of adaptability in dynamic systems such as EVs.

The work provides a foundation for model predictive control in dynamic systems, but it is not specifically focused on neuro-fuzzy EV speed control.

Song et al. [19] investigated the sliding mode control of electric drives. Sliding-mode control provides an alternative non-linear control strategy for electric drives, but it follows a different control principle from fuzzy-neural control. Therefore, it does not address the adaptive fuzzy-rule and neural-learning mechanism considered in this study. Wang et al. [20] proposed hybrid methods which highlights energy-efficient EV speed control, it combines fuzzy logic with other optimization techniques to maximize performance.

The use of hybrid optimization techniques demonstrates the importance of combining intelligent methods for energy-efficient EV control. However, the approach does not directly establish a neural adaptation mechanism for fuzzy speed control, which remains a focus of the present research. Wang et al. [21] used particle swarm optimization for speed optimization methods of hybrid EVs; thereby enhancing the efficiency of the queues. 

Table 1. Comparative study of existing neuro-fuzzy-based speed management techniques

Note: proportional-integral-derivative (PID); particle swarm optimization (PSO); electric vehicle (EV); fuzzy logic controller (FLC); fuzzy neural network (FNN); adaptive neuro-fuzzy controller (ANFC).

The study uses PSO for speed optimization in hybrid EV applications. However, PSO-based optimization is different from online neural adaptation of fuzzy-control parameters, and therefore the specific contribution of a neuro-fuzzy controller is not established by this work. Benmouna et al. [23] addressed the problem of non-linear control and optimization of hybrid EVs under resource constraints.

The study addresses non-linear control and optimization of hybrid EVs, but the specific neuro-fuzzy speed-control mechanism considered in the present research is not investigated. The difference in control architecture leaves scope for further comparison of intelligent control methods.

Zaid et al. [22] developed a powerful photovoltaic energy model, which emphasized on improved response and energy management.

The work is mainly concerned with photovoltaic energy and EV charging rather than direct propulsion-speed control. Therefore, its relevance to the present study is limited to energy-management and control-system considerations and does not directly address neuro-fuzzy EV speed regulation.

The proposed speed control system is based on a neuro-fuzzy controller, which combines neural networks and fuzzy logic in a single system to overcome the drawbacks of the traditional PID controller and fuzzy PID controller. Below is the process for developing and implementing the system.

The ANFC system's speed management logic is illustrated in the flowchart in Figure 1. The first step in this process is to establish a reference velocity. The hall sensor signal is checked and the reference speed is compared with a carrier signal to create PWM pulses that are used to control the BLDC motor via a 3-phase inverter, when the system is in the OPEN LOOP mode. The actual speed is calculated using feedback from hall sensors in closed loop and the speed error is calculated by comparing the desired speed with the actual speed. The error is fed into the PI controller which generates a control signal to correct the error and changes the PWM pulses to keep the speed as desired. This loop is repeated until the desired speed is attained, which enables accurate and adaptive motor speed control.

tu_pian_1.png

Figure 1. Adaptive neuro-fuzzy controller (ANFC) model for speed management in automatic EV

4.1 Control flow description for adaptive neuro-fuzzy controller model for speed management in automatic EV

The ANFC model for speed management in an automatic EV is presented in this algorithm. This is the algorithm of the ANFC model for speed management in an automatic EV. The ANFC of an automatic EV is a hybrid controller that combines traditional and intelligent control techniques to ensure accurate performance and adaptability of the motor.

Initially, the system starts by setting a desired speed as the reference value. Open loop control: The control system directly utilizes the hall sensor signals to output PWM signals, which are compared with a carrier value and then sent to the BLDC motor through a 3-phase inverter. The system adjusts the speed in closed-loop by checking the actual motor speed through hall sensors and error is the difference between the reference speed and the actual speed. The error can be handled by either a PI controller that provides proportional and integral gains to speed up the PWM signals or by an adaptive neuro-fuzzy logic system that automatically adjusts the PWM signals using fuzzy rules and optimizes the neural network.

The ANFC is an adaptive strategy that can learn and tune the control strategy on the fly, efficiently handling changes in load, road conditions or disturbances. The loop is repeated until the desired speed is reached, combining accurate error correction (PI) and smart adaptability (neuro-fuzzy) control. With this logical integration, the EV operates efficiently, with stability and without any problems under all circumstances.

4.1.1 Control flow description steps

Start: Initialize the system.

Set the Desired Speed Value: Assign a reference speed value.

Check Open Loop Condition

•If the system operates in an open-loop mode, proceed to Step 4.

•If not, skip to Step 7.

Check Hall Sensor Signals: Read the signals from the hall sensors.

Compare Reference and Carrier Values: Compare the desired speed (reference value) with the carrier value.

Generate PWM Pulses: Use the comparison results to generate the necessary PWM pulses.

Control the BLDC Motor

•Apply the generated PWM pulses to the 3-Phase Inverter to drive the BLDC motor.

Calculate Actual Speed: Measure the actual speed of the motor.

PI Controller

•Calculate the speed error: Error = (Set Speed - Actual Speed)

•Use the error to generate the PI output for adjusting the PWM pulses.

Adjust PWM Pulses: Update the PWM signals based on the PI controller output.

Repeat Steps 8–10: Continue until the desired speed is achieved.

End: Stop the control loop as required.

4.2 Fuzzy neuro speed control equation

The equations which are used primarily on classical control methods (e.g., Proportional-Integral (PI) control) and motor dynamics. While they form the foundation for understanding motor behavior, they need to be adapted or extended for fuzzy neuro speed control.

•Fuzzy Logic Control (FLC): Uses a rule-based approach to control the motor by mapping speed errors and their rate of change into fuzzy sets (e.g., "low," "medium," "high") and deciding the control output based on predefined rules.

•Neural Networks: Can be integrated with fuzzy logic to optimize the fuzzy rules and membership functions dynamically, enabling more adaptive control in uncertain environments. Therefore, the neuro-fuzzy control is more data-driven and rule-based.

The fuzzy neuro control is more data-driven and rule-based; the fundamental equations, such as error calculation and motor dynamics, are still relevant. Here's how they integrate:

Based on the proposed system flow, the neuro-fuzzy-based control system speed controller model can be mathematically modelled step by step. Here are the equations representing each step:

Reference Speed (set value): Let the desired speed (set value) be

$Wref=Set~Speed$           (1)

Hall Sensor Signals: The hall sensors provide the rotor angular position θ(t), which is used to estimate the actual speed.

${{W}_{actual}}\left( t \right)=\frac{d\theta \left( t \right)}{dt}$           (2)

Error Calculations:

$E\left( t \right)=Wref-Wactual$          (3)

PI Controller Replacement: In a fuzzy neuro system, the PI controller equation

$\mu \left( t \right)={{K}_{p}}e\left( t \right)+{{K}_{i}}\mathop{\int }_{0}^{t}e\left( t \right)d$          (4)

PWM Signal Generation: The control output $\mu \left( t \right)~$from the fuzzy/neuro system adjusts the PWM duty cycle:

$\begin{gathered}\text PWM ~DutyCycle= \text { Wfuzzy/neuro }\left(e(t),\left(\frac{d e(t)}{d t}\right)\right)\end{gathered}$          (5)

Motor Dynamics: The motor dynamics equation:

$J\frac{d{{W}_{actual}}}{dt}+B{{W}_{actual}}={{T}_{actual}}-{{T}_{load}}$         (6)

where, J is the equivalent rotational inertia of the motor and drivetrain. B is the viscous friction coefficient. T_actual is the actual motor torque, and T_load represents the equivalent load torque acting against the motor.

The fuzzy controller generates the commanded motor torque (T_CMD) from the controller inputs. The commanded torque is applied to the vehicle/motor dynamic model as

${{T}_{CMD}}=Sat\left( {{T}_{CMD}},~{{T}_{min}},~{{T}_{max}} \right)$        (7)

where,

Sat represents the actuator saturation function T_min and T_max are the minimum and maximum allowable motor torque values.

The neuro-fuzzy-based equations (error calculation, PWM generation, and motor dynamics) are applicable and are reinterpreted and integrated into the fuzzy neuro framework. Specifically, the PI controller output equation is replaced by the fuzzy/neuro logic.