Direct torque control of electric vehicle drives using hybrid techniques

ABSTRACT

: Field-oriented control

INTRODUCTION
In past decades, PMSMs were identified as promising materials for EV applications because of their low weight, high reliability, speed control variation, low distortion, and efficient performance [1]. In the context of EV traction, a quick and reliable torque response of the electrical drive is necessary to satisfy the instantaneous torque demand ordered by the operator via drive movement [2]. Additionally, the PMSM drive should be capable of recovering quickly from system disruptions in terms of torque and speed. These are a few of the demands on the EV's performance that must be met to match the driving characteristics [3]. In order to torque ripple minimization with enhancing the steady-state performance, FOC is utilized. FOC enhances the efficacy of the dynamic response [4], [5]. However, when the temperature varies, this technique becomes susceptible to even minor changes in the parameter.
To enhance the performance of the PMSM motor, numerous control strategies are suggested. The shortcomings of FOC were replaced by a brand-new method known as DTC by Bao et al. [6]. Torque and flux can be immediately controlled by DTC by making the PMSM drive more effective. By using the torque error and motor speed to determine the active vector, complicated mathematical processes are avoided [7]. But it also makes the monitoring system noisy [8]. To provide appropriate steady-state performance, torque, stator flux disturbances and high-frequency components have to be eliminated [9]. The existing solutions completely disregard rotating components and solely use active vectors, or a combination of dynamic matrices and zero matrices [10].
To solve the synchronous torque and flux in the PMSM machine, Pushparajesh et al. [11] developed TSMPC based on lexicographic technique. However, the assumptions made in this study are that a best voltage vector as well as every cost function's limitation was confined to a mere lower limit. For the DTC of PMSM, Nasr et al. [12] suggested a FST. ZST generally does not permit the reversal of revolution, which is its fundamental error. By increasing the area with a duty cycle angle to improve the torque and speed response, Zhang et al. [13] showed a methodology that was known as: MPDTC. A model-free deep reinforcement learning torque regulator was developed by Schenke and Wallscheid [14]. Yet, the reinforcement learning was not applied to dynamic stability theory. To significantly minimize the torque fluctuations for PMSM drive, Mohammed et al. [15] demonstrated an IL-DTC. Unfortunately, parameter variations restricted the usage of SPMSM which reduced the motor's ability to manage torque disturbances. In PMSM, to enhance the efficiency, a combined mechanism [16] was developed. But leakage current was the root of the problem. For PMSM, SVPWM-DTC [17] was proposed; however, because of the bandwidth restriction, the motor experienced significant torque ripple. For PMSM drives, a brand-new DCF-MPDSC [18] with duty ratio minimization was introduced. Every DTC computation currently in use is carried out in a stationary reference frame without the qualitative data of the rotor location [19]. Because of its simplicity, it can run each computing cycle quickly and with a high sample frequency [20]. The ripple will roughly decrease for every increase in the sampling frequency [21]. The excessive torque ripple of the DTC control was a significant drawback. This generally does not pose a problem in some applications since the ripple is filtered by the rotor and the load inertia, but in other applications like electric vehicles, low torque ripple is crucial [22].
In this research, an approach based on JSO-ANFIS is used to present the solution to the aforementioned problem. The main contribution of current study is stated as: i) the JSO-ANFIS-based DTC of the PMSM drive is adopted to eliminate torque ripples and improve the speed; ii) a variable switching table for the DTC of the PMSM is computed to enhance steady state and dynamic functionality; and iii) in stationary reference, the PMSM's dynamic behavior is described to prevent complex frame transportation.
The organization of current research is specified; the proposed method with the statistical equations are explained in section 2. The simulation outcomes as well as the comparison evaluation with conventional approaches are represented in section 3. At last, the research conclusion is briefly and clearly explained in section 4.

METHOD
Here, the speed and torque of the PMSM drive depend on JSO-ANFIS-based DTC control. JSO is used to optimize the ANFIS learning variable in a variety of functioning scenarios. Creating a DTC system to enhance the PMSM drive's effectiveness is considered as the major goal of the current research which enables smooth torque output as well as better speed control monitoring.

DTC of PMSM
In the traditional DTC scheme, the regulation of a PMSM motor in an EV obtains various fluctuations, those of which are surpassed by employing the proposed DTC control. Figure 1 illustrates the overall block diagram for the suggested technique. During the operation of classical DTC, the steady state signals show clear flux linkage vectors within the six sectors boundary [23]. Enhancing the effectiveness of the PMSM by focusing on continuous torque response and speed by using hybrid JSO-ANFIS-based DTC control is considered as the main objective in this work.

Jellyfish search optimizer
Globally, jellyfish are mostly available in greater depths under the sea [24]. By combining all the paths in the sea used by the jellyfish, the ocean current's main trend is found which is represented in (1).
Then fix = Consequently, (3) represents the updated ( ⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗ ), is number of jellyfish, is attraction control variable, * is best position of jellyfish. Every jellyfish's mean position, and the distance between a current best location of jellyfish as well as the average location is which is given in (4).
Then, (10) is modified and as (11), where, > 0 is the distribution coefficient, In both passive (type A) as well as active (type B) ways, Jellyfish can swim with their groups. Type A is simply known as motion of jellyfish within their locations, and (12) represents all jellyfish's new location (12), is Upper bound, is Lower bound, > 0 is coefficient of movement, whereas, = 0.1 is obtained using (12). Moreover, (13) expresses this effort which is deliberated as a local exploration's operative manipulation, Here, ⃗⃗⃗⃗⃗⃗⃗⃗⃗ = (0,1) × ⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗ (14) here, is location 's function.
Thus, ( + 1) = ( ) + ⃗⃗⃗⃗⃗⃗⃗⃗⃗ The movement type over a period of time is defined by utilizing a temporal control mechanism. The time control system comprises of a time control function ( ) and constant which are used to control the ability of jellyfish. The time control randomly changes between 0 &1, thus the 's accuracte value remains unknown. So, is fixed to 0.5, which is expressed in (17).
Thus, the (18) represents the written map, here, is ℎ jellyfish location's chaotic value, 0 is initial population in [0,1], where η = 4.0. Due to spherical shape of earth, a jellyfish can get back to the contrary bound when it gets out from the search area which is given in (19), , is ℎ jellyfish's location in ℎ dimension, and ′ , is updated position.

ANFIS
This control technique actually combines the processes of fuzzy as well as ANN [25]. ANFIS is used to create the torque standard from speed error variations, each time. The network comprises 20 nodes and undergoes 12 epochs of training. Two-Gaussian membership functions are analyzed by each node in this layer, after which, the input parameters are transferred to the following layer. The non-linear element is included in each iterative count which is described as input MF , , which is modified via the (20).
Step time is stated as ; learning rate is stated as ; steepest momentum constant is referred to as .

JSO-optimized ANFIS
In this section, under different PMSM drive operating conditions JSO-optimized learning parameters for ANFIS are described. To minimize the next objective function denoted by (21), the algorithm constraints are chosen. After that, the selected constraints are intended as the fitness function for optimization to ensure consistency and achieve higher restraint in the event of unanticipated load disruption and speed differences.
is stated as the speed at maximum level; and is referred to as an actual and reference speed; is stated as Sample count; the range of tuning variables are declared in (22) and (23).

SIMULATION RESULTS AND DISCUSSION
Numerous features of direct torque control for PMSM drives are discussed in this study. The JSO-ANFIS is implemented and simulated using following setup: This research is implemented in MATLAB R2020a tool using the Intel Core i3 processor, Windows 8 OS and a RAM of 8 GB. Moreover, Table 2 shows the ratings of the simulation model. The creation of DTC regulation in the PMSM motor utilizing JSO-ANFIS, is experimentally executed by using MATLAB is demonstrated in Figure 2. Table 2 contains all of the control variables for the simulation environment. A 10-second sample rate is specified. Specifically, the inertia bands are set to 1.8 Nm and 0.088 Wb. Figure 3 display the speed performance of ANFIS. The motor's speed is being used as a response, and the difference between it and to determine the speed inconsistency, the reference speed is used. A reference value of 4,000 along with the projected speed error, variation, and error, are all sent to the JSO-ANFIS controller. Similarly, Figure 4 display the ANFIS torque.  The simulation findings demonstrate that JSO-ANFIS achieves superior performance related to the ANFIS. Additionally, it might make the system more stable. The suggested JSO-ANFIS outperformed conventional systems by achieving a stable state in 1.7 seconds. The suggested JSO-optimized ANFIS, conventional proportional integral derivative (PID), and ANFIS are used to evaluate the steady-state assessment and other constraints (settling time, rising time, and overshoot time) which are tabulated in Table 3. Figures 5 to 7 show the JSO-ANFIS controller's findings for speed and torque. The comparison of transient response with conventional controllers is shown in Table 4. Even while traditional PID and ANFIS operate superior in terms of rise and settling times, overshoot is more obvious and invalidates all other transient quality objectives.  The evaluation of torque ripple is shown in Table 4 which evidently proves that compared to current SVPWM-DTC [16] and ANFIS, the suggested JSO-ANFIS achieves a smaller torque ripple. Also, when related with previous models of SVPWM-DTC [16], ANFIS, and GEO-ANFIS, which reached 0.6 Nm, 0.53 Nm, and 0.44 Nm correspondingly, the suggested JSO-ANFIS obtains a smaller torque ripple of 0.26 Nm. The harmonic evaluation of the PMSM motor is shown in Figure 7. The proposed JSO-ANFIS controller exceeds the currently used DCF-MPDSC [17] and ANFIS controller in the control method.

CONCLUSION
For the past few years, real-time control and DTC control of PMSM drives have achieved a lot of attention as it improves the effectiveness of the system. The outcome of the conventional direct torque control is a large torque as well as flux linkage ripple centered on the 60 degrees sector that included the position signal to choose the appropriate space voltage vector. Here, JSO is used in many situations to maximize the ANFIS learning variables. To achieve closed-loop operation, a speed controller called the JSO-ANFIS is used. The MATLAB/Simulink results demonstrate that the suggested JSO-ANFIS can eliminate the uncertainty problems brought on by speed/load variations. In comparison to the DCF-MPDSC controller, the simulation observed that the proposed technique delivers minimal torque ripple (0.26 Nm) and less THD (1.21%) which enhances system efficiency. The strong formula and rule modules make it easy to design and install the JSO-ANFIS controller. This research project may be expanded in the future using further meta-heuristic approaches or hybrid intelligence methods.