A permanent magnet synchronous motor sliding mode direct torque control method and system based on a hybrid whale optimization algorithm

Abstract

The application belongs to the technical field of permanent magnet synchronous motor control, and discloses a permanent magnet synchronous motor sliding mode direct torque control method and system based on a hybrid whale optimization algorithm; a sensitivity evaluation and decoupling mapping are performed on a parameter set to be optimized, a multi-scale search space is constructed, and the speed outer loop control parameters and the torque flux inner loop sliding mode control parameters are hierarchically processed according to the sensitivity difference. In the optimization process, the characteristic indexes of different response stages are extracted, and a stage perception comprehensive fitness evaluation strategy is adopted. By introducing an evolutionary stagnation factor and a nonlinear dynamic conversion probability mechanism, the wide-area exploration mode of white whale swimming and the local development mode of whale spiral attack are intelligently switched. A directional disturbance whale falling mechanism is adopted to maintain the population diversity, and a phased parameter optimization strategy is implemented, that is, the stability parameter group is optimized first, and then the anti-chattering parameter group is optimized. The application can improve the dynamic response performance, steady-state accuracy and anti-interference ability of the motor control system.

Description

Technical Field

[0001] This invention relates to the field of permanent magnet synchronous motor control technology, and more specifically, to a sliding mode direct torque control method and system for permanent magnet synchronous motors based on a hybrid whale optimization algorithm. Background Technology

[0002] Existing sliding mode direct torque control systems for permanent magnet synchronous motors (PMSMs) suffer from numerous technical bottlenecks in parameter optimization, severely limiting their performance in key applications such as high-precision servo drives, electric vehicle drives, and industrial robots. First, the system involves multi-dimensional control parameters, including proportional-integral control parameters in the speed outer loop, sliding surface coefficients, switching gain, and linear gain in the torque and flux linkage inner loops. These parameters exhibit complex nonlinear coupling relationships. Traditional step-by-step debugging methods are not only time-consuming and labor-intensive but also often result in mutual interference between parameters. Engineers frequently encounter situations where improved dynamic response leads to reduced steady-state accuracy. Second, existing intelligent optimization algorithms, such as genetic algorithms and particle swarm optimization, commonly converge to local optima prematurely when facing such a high-dimensional and strongly coupled search space. This is particularly problematic under complex conditions (such as sudden load changes, parameter perturbations, and external disturbances). The algorithms often become trapped near suboptimal solutions in the early stages of the search, failing to find the truly globally optimal parameter combination. This leads to problems such as excessive overshoot, unstable oscillations, and poor anti-interference capabilities in the actual operation of the control system. Traditional methods neglect the varying sensitivity of different control parameters to system performance, employing uniform search step sizes and optimization strategies. This can lead to key parameters with significant impact on system performance missing their optimal values ​​due to overly coarse search granularity, while parameters with less impact on performance are over-optimized, wasting computational resources. Furthermore, existing fitness evaluation functions typically use fixed weight allocations, failing to adaptively adjust to the characteristics of different motor operation stages. For example, dynamic response performance should be prioritized during startup and acceleration, while torque ripple and efficiency should be emphasized during steady-state operation. This evaluation method struggles to accurately reflect the diversity and stage-specific characteristics of actual control requirements, thus affecting the convergence direction and final performance of the optimization algorithm.

[0003] In view of this, the present invention proposes a sliding mode direct torque control method and system for permanent magnet synchronous motors based on a hybrid whale optimization algorithm to solve the above problems. Summary of the Invention

[0004] To overcome the aforementioned deficiencies of the prior art and to achieve the above objectives, the present invention provides the following technical solution: a sliding mode direct torque control method for permanent magnet synchronous motors based on a hybrid whale optimization algorithm, comprising: The set of parameters to be optimized for the sliding mode direct torque control system of permanent magnet synchronous motor is obtained. The sensitivity of each parameter dimension in the set of parameters to be optimized is evaluated. Based on the evaluation results, the parameter dimensions are decoupled and mapped to construct a multi-scale search space. The set of parameters to be optimized includes the outer loop control parameters of speed and the inner loop sliding mode control parameters of torque and flux linkage. A hybrid whale population is initialized in a multi-scale search space. For each individual in the hybrid whale population, a set of control parameters is reconstructed based on the individual's position vector in the multi-scale search space and input into a preset motor control model for operation. Dynamic response data during operation is collected. Based on dynamic response data, feature indicators of the system at different response stages are extracted, and the fitness evaluation strategy is dynamically adjusted in combination with the feature indicators to calculate the individual's stage-perception comprehensive fitness value. In the iterative evolution of the mixed whale population, an evolutionary stagnation factor is introduced, and the nonlinear dynamic transition probability is calculated jointly based on the evolutionary stagnation factor and the current iterative state. When the random judgment value is less than the nonlinear dynamic transformation probability, the individual is controlled to enter the wide-area exploration mode, and the individual's position vector is updated based on the beluga whale swimming mechanism and combined with the dimension scaling factor of the multi-scale search space. When the random judgment value is greater than or equal to the nonlinear dynamic transformation probability, the individual is controlled to enter the local development mode, and the position vector of the individual is updated based on the whale spiral attack mechanism and combined with the dynamic convergence factor. After each location update, inferior individuals in the population are identified based on the phase-aware comprehensive fitness value. A targeted perturbation whale fall mechanism is executed on the inferior individuals to generate reset individuals, and the reset individuals replace the corresponding inferior individuals. Repeat the above iterative evolution process until the termination condition is met, output the target control parameter set corresponding to the global optimal position vector, and configure the target control parameter set into the permanent magnet synchronous motor sliding mode direct torque control system to drive the motor.

[0005] A sliding mode direct torque control system for permanent magnet synchronous motors based on a hybrid whale optimization algorithm, used to implement the aforementioned sliding mode direct torque control method for permanent magnet synchronous motors based on a hybrid whale optimization algorithm, includes: Main control unit, inverter, signal acquisition unit and permanent magnet synchronous motor; The signal acquisition unit is used to acquire real-time operating data of the permanent magnet synchronous motor. The main control unit is connected to the signal acquisition unit and the inverter. The main control unit includes a parameter optimization subunit and a direct torque control subunit. The parameter optimization subunit includes: The space construction module is used to obtain the set of parameters to be optimized for the sliding mode direct torque control system of permanent magnet synchronous motor, to evaluate the sensitivity of each parameter dimension in the set of parameters to be optimized, and to decouple and map each parameter dimension based on the evaluation results to construct a multi-scale search space. The model simulation module is used to initialize a hybrid whale population in a multi-scale search space. For each individual in the hybrid whale population, the control parameter set is reconstructed based on the individual's position vector in the multi-scale search space and input into a preset motor control model for operation, and dynamic response data during operation is collected. The fitness assessment module is used to extract characteristic indicators of the system at different response stages based on dynamic response data, and to dynamically adjust the fitness evaluation strategy in combination with the characteristic indicators to calculate the individual's stage-aware comprehensive fitness value. The transition probability calculation module is used to introduce an evolutionary stagnation factor during the iterative evolution of a hybrid whale population, and to jointly calculate the nonlinear dynamic transition probability based on the evolutionary stagnation factor and the current iterative state. The wide-area exploration module is used to control an individual to enter wide-area exploration mode when the random judgment value is less than the nonlinear dynamic transformation probability. It updates the individual's position vector based on the beluga whale swimming mechanism and combined with the dimension scaling factor of the multi-scale search space. The local development module is used to control an individual to enter local development mode when the random judgment value is greater than or equal to the nonlinear dynamic transformation probability. It updates the individual's position vector based on the whale spiral attack mechanism and combined with the dynamic convergence factor. The whale fall reset module is used to identify inferior individuals in the population based on the phase-aware comprehensive fitness value after each location update, perform a targeted perturbation whale fall mechanism on the inferior individuals to generate reset individuals, and replace the corresponding inferior individuals with the reset individuals. The optimization output module is used to control the repeated iterative evolution process of the above modules until the termination condition is met, and outputs the target control parameter set corresponding to the global optimal position vector; The direct torque control subunit is used to receive the target control parameter set output by the optimization output module to configure the controller parameters, and to generate a space vector pulse width modulation signal by combining the real-time operating data input by the signal acquisition unit, and output the space vector pulse width modulation signal to the inverter.

[0006] The technical effects and advantages of the sliding mode direct torque control method and system for permanent magnet synchronous motors based on the hybrid whale optimization algorithm of this invention are as follows: This invention constructs a multi-scale search space, enabling refined searching of high-sensitivity parameters and coarse-grained searching of low-sensitivity parameters. This effectively avoids the problem of missing optimal values ​​for high-sensitivity parameters due to excessively large search step sizes, while preventing over-optimization of low-sensitivity parameters, thus improving parameter optimization efficiency and reducing ineffective computation. The phase-aware integrated fitness evaluation strategy focuses on optimizing dynamic response characteristics during motor startup and acceleration, and on improving torque and flux pulsation during steady-state operation. The hybrid whale optimization algorithm, through the introduction of evolutionary stagnation factors and nonlinear dynamic transition probabilities, automatically adjusts the ratio of exploration to development based on the algorithm's convergence state, effectively overcoming the local convergence bottleneck of traditional optimization algorithms in high-dimensional parameter spaces and improving the success rate of finding the global optimum. The phased parameter optimization strategy, through hierarchical processing—first fixing anti-bounce parameters to optimize stability parameters, and then optimizing anti-bounce parameters based on stability parameters—significantly reduces coupling interference between parameters, enabling the orderly solution of complex multi-dimensional parameter optimization problems and greatly shortening engineering debugging time. Attached Figure Description

[0007] Figure 1 This is a schematic diagram of a sliding mode direct torque control method for permanent magnet synchronous motors based on a hybrid whale optimization algorithm, according to the present invention. Figure 2 This is a schematic diagram of a sliding mode direct torque control system for a permanent magnet synchronous motor based on a hybrid whale optimization algorithm, according to the present invention. Detailed Implementation

[0008] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0009] Please see Figure 1 The present invention proposes a sliding mode direct torque control method for permanent magnet synchronous motors based on a hybrid whale optimization algorithm, comprising the following steps: Step 1: Obtain the set of parameters to be optimized for the sliding mode direct torque control system of the permanent magnet synchronous motor. Perform sensitivity evaluation on each parameter dimension in the set of parameters to be optimized. Based on the evaluation results, decouple and map each parameter dimension to construct a multi-scale search space. The set of parameters to be optimized includes the outer loop control parameters of speed and the inner loop sliding mode control parameters of torque and flux linkage. Step 2: Initialize the hybrid whale population in the multi-scale search space. For each individual in the hybrid whale population, reconstruct the control parameter set based on the individual's position vector in the multi-scale search space, and input it into the preset motor control model for operation, and collect dynamic response data during the operation process. Step 2 specifically includes: Step 2.1: Initialize the hybrid whale population in the multi-scale search space by randomly generating an initial position vector for each individual. Specifically, the population size N is set to 30 individuals. This population size ensures sufficient coverage of the search space while avoiding excessive computational resource consumption due to an overly large population. For each individual i in the population, where i ranges from 1 to 30, an initial position vector generation operation is performed. The initial position vector is a vector containing 8 values, corresponding to the initial values ​​of 8 control parameters to be optimized. A uniformly distributed random initialization strategy is adopted. For each dimension j of the position vector, where j ranges from 1 to 8, random initial values ​​are generated based on the value range of the corresponding parameter for that dimension. For example, the first dimension corresponding to the speed outer loop proportional gain Kp has a value range of 0 to 100, and its initial value is obtained by generating a uniformly distributed random number between 0 and 1 and multiplying it by 100; the second dimension corresponding to the speed outer loop integral gain Ki has a lower bound of 0 and an upper bound of 500, and its initial value is obtained by generating a uniformly distributed random number between 0 and 1 and multiplying it by 500; the third dimension corresponding to the torque inner loop sliding surface coefficient cT has a lower bound of 10 and an upper bound of 200, and its initial value is obtained by generating a uniformly distributed random number between 0 and 1, multiplying it by 190, and then adding 10; the fourth dimension corresponding to the torque inner loop switching gain εT has a lower bound of 50 and an upper bound of 500, and its initial value is obtained by generating a uniformly distributed random number between 0 and 1, multiplying it by 450, and then adding 50; the torque inner loop linear gain k The 5th dimension corresponding to T has a lower bound of 100 and an upper bound of 1000. The initial value is obtained by generating a uniformly distributed random number between 0 and 1, multiplying it by 900, and then adding 100. The 6th dimension corresponding to the inner loop sliding surface coefficient cψ of the flux linkage has a lower bound of 10 and an upper bound of 200. The initial value is obtained by generating a uniformly distributed random number between 0 and 1, multiplying it by 190, and then adding 10. The 7th dimension corresponding to the inner loop switching gain εψ of the flux linkage has a lower bound of 50 and an upper bound of 500. The initial value is obtained by generating a uniformly distributed random number between 0 and 1, multiplying it by 450, and then adding 50. The 8th dimension corresponding to the inner loop linear gain kψ of the flux linkage has a lower bound of 100 and an upper bound of 1000. The initial value is obtained by generating a uniformly distributed random number between 0 and 1, multiplying it by 900, and then adding 100. By randomly generating values ​​dimension by dimension, the initial value for each dimension is calculated as follows: the initial value of the dimension equals the lower bound of the parameter plus a uniformly distributed random number between 0 and 1 multiplied by the difference between the upper and lower bounds of the parameter. This ensures that the generated initial values ​​are uniformly distributed within the range of the corresponding parameter, resulting in a complete initial position vector for each individual. This vector contains 8 numerical components, and these initial positions are uniformly distributed in the search space, providing diverse starting points for subsequent optimization iterations. Simultaneously, the historical best position record is initialized for each individual, with the initial value set to the current individual's initial position vector, and the historical best fitness value set to a maximum value for subsequent recording and updating of historical best information.

[0010] Step 2.2: For each individual in the mixed whale population, reconstruct the control parameter set based on the individual's position vector in the multi-scale search space. Specifically, this includes: for each individual in the population, reading the eight numerical components of its position vector. Establishing a one-to-one correspondence between the position vector dimension and the control parameters: the first numerical component corresponds to the outer loop proportional gain Kp of the rotational speed, and this value is directly assigned to the Kp parameter; the second numerical component corresponds to the outer loop integral gain Ki of the rotational speed; the third numerical component corresponds to the inner loop sliding surface coefficient cT of the torque; the fourth numerical component corresponds to the inner loop switching gain εT of the torque; the fifth numerical component corresponds to the inner loop linear gain kT of the torque; the sixth numerical component corresponds to the inner loop sliding surface coefficient cψ of the flux linkage; the seventh numerical component corresponds to the inner loop switching gain εψ of the flux linkage; and the eighth numerical component corresponds to the inner loop linear gain kψ of the flux linkage. Each component of the position vector is assigned a value to its corresponding control parameter, thus reconstructing the complete control parameter set for that individual object. This parameter set contains the specific values ​​of eight parameters: Kp, Ki, cT, εT, kT, cψ, εψ, and kψ. For example, assuming the position vector of an object is 35.2, 120.5, 55.3, 210.7, 450.3, 48.9, 185.6, and 380.7, the reconstructed control parameter set would be Kp = 35.2, Ki = 120.5, cT = 55.3, and so on. The reconstructed control parameter set provides complete parameter input for the subsequent operation of the motor control model.

[0011] Step 2.3 involves inputting the reconstructed control parameter set into the preset motor control model, configuring the controller parameters, and starting the model. Specifically, this includes receiving the reconstructed control parameter set and performing parameter configuration operations on the motor control model. The preset motor control model is built in the Matlab / Simulink simulation environment and includes a permanent magnet synchronous motor body model, a three-phase inverter model, a space vector pulse width modulation module, a coordinate transformation module, a speed outer-loop PI controller, a flux sliding mode controller, a torque sliding mode controller, a sensor module, and a feedback loop. For the speed outer-loop PI controller, the reconstructed Kp value is configured to the proportional gain parameter position, and the Ki value is configured to the integral gain parameter position. For the flux sliding mode controller, the cψ value is configured to the integral sliding surface design, the εψ value is configured to the switching gain parameter position, and the kψ value is configured to the linear gain parameter position. For the torque sliding mode controller, the cT value is configured to the torque sliding surface design, the εT value is configured to the switching gain parameter position, and the kT value is configured to the linear gain parameter position.

[0012] After configuring the controller parameters, the simulation parameters for the motor control model were set. The rated parameters of the permanent magnet synchronous motor were set as follows: stator resistance Rs = 2.875 ohms, stator inductance Ls = 8.5 millihenries, permanent magnet flux linkage ψf = 0.175 Webers, pole pair number pn = 4, moment of inertia J = 0.003 kg / m², and coefficient of friction B = 0.008 N·m / s. The simulation time Tsim was set to 2 seconds, and the simulation step size was set to 0.00001 seconds to ensure numerical calculation accuracy. The reference speed signal was set as a step input, stepping from 0 radians / second to 100 radians / second at time 0 seconds, and a load torque disturbance of 5 N·m was applied at time 1 seconds to test the dynamic response and disturbance rejection performance of the system. The reference flux linkage amplitude was set to 0.5 Webers. After all parameters are configured, the motor control model is started for simulation. The simulation program progresses step by step according to the set time step. Throughout the simulation, the control model executes closed-loop control according to the configured control parameter set to drive the permanent magnet synchronous motor model.

[0013] Step 2.4: During the operation of the motor control model, real-time dynamic response data of the system is collected. Specifically, this includes: real-time collection of key dynamic response data within the 2-second simulation time of the motor control model. The data sampling frequency is set to 10000 Hz, i.e., sampling once every 0.0001 seconds, for a total of 20000 data points collected within the 2-second simulation time. The collected dynamic response data includes the time series of the following variables: actual speed time series, in radians per second, recording the complete speed response process of the motor from start-up to steady state; actual electromagnetic torque time series, in Newton-meters, recording the dynamic changes and steady-state pulsations of torque; actual stator flux linkage amplitude time series, in Webers, recording the tracking process and steady-state pulsations of the flux linkage; stator d-axis current time series and q-axis current time series, in amperes, reflecting the regulation effect of the controller; and the d-axis reference voltage time series and q-axis reference voltage time series output by the controller, in volts, used to evaluate the chattering degree of the sliding mode controller.

[0014] During the data acquisition process, the sampled data at each moment is stored in the corresponding time series array in chronological order. Simultaneously, based on the acquired actual and reference values, the time series of intermediate variables are calculated in real time. The speed error time series is calculated, where the speed error at each moment equals the reference speed of 100 radians per second minus the actual speed value at that moment. The torque error time series is calculated, where the torque error at each moment equals the reference torque output by the PI controller at that moment minus the actual electromagnetic torque value. The flux linkage error time series is calculated, where the flux linkage error at each moment equals the reference flux linkage of 0.5 Weber minus the actual stator flux linkage amplitude. These error signal time series provide the raw data for the subsequent calculation of the time-multiplied absolute error integral index.

[0015] After data collection, the complete dynamic response dataset corresponding to each individual is organized and stored. The dataset includes time series data such as actual rotational speed, torque, flux linkage, current, and reference voltage, as well as calculated time series data for rotational speed error, torque error, and flux linkage error. Each dataset contains 20,000 sampling points. This complete dynamic response data provides a sufficient data foundation for calculating the stage-aware integrated fitness value in step 3. For each individual in the population, steps 2.2 to 2.4 are repeated. After data collection for all 30 individuals, 30 complete dynamic response datasets are obtained, preparing the data for the first iteration of fitness evaluation.

[0016] Step 3: Based on dynamic response data, extract the characteristic indicators of the system at different response stages, and dynamically adjust the fitness evaluation strategy in combination with the characteristic indicators to calculate the individual's stage-perception comprehensive fitness value. Step 4: In the iterative evolution of the mixed whale population, an evolutionary stagnation factor is introduced, and the nonlinear dynamic transition probability is jointly calculated based on the evolutionary stagnation factor and the current iterative state. Step 5: When the random judgment value is less than the nonlinear dynamic transformation probability, control the individual to enter the wide-area exploration mode, and update the individual's position vector based on the beluga whale swimming mechanism and combined with the dimensional scaling factor of the multi-scale search space. Step 6: When the random judgment value is greater than or equal to the nonlinear dynamic conversion probability, control the individual to enter the local development mode, and update the individual's position vector based on the whale spiral attack mechanism and combined with the dynamic convergence factor. Step 7: After each location update, identify inferior individuals in the population based on the stage-aware comprehensive fitness value, perform a directional perturbation whale fall mechanism on the inferior individuals to generate reset individuals, and replace the corresponding inferior individuals with the reset individuals. Step 8: Repeat the above iterative evolution process until the termination condition is met, output the target control parameter set corresponding to the global optimal position vector, and configure the target control parameter set into the permanent magnet synchronous motor sliding mode direct torque control system to drive the motor.

[0017] In this embodiment of the invention, the invention employs a multi-scale search space to construct a multi-scale search space by acquiring the set of parameters to be optimized in the sliding mode direct torque control system of a permanent magnet synchronous motor and performing sensitivity evaluation. In this multi-scale space, a hybrid whale population is initialized, and a motor model is run based on individual position vectors to collect dynamic response data. Based on the dynamic response data, characteristic indicators of different response stages are extracted to dynamically adjust the fitness evaluation strategy. An evolutionary stagnation factor is introduced during iterative evolution to jointly calculate the nonlinear dynamic transition probability. Based on the transition probability, individuals are controlled to enter a wide-area exploration mode or a local development mode. A directional perturbation whale fall mechanism is implemented for inferior individuals. Finally, the target control parameter set corresponding to the globally optimal position vector is output and configured into the motor control system. This technical approach effectively overcomes the technical problems in existing permanent magnet synchronous motor sliding mode direct torque control systems, such as the difficulty of manual parameter tuning due to high-dimensional coupling of multiple parameters, the tendency of traditional optimization algorithms to get trapped in local optima and insufficient convergence accuracy, and the lack of an adaptive parameter tuning mechanism leading to large system torque ripple and low flux tracking accuracy. This achieves the technical effects of efficient global optimization and adaptive tuning of control parameters, effectively breaking through the bottleneck of local optima to improve algorithm convergence accuracy, significantly reducing system torque ripple and high-frequency chattering, and improving flux tracking accuracy and dynamic response performance.

[0018] In a preferred embodiment of the present invention, the above-mentioned sensitivity evaluation of each parameter dimension in the parameter set to be optimized, and the decoupling mapping of each parameter dimension based on the evaluation results to construct a multi-scale search space, includes: Step 1.1: Apply preset perturbations to the proportional gain and integral gain of the speed outer loop, and the sliding surface coefficient, switching gain, and linear gain of the torque inner loop and flux linkage inner loop in the parameter set to be optimized. Collect the rate of change of the output variables of the motor control model as a sensitivity index. Specifically, in the sliding mode direct torque control system of a permanent magnet synchronous motor, first identify the eight key control parameters in the parameter set to be optimized, including the proportional gain Kp and integral gain Ki of the speed outer loop, the sliding surface coefficient cT, switching gain εT, and linear gain kT of the torque inner loop, and the sliding surface coefficient cψ, switching gain εψ, and linear gain kψ of the flux linkage inner loop. For each control parameter, apply positive and negative preset perturbations based on its initial setpoint. The amplitude of the preset perturbation is set to 5% of the initial value to ensure that the perturbation is small enough to maintain system stability, and large enough to produce observable output changes. For each parameter perturbation, the perturbed parameter values ​​are configured into a preset motor control model. This model is constructed based on the mathematical model of a permanent magnet synchronous motor, including the stator voltage equation, electromagnetic torque equation, and a complete implementation of the sliding mode direct torque controller in the dq rotating coordinate system. The motor control model is run, a standard load step signal is applied, and the changes in system output variables, including electromagnetic torque Te, stator flux linkage amplitude ψs, and speed ω, before and after the perturbation are collected. The rate of change of each output variable relative to the parameter perturbation is calculated, and a sensitivity index is defined as: Sensitivity Index = (Change in Output Variable) / (Parameter Perturbation). This index reflects the degree of influence of the control parameter on system performance; a larger sensitivity index value indicates a more significant impact of the parameter on the system output. The sensitivity indices for electromagnetic torque, stator flux linkage, and speed are calculated for each of the eight control parameters, and an 8×3-dimensional sensitivity matrix is ​​constructed to provide a quantitative basis for subsequent parameter grouping.

[0019] Step 1.2: Based on the sensitivity index, the set of parameters to be optimized is divided into high-sensitivity parameter combinations and low-sensitivity parameter combinations. Specifically, this includes: receiving the calculated sensitivity matrix; calculating the comprehensive sensitivity index for each control parameter using a weighted average method; and weighting the sensitivity indices of this parameter to electromagnetic torque, stator flux linkage, and rotational speed with weighted coefficients set to 0.5, 0.3, and 0.2, respectively, reflecting the dominant role of torque performance in sliding mode direct torque control. Based on the numerical distribution of the comprehensive sensitivity index, the K-means clustering algorithm is used to divide the eight control parameters into two combinations: high-sensitivity parameter combinations and low-sensitivity parameter combinations. During the clustering process, the comprehensive sensitivity index is selected as the clustering feature, the number of clusters is set to 2, and the cluster centers are determined through iterative calculation. Each parameter is then assigned to the combination belonging to the nearest cluster center. Cluster analysis revealed that the outer loop proportional gain Kp, integral gain Ki, and inner loop switching gain εT of the torque were classified as high-sensitivity parameter combinations, significantly affecting the system's dynamic response and torque ripple. Conversely, the inner loop sliding mode coefficient cT and linear gain kT of the torque, and the inner loop sliding mode coefficient cψ, switching gain εψ, and linear gain kψ of the flux linkage were classified as low-sensitivity parameter combinations, with relatively milder impacts on system performance. This parameter combination classification provides a scientific basis for the subsequent design of differentiated search strategies.

[0020] The outer loop proportional gain Kp of the speed control provides an immediate control response based on the magnitude of the speed error. A larger Kp results in a faster response, but excessive Kp can lead to overshoot and oscillation. The outer loop integral gain Ki eliminates the steady-state speed error. A larger Ki results in higher steady-state accuracy, but excessive Ki can reduce system stability. The inner loop sliding surface coefficient cT determines the weight of the torque error integral in the sliding surface. A larger cT results in a faster system speed to reach the sliding surface, but may increase chattering. The inner loop switching gain εT controls the switching speed of the sliding mode approaching law to overcome uncertainties and disturbances. A larger εT results in stronger disturbance rejection but more severe chattering. The inner loop linear gain kT provides linear feedback of the sliding mode approaching law to weaken chattering. A larger kT results in faster approach speed and less chattering. The inner loop sliding surface coefficient cψ, switching gain εψ, and linear gain kψ have similar functions to the corresponding parameters in the inner loop of the torque control, controlling the integral weight of the flux error, the disturbance rejection capability of the flux control, and the chattering suppression effect, respectively.

[0021] Step 1.3 involves configuring a first search step size interval for high-sensitivity parameter combinations and a second search step size interval for low-sensitivity parameter combinations. The upper limit of the step size in the first search step size interval is smaller than the upper limit of the step size in the second search step size interval. Specifically, this includes designing differentiated search step size configuration strategies based on the different characteristics of high-sensitivity and low-sensitivity parameter combinations. For high-sensitivity parameter combinations, since these parameters have a significant impact on system performance, even small changes in these parameters can lead to large fluctuations in system performance. Therefore, a smaller search step size is needed to achieve refined optimization. The first search step size interval is configured with an upper limit set to 2% of the parameter value range and a lower limit set to 0.1% of the parameter value range. This ensures that high-sensitivity parameters are searched with small steps during optimization, avoiding missing local optima or causing system instability due to excessively large step sizes. For low-sensitivity parameter combinations, since these parameters have a relatively mild impact on system performance, a larger search step size is needed to improve optimization efficiency and quickly cover a broad parameter space. A second search step size interval is configured, with the upper limit set to 10% of the parameter value range and the lower limit set to 1%. This allows low-sensitivity parameters to explore with larger step sizes during optimization, improving global search capabilities. Through differentiated search step size configurations, the search efficiency of low-sensitivity parameters is improved while ensuring refined optimization of high-sensitivity parameters, achieving a balance in overall optimization performance.

[0022] Step 1.4 maps high-sensitivity parameter combinations and low-sensitivity parameter combinations to different dimensional intervals within the same search space, generating a multi-scale search space. Within this multi-scale search space, dimensional scaling factors corresponding to either the first or second search step size interval are assigned to different dimensional intervals. Specifically, this includes: constructing a unified multi-scale search space containing eight dimensions, each corresponding to one of the eight control parameters in the parameter set to be optimized. The dimensions corresponding to the high-sensitivity parameter combinations are defined as the refined search dimensional intervals, and the dimensions corresponding to the low-sensitivity parameter combinations are defined as the wide-area search dimensional intervals, achieving decoupling mapping of different parameter combinations within the same search space. For the refined search dimensional intervals, a dimensional scaling factor is configured based on the first search step size interval, with a value between 0.02 and 0.1. This factor is used to adjust the step size of individual position updates during algorithm iteration, ensuring that high-sensitivity parameters update their positions with small steps. For the wide-area search dimensional intervals, a dimensional scaling factor is configured based on the second search step size interval, with a value between 0.1 and 0.5. This factor allows low-sensitivity parameters to update their positions with larger steps, improving search efficiency. In the multi-scale search space, each dimension carries a corresponding dimension scaling factor attribute, which is invoked in the position update calculation of the hybrid whale optimization algorithm to achieve adaptive step size adjustment for different parameter dimensions. The construction of the multi-scale search space effectively solves the problem that the traditional single-scale search space cannot simultaneously meet the optimization needs of different sensitivity parameters, providing a spatial foundation for subsequent efficient optimization.

[0023] In this embodiment of the invention, by applying preset perturbations to each parameter in the parameter set to be optimized and collecting the rate of change of the output variable of the motor control model as a sensitivity index, the parameters are divided into high-sensitivity parameter combinations and low-sensitivity parameter combinations based on the sensitivity index. Different search step size intervals are configured for different combinations, and the two sets of parameters are mapped to different dimensional intervals of the same search space and assigned corresponding dimensional scaling factors to generate a multi-scale search space. Therefore, this invention overcomes the technical problems of existing parameter optimization methods, such as the inability to take into account the optimization needs of different sensitivity parameters when using a uniform search step size, insufficient accuracy due to excessively large step sizes for high-sensitivity parameters and low efficiency due to excessively small step sizes for low-sensitivity parameters, and difficulty in searching high-dimensional parameter spaces due to the lack of parameter dimension decoupling mechanisms. Thus, it achieves the technical effect of realizing differentiated search strategies based on parameter sensitivity differences, ensuring both refined optimization of high-sensitivity parameters and efficient exploration of low-sensitivity parameters through the construction of a multi-scale search space, and effectively improving the convergence speed and optimization accuracy of the hybrid whale optimization algorithm in the optimization of control parameters of permanent magnet synchronous motors.

[0024] In a preferred embodiment of the present invention, the above-mentioned extraction of feature indicators of the system at different response stages based on dynamic response data, and dynamic adjustment of the fitness evaluation strategy in combination with the feature indicators to calculate the individual's stage-perception comprehensive fitness value, includes: Step 3.1: Identify the current response stage based on the speed error change rate in the dynamic response data. The current response stage includes a dynamic adjustment stage or a steady-state maintenance stage. Specifically, this includes: receiving the collected dynamic response data, which contains time series data of the motor's speed ω, electromagnetic torque Te, and stator flux linkage amplitude ψs during operation at a given reference speed. Calculate the speed error eω as the difference between the reference speed ωref and the actual speed ω, i.e., eω = ωref - ω. Perform numerical differentiation on the speed error time series to obtain the speed error change rate, which reflects the trend of speed error change over time. Set a response stage identification threshold. When the absolute value of the speed error change rate is greater than a preset dynamic threshold (e.g., set to 10 rad / s²), the system is determined to be in the dynamic adjustment stage. This stage is characterized by rapid changes in speed error, and the system is transitioning from an initial or disturbed state to a stable operating state. At this time, the system's dynamic response performance (such as rise time and overshoot) becomes the main focus. When the absolute value of the rate of change of speed error is less than or equal to the dynamic threshold, the system is determined to be in the steady-state maintenance stage. This stage is characterized by slow changes in speed error, indicating that the system has approached or reached a stable operating state. At this time, the steady-state performance of the system (such as torque pulsation and flux linkage pulsation) becomes the main focus. Accurate identification of the response stage provides a stage basis for the dynamic adjustment of the subsequent fitness evaluation strategy, ensuring that the optimization process can accurately evaluate the characteristic requirements of different operating stages.

[0025] Step 3.2: In the dynamic adjustment phase, overshoot and rise time features are extracted as primary indicators. The dynamic weighting coefficients of the speed error term and overshoot penalty term in the fitness function are increased. Specifically, when the system is identified as being in the dynamic adjustment phase, the dynamic response performance evaluation process is initiated. Overshoot features are extracted from the dynamic response data. Overshoot OSω is defined as the ratio of the difference between the maximum value of the actual speed exceeding the reference speed and the reference speed. The calculation formula is: OSω=(ωmax-ωref) / ωref×100%, where ωmax is the peak value of the actual speed during the dynamic process. Overshoot features reflect the stability of the system's dynamic response; excessive overshoot may lead to motor overcurrent or mechanical damage. Rise time features are extracted. Rise time tr is defined as the time required for the speed to rise from 10% to 90% of the reference speed. This feature reflects the system's response speed. Based on the extracted overshoot and rise time features, the weighting coefficients in the fitness function are dynamically adjusted. Specifically, the fitness function includes a speed error integral term ∫|eω|dt, an overshoot penalty term λ·OSω, a torque error term ∫|eT|dt, and a flux linkage error term ∫|eψ|dt. During the dynamic adjustment phase, the dynamic weight coefficient w3 of the speed error term is increased, for example from the base value of 0.05 to 0.15, to enhance the focus on speed tracking performance. Simultaneously, the penalty coefficient λ of the overshoot penalty term is increased, for example from the base value of 8500 to 9500, to strengthen the suppression of overshoot. The dynamic adjustment of the weight coefficients ensures that the optimization algorithm prioritizes optimizing the system's response speed and overshoot suppression performance during the dynamic adjustment phase.

[0026] Step 3.3: During the steady-state maintenance phase, extract torque ripple characteristics and flux linkage ripple characteristics as the dominant indicators, and increase the dynamic weight coefficients of the torque error term and flux linkage error term in the fitness function. Specifically, when the system is identified as being in the steady-state maintenance phase, initiate the steady-state performance evaluation process. Extract torque ripple characteristics from the dynamic response data. Torque ripple is defined as the peak-to-peak value of the actual electromagnetic torque during the steady-state phase, and the calculation formula is: ,in and These represent the maximum and minimum values ​​of the electromagnetic torque during the steady-state phase. Torque ripple characteristics reflect the smoothness of the system's steady-state operation; large torque ripples can lead to mechanical vibration and noise. The flux linkage ripple characteristic is extracted; flux linkage ripple is defined as the peak-to-peak value of the stator flux linkage amplitude during the steady-state phase, and the calculation formula is: ,in and These represent the maximum and minimum values ​​of the stator flux linkage amplitude during the steady-state phase. Flux linkage pulsation characteristics reflect flux linkage tracking accuracy; smaller pulsation indicates better flux linkage control. Based on the extracted torque pulsation and flux linkage pulsation characteristics, the weighting coefficients in the fitness function are dynamically adjusted. Specifically, during the steady-state maintenance phase, the dynamic weighting coefficient w2 of the torque error term is increased, for example, from the base value of 5 to 8, to enhance the focus on torque pulsation suppression; simultaneously, the dynamic weighting coefficient w1 of the flux linkage error term is increased, for example, from the base value of 500 to 800, to strengthen the emphasis on flux linkage tracking accuracy. This dynamic adjustment of the weighting coefficients ensures that the optimization algorithm prioritizes optimizing the system's torque stability and flux linkage tracking performance during the steady-state maintenance phase.

[0027] Step 3.4: Based on the dynamic weight coefficients corresponding to the current response stage, the integral of time multiplied by absolute error (ITAE) index is weighted and calculated, and combined with the overshoot penalty term to generate the stage-aware comprehensive fitness value. Specifically, this includes: receiving the identification result of the current response stage and the corresponding adjusted dynamic weight coefficients, and constructing the stage-aware comprehensive fitness function. The comprehensive fitness function uses the integral of time multiplied by absolute error (ITAE) index as the basic evaluation index. This index can simultaneously consider the error magnitude and error duration. The calculation formula is: ITAE=∫t·|e(t)|dt, where t is time and e(t) is the error value at time t. The corresponding ITAE indices are calculated for speed error, torque error, and flux linkage error respectively, obtaining ITAEω, ITAEeT, and ITAEeψ. Based on the dynamic weight coefficients w3, w2, and w1 corresponding to the current response stage, the three ITAE indices are weighted and summed to calculate the weighted error term: Weighted error = w3·ITAEω + w2·ITAEeT + w1·ITAEeψ. The weighted error term comprehensively reflects the system's performance in the current response phase, and the dynamic adjustment of the weight coefficients ensures that the performance indicators of interest at different stages are given reasonable attention. An overshoot penalty term is further introduced. When speed overshoot is detected (i.e., OSω > 0), a penalty term λ·OSω is superimposed on the weighted error term, where λ is the overshoot penalty coefficient, and its value is dynamically adjusted according to the current response phase. The final formula for calculating the phase-aware comprehensive fitness value is: F(X) = w3·ITAEω + w2·ITAEeT + w1·ITAEeψ + λ·OSω, with the optimization objective being to find the control parameter vector X that minimizes F(X). The phase-aware comprehensive fitness value can dynamically adjust the evaluation focus according to different characteristics of the system's operating phase, providing a precise optimization objective for the hybrid whale optimization algorithm.

[0028] In this embodiment of the invention, the technical means of identifying the current response stage based on the rate of change of speed error in dynamic response data, extracting overshoot and rise time features and increasing the weight coefficients of speed error and overshoot penalty terms in the dynamic adjustment stage, extracting torque pulsation and flux pulsation features and increasing the weight coefficients of torque and flux error terms in the steady-state maintenance stage, and generating a stage-aware comprehensive fitness value based on the weighted calculation of the time multiplied by the absolute error integral index using dynamic weight coefficients and combining it with the overshoot penalty term, overcomes the technical problems of existing fitness functions using fixed weight coefficients that cannot adapt to the performance evaluation needs of different operating stages, the different focus of the dynamic response stage and steady-state operation stage but the single evaluation standard leading to unsatisfactory optimization results, and the lack of a stage-aware mechanism that cannot specifically optimize performance indicators of different stages. This achieves dynamic adaptive adjustment of the fitness evaluation strategy based on response stage identification, prioritizing the optimization of response speed and overshoot suppression performance in the dynamic adjustment stage, prioritizing the optimization of torque pulsation suppression and flux tracking accuracy in the steady-state maintenance stage, and providing accurate multi-stage performance evaluation of the optimization algorithm through the stage-aware comprehensive fitness value, effectively improving the technical effect of parameter optimization in terms of targeting and comprehensiveness.

[0029] In a preferred embodiment of the present invention, the above-mentioned introduction of an evolutionary stagnation factor, and the joint calculation of the nonlinear dynamic transition probability based on the evolutionary stagnation factor and the current iteration state, includes: Step 4.1: Record the number of times the global optimal fitness value of the hybrid whale population has not improved within a preset number of generations. The ratio of the number of times the global optimal fitness value has not improved to the preset number of generations is used as the evolutionary stagnation factor. Specifically, this includes: monitoring the changes in the global optimal fitness value of the population in real time during the iterative evolution of the hybrid whale optimization algorithm. The preset number of generations is set to 20 generations, meaning that the improvement of the global optimal fitness value is continuously tracked during 20 consecutive iterations. After each iteration, the global optimal fitness value Fbest(t) of the current generation is compared with the global optimal fitness value Fbest(t-1) of the previous generation to determine whether an improvement has occurred. Here, the global optimal fitness value refers to the smallest fitness value among all individuals currently discovered during the iterative evolution of the hybrid whale optimization algorithm. The position of the individual corresponding to this value is the global optimal individual position. Fbest(t) represents the global optimal fitness value at the end of the t-th generation iteration, used to calculate the evolutionary stagnation factor and determine the convergence state of the algorithm.

[0030] The improvement criterion is defined as follows: when |Fbest(t)-Fbest(t-1)|>δ, where δ is the improvement threshold (e.g., set to 0.001), effective improvement is determined; otherwise, no improvement is determined. Within a consecutive preset generation time window, the number of times no improvement occurs, Nstag, is accumulated. This counter increments by 1 each time no improvement occurs and resets to 0 when improvement occurs. The evolutionary stagnation factor Sf is calculated, defined as the ratio of the number of times no improvement occurs to the consecutive preset generation: Sf=Nstag / 20. The evolutionary stagnation factor ranges from 0 to 1. When Sf is close to 0, it indicates active population evolution and continuous improvement of the global optimum; when Sf is close to 1, it indicates stagnation in population evolution, with the global optimum not improving for a long time and potentially trapped in a local optimum. The evolutionary stagnation factor provides a quantitative basis for adjusting the algorithm's search strategy and effectively identifies the algorithm's evolutionary state.

[0031] Let's take a specific iteration as an example to illustrate the calculation of the evolutionary stagnation factor: Assume the current iteration is the 180th generation, with a preset generation count of 20. Looking back at the global optimal fitness values ​​of the previous 20 generations: the global optimal fitness value for the 160th generation is 125.6; for the 161st generation it is 125.6 (no improvement); for the 162nd generation it is 124.8 (improvement), with the number of times it didn't improve reset to 0; for the 163rd generation it is 124.8 (no improvement); for the 164th generation it is 124.8 (no improvement), and so on. In the 19 generations from the 162nd to the 180th, assuming improvement only occurred in the 162nd, 168th, and 175th generations, and no improvement in the remaining 16 generations, the cumulative number of times it didn't improve (Nstag) is 16. The evolutionary stagnation factor Sf equals 16 / 20, which equals 0.8. This value is close to 1, indicating a significant stagnation in population evolution, requiring an increase in the transition probability to enhance wide-area exploration capabilities.

[0032] Step 4.2: Obtain the iteration ratio of the current iteration count to the maximum iteration count, and calculate the basic transformation probability based on the iteration ratio. Specifically, this includes: obtaining the current iteration count t and the preset maximum iteration count tmax of the hybrid whale optimization algorithm, with the maximum iteration count set to 500 generations as the algorithm termination condition. Calculate the iteration ratio r = t / tmax, which reflects the current iteration progress and ranges from 0 to 1. In the early stages of iteration, r is close to 0, and in the later stages, r is close to 1. Design a nonlinear basic transformation probability calculation function based on the iteration ratio, using an exponential decay form to ensure that the transformation probability decreases nonlinearly with iteration. The calculation formula is: Pcbase = Pmax - (Pmax - Pmin) × (t / tmax)², where Pmax is the initial maximum transformation probability set to 0.9, and Pmin is the final minimum transformation probability set to 0.2. It should be noted that Pmax refers to the maximum value of the base transition probability Pcbase in the early stage of iteration, i.e., the first iteration. This parameter is set to 0.9, meaning that at the beginning of the optimization algorithm, the generated random decision value r1 has a 90% probability of being less than the transition probability Pc, thus giving the individual a 90% probability of entering the wide-area exploration mode. Pmin refers to the minimum value of the base transition probability Pcbase in the later stage of iteration, i.e., the 500th iteration, the maximum number of iterations. This parameter is set to 0.2, meaning that near the end of the optimization algorithm, there is still a 20% probability of the individual entering the wide-area exploration mode and an 80% probability of entering the local exploration mode.

[0033] The basic transition probability calculation function embodies the algorithm's design philosophy of gradually shifting from wide-area exploration in the early stages of iteration to local development in the later stages: In the early stages of iteration, Pcbase is close to 0.9, and the higher transition probability encourages more individuals to enter wide-area exploration mode, quickly covering the search space; in the later stages of iteration, Pcbase is close to 0.2, and the lower transition probability encourages more individuals to enter local development mode, refining the search for the optimal solution. The basic transition probability provides the foundational components for subsequent nonlinear dynamic transition probability calculations.

[0034] Step 4.3 involves using the evolutionary stagnation factor as a stagnation penalty term. This penalty term is then superimposed on the base transition probability to generate a nonlinear dynamic transition probability. A larger evolutionary stagnation factor results in a higher nonlinear dynamic transition probability, increasing the probability of entering a wide-area exploration mode. Specifically, this includes receiving the calculated evolutionary stagnation factor Sf and the base transition probability Pcbase, and designing an adaptive transition probability adjustment mechanism. The evolutionary stagnation factor, as a stagnation penalty term, reflects the degree of stagnation in population evolution. When the stagnation degree is high, it is necessary to enhance global exploration capabilities to escape local optima. The nonlinear dynamic transition probability calculation formula is designed as: Pc = Pcbase + α × Sf × (1 - Pcbase), where α is the stagnation response coefficient set to 0.5. This coefficient controls the influence of the evolutionary stagnation factor Sf on the transition probability. In the formula, the Sf×(1-Pcbase) term ensures that the increment of the stagnation penalty increases as the basic transition probability decreases. In the later stages of iteration, when the basic transition probability is low, if evolutionary stagnation occurs, it can significantly increase the transition probability and strengthen wide-area exploration. When the evolutionary stagnation factor Sf is large (close to 1), it indicates that the population has not improved in the long term, and the nonlinear dynamic transition probability Pc increases significantly, causing more individuals to enter the wide-area exploration mode and break through local optima by exploring new search areas. When the evolutionary stagnation factor Sf is small (close to 0), it indicates that the population is actively evolving, and the nonlinear dynamic transition probability Pc is mainly determined by the basic transition probability, with a balance between exploration and development according to the normal iteration progress. The final generated nonlinear dynamic transition probability Pc comprehensively considers the iteration progress and evolutionary state, providing the algorithm with an adaptive search mode switching mechanism.

[0035] In this embodiment of the invention, by recording the number of times the global optimal fitness value of the hybrid whale population has not improved within a consecutive preset number of generations and calculating the evolutionary stagnation factor, obtaining the iteration ratio of the current iteration number to the maximum iteration number and calculating the basic transition probability, and superimposing the evolutionary stagnation factor as a stagnation penalty term with the basic transition probability to generate a nonlinear dynamic transition probability, the technical problem of existing intelligent optimization algorithms using fixed or only iteration-dependent transition probabilities, being unable to dynamically adjust according to the population evolution state leading to difficulty in escaping local optima, and lacking an evolutionary stagnation identification mechanism to enhance global exploration capabilities in a timely manner is overcome. This achieves the technical effect of real-time perception of the population evolution state based on the evolutionary stagnation factor, adaptively increasing the transition probability to enhance wide-area exploration capabilities when evolutionary stagnation occurs, effectively preventing the algorithm from falling into local optima and improving the ability to escape local optima, and realizing an intelligent balance between exploration and development through nonlinear dynamic transition probabilities, significantly improving the global optimization performance and convergence reliability of the hybrid whale optimization algorithm.

[0036] In a preferred embodiment of the present invention, when the random determination value is less than the nonlinear dynamic transition probability, the individual is controlled to enter the wide-area exploration mode, and the position vector of the individual is updated based on the beluga whale swimming mechanism combined with the dimension scaling factor of the multi-scale search space, comprising: Step 5.1: When the random determination value is less than the nonlinear dynamic transition probability, a reference individual position is randomly selected from the current population, which specifically includes: in each iteration process, for each individual in the hybrid whale population, generating a random determination value r1 within the interval [0, 1], and the random value is used to determine the search mode adopted by the individual. The generated random determination value r1 is compared with the calculated nonlinear dynamic transition probability Pc. When r1<Pc, it is determined that the individual enters the wide-area exploration mode in the current iteration, and the beluga whale swimming mechanism is adopted for position update. In the current population, excluding the individual itself, one individual is randomly selected from all the remaining individuals as a reference individual, and the position vector Xrand(t) of the reference individual is recorded. The position vector Xrand(t) refers to the position vector of the reference individual randomly selected from the current population in the t-th iteration, and the position vector is a parameter vector containing 8 dimensions, corresponding to 8 control parameters to be optimized. The random selection mechanism ensures that the wide-area exploration mode can make full use of the position information of different individuals in the population, enhance the diversity of search, and avoid the loss of population diversity caused by all individuals gathering towards the global optimal position. The reference individual position Xrand(t) provides a spatial reference point for the subsequent beluga whale swimming position update calculation.

[0037] When r1≥Pc, it is determined that the individual enters the local exploitation mode in the current iteration, and the whale spiral attack mechanism is adopted for position update. This mechanism guides the current individual to perform refined search near the optimal solution based on the global optimal individual position of the current population.

[0038] Step 5.2: Obtain the dimension scaling factor corresponding to the dimension interval where the individual is located. Use the dimension scaling factor as the adaptive step size adjustment coefficient. Specifically, this includes: reading the position vector of the current individual Xi(t) from the multi-scale search space. This vector contains 8 dimensional components, each corresponding to a parameter to be optimized. Based on the parameter type corresponding to each dimension, identify whether the dimension belongs to the refined search dimension interval or the wide-area search dimension interval. For dimensions belonging to the refined search dimension interval (corresponding to high-sensitivity parameter combinations), obtain the dimension scaling factor configured for that dimension, with a value range of 0.02 to 0.1; for dimensions belonging to the wide-area search dimension interval (corresponding to low-sensitivity parameter combinations), obtain the dimension scaling factor configured for that dimension, with a value range of 0.1 to 0.5. Denote the obtained dimension scaling factor as a vector S=[s1,s2,...,s8], where each component si corresponds to the scaling factor of the i-th dimension. In the location update calculation of the wide-area exploration mode, the dimension scaling factor vector S is used as the adaptive step size adjustment coefficient. This coefficient is used to adjust the location update amplitude of different dimensions, ensuring that the high-sensitivity parameter dimension is explored with a smaller step size and the low-sensitivity parameter dimension is explored with a larger step size, so as to realize the differentiated exploration strategy of different parameter dimensions.

[0039] Step 5.3: Based on the reference individual's position, the individual's current position, and the adaptive step size adjustment coefficient, calculate the individual's swimming displacement in the multi-scale search space. Specifically, this includes: using the beluga whale swimming mechanism to calculate the individual's position update vector. The beluga whale swimming mechanism is based on the exploration strategy of the beluga whale optimization algorithm, achieving wide-area coverage of the search space by simulating the swimming behavior of a beluga whale. The basic formula for calculating the swimming displacement is: ΔX=α1×S⊙(Xrand(t)-Xi(t)), where α1 is the step size control coefficient set to 0.3, S is the dimension scaling factor vector, Xrand(t) is the reference individual's position vector, Xi(t) is the current individual's position vector, and ⊙ represents element-wise multiplication (Hadamard product). In the formula, (Xrand(t)-Xi(t)) calculates the position difference vector between the current individual and the reference individual, which indicates the direction of wide-area exploration. S⊙(Xrand(t)-Xi(t)) multiplies each dimension of the position difference vector by the corresponding dimension scaling factor to achieve differentiated step size adjustment for different dimensions. The displacement of dimensions with high sensitivity parameters is smaller, while the displacement of dimensions with low sensitivity parameters is larger. α1×S⊙(Xrand(t)-Xi(t)) further adjusts the overall amplitude of the swimming displacement uniformly through the step size control coefficient α1. The calculated swimming displacement ΔX is an 8-dimensional vector, where each dimension component represents the position change of the corresponding parameter in wide-area exploration mode.

[0040] Taking the position update of an individual in wide-area exploration mode as an example, the calculation process is explained as follows: The current individual's position vector is 35.2, 120.5, 55.3, 210.7, 450.3, 48.9, 185.6, 380.7. The randomly selected reference individual's position vector is 42.8, 95.3, 68.5, 195.2, 520.6, 55.7, 210.3, 425.5. First, the position difference vector is calculated. The position difference in the first dimension is 7.6, the position difference in the second dimension is -25.2, the position difference in the third dimension is 13.2, and so on, resulting in position difference vectors of 7.6, -25.2, 13.2, -15.5, 70.3, 6.8, 24.7, 44.8. The preset dimension scaling factor vector is 0.1, 0.1, 0.3, 0.1, 0.3, 0.3, 0.1, 0.3. Multiplying each component of the position difference vector by its corresponding scaling factor, the adjusted displacement is 0.76 for the first dimension, -2.52 for the second, 3.96 for the third, and so on, resulting in adjusted displacement vectors of 0.76, -2.52, 3.96, -1.55, 21.09, 2.04, 2.47, and 13.44. Finally, multiplying the entire vector by a step size control coefficient of 0.3 yields the final floating displacement vectors of 0.228, -0.756, 1.188, -0.465, 6.327, 0.612, 0.741, and 4.032. This floating displacement is then superimposed on the current position to obtain the updated position.

[0041] Step 5.4 involves superimposing the displacement amount with the individual's current position to generate an updated position vector for the individual in wide-area exploration mode. Specifically, this includes receiving the calculated displacement amount ΔX and the individual's current position vector Xi(t), and performing a position update operation. The calculation formula is: Xi(t+1) = Xi(t) + ΔX. This formula superimposes the displacement amount onto the current position, generating a new position vector Xi(t+1) for the individual in the next iteration. After the position update, boundary condition checks are performed to ensure that all components of the updated position vector are within the preset parameter value range. If a component exceeds the upper bound Xub, it is restricted to the upper bound value Xub; if a component is below the lower bound Xlb, it is restricted to the lower bound value Xlb. The boundary values ​​are set to 100 and -100. Boundary handling ensures that all individual positions remain within the valid search space, avoiding physically infeasible parameter combinations. After boundary processing, the updated position vector Xi(t+1) is used as the final updated position of the individual in the wide-area exploration mode. This position will be used in the next iteration for motor control model simulation and fitness evaluation. By using the beluga whale swimming mechanism combined with the wide-area exploration position update using the dimension scaling factor, the individual can perform adaptive wide-area exploration in a multi-scale search space, achieving efficient wide-area coverage of low-sensitivity parameters while maintaining fine-grained search for high-sensitivity parameters.

[0042] In this embodiment of the invention, because a reference individual position is randomly selected from the current population when the random judgment value is less than the nonlinear dynamic transformation probability, and the dimension scaling factor corresponding to the dimension interval where the individual is located is obtained as the adaptive step size adjustment coefficient, the swimming displacement is calculated based on the reference individual position, the current position of the individual, and the adaptive step size adjustment coefficient, and the swimming displacement is superimposed with the current position to generate the updated position vector in the wide-area exploration mode, the technical means of overcoming the existing intelligent optimization algorithm's inability to adapt to the sensitivity differences of different parameter dimensions when using a uniform step size in wide-area exploration, the decrease in search accuracy due to excessively large step sizes in wide-area exploration of high-sensitivity parameters, and the difficulty in balancing efficiency and accuracy in wide-area exploration due to the lack of a multi-scale search mechanism are achieved. Thus, the technical effect of achieving adaptive differentiated wide-area exploration based on the beluga whale swimming mechanism and combined with the dimension scaling factor of the multi-scale search space is achieved, ensuring both the search accuracy of high-sensitivity parameters and the search efficiency of low-sensitivity parameters during the wide-area exploration process, effectively improving the search capability and adaptability of the hybrid whale optimization algorithm in the wide-area exploration mode.

[0043] In a preferred embodiment of the present invention, the above-mentioned control of an individual to enter a local development mode when the random determination value is greater than or equal to the nonlinear dynamic transformation probability, and updating the individual's position vector based on the whale spiral attack mechanism and combined with a dynamic convergence factor, includes: Step 6.1: When the random decision value is greater than or equal to the nonlinear dynamic transition probability, obtain the global optimal individual position of the current population. Specifically, this includes: In each iteration, for each individual in the mixed whale population, when the generated random decision value r1 is compared with the nonlinear dynamic transition probability Pc and r1 ≥ Pc, the individual is determined to enter the local development mode in the current iteration, and the position is updated using the whale spiral attack mechanism. From all individuals in the current population, based on the stage-aware comprehensive fitness value evaluation result, the individual with the smallest fitness value (i.e., the best performance) is selected as the global optimal individual, and its position vector Xbest(t) is recorded. The position vector contains 8 dimensional components, corresponding to the optimal control parameter combination found in the current iteration. The global optimal individual position represents the optimal solution region found by the current population in the search space. In the local development mode, individuals achieve refined search of local regions by moving towards the global optimal position. The obtained global optimal individual position Xbest(t) serves as the target attraction point for local development, guiding individuals to perform spiral encirclement search.

[0044] Step 6.2 calculates the spatial distance between the individual's current position and the global optimal individual position, and constructs a dynamic convergence factor that decreases with each iteration based on the current iteration number. Specifically, this includes calculating the Euclidean distance between the individual's current position vector Xi(t) and the global optimal individual position vector Xbest(t), defining the spatial distance vector as: Di = |Xbest(t) - Xi(t)|, where each dimension component represents the distance between the individual and the global optimal solution in the corresponding parameter dimension. The spatial distance Di reflects the closeness of the individual to the currently known optimal solution; the smaller the distance, the closer the individual is to the optimal solution region. A dynamic convergence factor is constructed based on the current iteration number t. This factor controls the convergence speed of local development, and the formula for calculating the dynamic convergence factor that decreases with each iteration is: A = 2 × (1 - t / Tmax), where Tmax is the maximum iteration number of 500. In the early stages of iteration, when t is small, A is close to 2, and a larger convergence factor allows for a larger range of local searches; in the later stages of iteration, when t is close to Tmax, A is close to 0, and a smaller convergence factor shrinks the local search range, strengthening refined development. The decreasing trend of the dynamic convergence factor A ensures that the algorithm gradually shifts from a large-scale local search to a fine search near the optimal solution during the iteration process, thereby improving the convergence accuracy.

[0045] Step 6.3, based on the dynamic convergence factor and the logarithmic spiral constant, calculate the spiral encirclement displacement of the individual towards the globally optimal individual position. Specifically, this includes: using a whale spiral attack mechanism to calculate the spiral encirclement displacement of the individual. This mechanism simulates the spiral bubble net predation behavior of humpback whales, achieving high-precision local exploitation by approaching the target position through a spiral trajectory. Set the logarithmic spiral constant b=1, which defines the shape characteristics of the spiral. Generate a random number l in the interval [-1, 1] to introduce the randomness of the spiral search. Calculate the spiral encirclement displacement vector: ΔXspiral=Di⊙exp(b·l)⊙cos(2π·l)-Di⊙A, where Di is the spatial distance vector, A is the dynamic convergence factor, exp(b·l) and cos(2π·l) constitute the logarithmic spiral function, and ⊙ represents element-wise multiplication. In the formula, exp(b·l)⊙cos(2π·l) generates the spiral coefficient, which spirals within the range [-e, e] as the random number l changes. Di⊙exp(b·l)⊙cos(2π·l) applies the spiral coefficient to each dimension of the spatial distance vector, achieving spiral encirclement search in different parameter dimensions. Subtracting the Di⊙A term introduces a dynamic convergence factor, which, as iteration progresses, causes A to decrease, leading to the displacement vector gradually shrinking towards the global optimum. The calculated spiral encirclement displacement ΔXspiral combines the exploratory nature of spiral search with the developmental nature of dynamic convergence, achieving efficient approximation to the optimal solution.

[0046] The calculation process is illustrated using the example of updating the position of an individual in a local development mode: The current iteration count t is 300, the maximum iteration count Tmax is 500, and the current individual position vectors are 38.5, 135.2, 62.3, 225.8, 480.5, 52.6, 198.7, and 405.3. The globally optimal individual position vectors are 41.2, 128.6, 58.7, 218.3, 465.2, 50.1, 205.4, and 395.8. First, calculate the spatial distance vectors: the first dimension distance is 2.7, the second dimension distance is 6.6, and so on, resulting in distance vectors of 2.7, 6.6, 3.6, 7.5, 15.3, 2.5, 6.7, and 9.5. Calculate the dynamic convergence factor A: the iteration ratio is 300 divided by 500, which equals 0.6; the remaining iteration ratio is 0.4; therefore, the dynamic convergence factor A equals 0.4 multiplied by 2, which equals 0.8. The random number generator l is set to 0.3, and the logarithmic spiral constant b is set to 1. The spiral coefficient is calculated: the exponential function value is e raised to the power of 0.3, approximately 1.35; the cosine function angle is 0.3 × 2π ≈ 1.88 radians, the cosine value is approximately -0.29, and the spiral coefficient is 1.35 × -0.29, approximately -0.39. The spiral displacement term is calculated: the first dimension is 2.7 × -0.39 ≈ -1.05, the second dimension is 6.6 × -0.39 ≈ -2.57, and so on. The convergence adjustment term is calculated: the first dimension is 2.7 × 0.8 = 2.16, the second dimension is 6.6 × 0.8 = 5.28, and so on. The final spiral encircling displacement has a first dimension of -1.05 - 2.16 = -3.21, a second dimension of -2.57 - 5.28 = -7.85, and so on, to obtain the complete spiral encircling displacement vector. This vector is superimposed with the global optimal position to obtain the updated position.

[0047] Step 6.4 combines the spiral encircling displacement with the globally optimal individual position to generate the updated position vector of the individual in the local development mode. Specifically, this includes: based on the calculated spiral encircling displacement ΔXspiral and the globally optimal individual position Xbest(t), performing a local development position update, calculated as: Xi(t+1) = Xbest(t) + ΔXspiral. This formula uses the globally optimal individual position as a reference, superimposed with the spiral encircling displacement, to generate the new position vector Xi(t+1) for the individual in the next iteration. This position update method ensures that the individual always performs a spiral search around the globally optimal position, achieving high-precision development of the local region. After the position update, a boundary condition check is performed to ensure that all components of the updated position vector are within the preset parameter value range. The same boundary processing method as in the wide-area exploration mode is used to limit components exceeding the boundary to the range [Xlb,Xub]. After boundary processing, the updated position vector Xi(t+1) is used as the final updated position of the individual in the local development mode. This position will be used for the motor control model simulation and fitness evaluation in the next iteration. By employing the whale spiral attack mechanism and combining it with local exploitation position updates based on dynamic convergence factors, individuals can perform efficient spiral searches near the global optimum, gradually converging to the optimum as iterations proceed, thus achieving high-precision local exploitation.

[0048] In this embodiment of the invention, the technical means of obtaining the global optimal individual position of the current population when the random judgment value is greater than or equal to the nonlinear dynamic transformation probability, calculating the spatial distance between the current individual position and the global optimal position, constructing a dynamic convergence factor that decreases with the number of iterations based on the number of iterations, calculating the spiral encirclement displacement based on the dynamic convergence factor and the logarithmic spiral constant, and combining the spiral encirclement displacement with the global optimal position to generate the updated position vector in the local development mode, overcomes the technical problems of existing intelligent optimization algorithms that use a fixed convergence speed in local development, which cannot adapt to the development needs of different iteration stages, have insufficient convergence speed in the later stages of iteration leading to decreased accuracy, and lack a dynamic convergence mechanism that cannot achieve a smooth transition from coarse search to fine search. Thus, it achieves the technical effect of adaptive local development based on the whale spiral attack mechanism and combined with a dynamic convergence factor that decreases with iteration, maintaining a large search range in the early stage of iteration to avoid premature convergence, and strengthening fine search in the later stage of iteration to improve convergence accuracy, effectively improving the development efficiency and convergence accuracy of the local development mode of the hybrid whale optimization algorithm.

[0049] In a preferred embodiment of the present invention, the process of identifying inferior individuals in the population based on the stage-aware comprehensive fitness value after each location update, and performing a directional perturbation whale fall mechanism on these inferior individuals to generate reset individuals, includes: Step 7.1: After each position update, sort the individuals in the population according to their stage-aware comprehensive fitness value. Identify the individuals with fitness values ​​at the bottom of a predetermined proportion as inferior individuals. Specifically, after the mixed whale population completes the position update for the current iteration, perform performance evaluation based on the stage-aware comprehensive fitness value F(Xi) of each individual. Sort all individuals in the population in ascending order of fitness value, with smaller fitness values ​​indicating better control performance and larger fitness values ​​indicating worse control performance. Set the proportion of inferior individuals to be identified as 15% of the population size. For example, when the population size is 30 individuals, identify the last 5 individuals as inferior individuals. The specific identification method is as follows: calculate the number of inferior individuals Nbad = ⌈N × 0.15⌉, where N is the population size and ⌈⌉ represents rounding up; in the sorted individual sequence, select the individuals with fitness values ​​at the bottom Nbad positions and mark them as inferior individuals. These suboptimal individuals represent the worst-performing individuals in the current population. The control parameter combinations corresponding to their position vectors result in poor motor control performance, contributing limited value to the optimization process while consuming significant computational resources. Identifying these suboptimal individuals provides the processing targets for the subsequent directional perturbation whale fall mechanism.

[0050] Step 7.2: For suboptimal individuals, obtain the current global best position and the historical best position of the suboptimal individual. Specifically, this includes: for each individual Xi identified as a suboptimal individual, performing a historical information extraction operation. Obtain the historical best position Xi reached by the individual throughout the entire iteration process from its historical records. pbest The historical best position is the position vector with the minimum fitness value for an individual across all iterations, representing the individual's best historical experience. The historical best position contains information about local advantages discovered by the individual during evolution; even if the individual's current performance is poor, its historical best position may contain parameter combinations valuable for global optimization. Simultaneously, the global best individual position Xbest(t) of the current population is obtained, representing the currently discovered global optimal solution. By simultaneously obtaining the historical best position and the current global best position of inferior individuals, dual reference information is provided for directional perturbation, ensuring that the reset individual can inherit valuable historical information and be guided towards the global optimum.

[0051] Step 7.3: Within the oriented neighborhood formed by the historical best position of the inferior individual and the current global best individual position, a Gaussian perturbation operator is introduced to generate a perturbation position. Specifically, this includes: based on the historical best position Xi of the inferior individual... pbest Construct a directional perturbation neighborhood with the current globally optimal individual position Xbest(t). Define the center position of the neighborhood as the weighted average of the two: Xcenter = 0.6 × Xbest(t) + 0.4 × Xi pbest, the setting of weight coefficients 0.6 and 0.4 makes the neighborhood center more inclined to the global optimal position, while retaining part of the historical optimal information. The neighborhood range radius is defined as a certain proportion of the distance between two positions: R=0.3×|Xbest(t)-Xi pbest |, each dimension component of this radius vector determines the perturbation range of the corresponding parameter dimension. A perturbed position is generated within the directional neighborhood, and a Gaussian perturbation operator is introduced, with the calculation formula: Xperturb=Xcenter+R⊙G, where G is an 8-dimensional Gaussian random vector, each dimension component is independently sampled from the standard normal distribution N(0, 1), and ⊙ denotes element-wise multiplication. The random perturbation generated by the Gaussian perturbation operator follows a normal distribution, most of the perturbation values are concentrated near the neighborhood center, and a small part of the perturbation values extend to the edge of the neighborhood. This distribution characteristic ensures that the perturbed position not only remains in a valuable region but also has a certain degree of random exploration. The generated perturbed position Xperturb integrates global optimal information, historical optimal information of inferior individuals, and random exploration, providing directional and diversified reset positions for inferior individuals.

[0052] Step 7.4, performing boundary over-limit check and correction on the perturbed position, and taking the corrected perturbed position as the reset individual, so that the reset individual retains the historical evolution information of the inferior individual and performs local optimization in the optimal neighborhood, which specifically includes: receiving the generated perturbed position Xperturb, executing the boundary over-limit check, and checking dimension by dimension whether each parameter component of the perturbed position is within the preset parameter value range [Xlb, Xub]. For components exceeding the upper bound, a boundary reflection correction strategy is adopted: if Xperturb,i>Xub, it is corrected to Xperturb,i=Xub-(Xperturb,i-Xub), that is, the excess part is reflected back into the boundary; if it still exceeds the boundary after correction, it is directly limited to the boundary value Xub. For components below the lower bound, the same reflection correction strategy is adopted: if Xperturb,i<Xlb, it is corrected to Xperturb,i=Xlb+(Xlb-Xperturb,i); if it still exceeds the boundary after correction, it is directly limited to the boundary value Xlb. Compared with directly limiting to the boundary value, the boundary reflection correction strategy can better retain the perturbation direction information and improve the diversity of reset positions. After completing the boundary correction, the corrected perturbed position Xperturb is taken as the position vector of the reset individual, replacing the current position of the original inferior individual. The reset individual inherits the historical optimal position information of the inferior individual, and restarts the search in the directional neighborhood formed by the global optimal position and the historical optimal position, which not only avoids the loss of valuable information caused by completely random reset, but also introduces new exploration possibilities through Gaussian perturbation, effectively balancing the retention of historical information and the exploration of new regions. The directional perturbation whale fall mechanism ensures that inferior individuals can efficiently re-participate in the optimization process and improve the overall quality of the population.

[0053] In this embodiment of the invention, by employing a technique of sorting the population individuals according to the stage-perceived comprehensive fitness value after each position update and identifying the last preset proportion of inferior individuals, obtaining the current global optimal position and its historical optimal position for the inferior individuals, and introducing a Gaussian perturbation operator to generate perturbation positions within the directional neighborhood formed by the two, and then using the perturbation positions for boundary violation verification and correction as reset individuals to replace inferior individuals, the invention overcomes the technical problems of existing intelligent optimization algorithms, such as the whale fall mechanism using completely random reset, loss of valuable information accumulated during the historical evolution of inferior individuals leading to decreased optimization efficiency, and the lack of a directional perturbation mechanism to guide reset individuals to high-value areas. This achieves the technical effect of retaining the historical optimal information of inferior individuals while eliminating them through the directional perturbation whale fall mechanism, ensuring that reset individuals re-search in high-value areas based on the global optimal and historical optimal construction of a directional neighborhood, and avoiding immediate local optima after reset by introducing random exploration through Gaussian perturbation, effectively improving the population diversity maintenance ability and global optimization efficiency of the hybrid whale optimization algorithm.

[0054] In a preferred embodiment of the present invention, the process of repeating the above iterative evolution process until the termination condition is met further includes: Step 8.1 divides the set of parameters to be optimized into a stability parameter group and an anti-chattering parameter group. The stability parameter group includes the outer loop gain and inner loop linear gain of the speed, while the anti-chattering parameter group includes the sliding surface coefficient and switching gain. Specifically, based on the physical functional characteristics of each parameter in the sliding mode direct torque control system, the eight parameters in the set of parameters to be optimized are functionally grouped. The stability parameter group is identified, which contains four parameters: the outer loop proportional gain Kp of the speed, the outer loop integral gain Ki of the speed, the inner loop linear gain kT of the torque, and the inner loop linear gain kψ of the flux linkage. These parameters mainly affect the stability and dynamic response performance of the system and are fundamental parameters for the stable operation of the control system. They need to be determined first to ensure the basic stability of the system. The anti-chattering parameter group is identified, comprising four parameters: the inner torque loop sliding surface coefficient cT, the inner torque loop switching gain εT, the inner flux loop sliding surface coefficient cψ, and the inner flux loop switching gain εψ. These parameters primarily affect the chattering degree and approach speed of sliding mode control. They are parameters used to further optimize control quality while ensuring system stability and require fine-tuning after the stability parameters are determined. The parameter grouping is based on the physical mechanism of sliding mode direct torque control: the stability parameter group determines the system's basic dynamic performance and stability margin, while the anti-chattering parameter group optimizes the smoothness and accuracy of sliding mode control based on a stable system. The parameter grouping provides a theoretical basis for subsequent phased optimization strategies.

[0055] Step 8.2: In the early stage of iterative evolution, the variable values ​​of the anti-bounce parameter set are fixed, and the first-stage optimization evolution of the stability parameter set is performed only in the multi-scale search space until the fitness value of the stability parameter set meets the first-stage convergence condition. Specifically, this includes: setting the early stage of iterative evolution to the first 200 iterations, performing the first-stage optimization in this stage, focusing on the optimization of the stability parameter set. The four parameters of the anti-bounce parameter set are fixed to preset empirical initial values: torque inner loop sliding surface coefficient cT=50, torque inner loop switching gain εT=100, flux inner loop sliding surface coefficient cψ=50, flux inner loop switching gain εψ=100. These initial values ​​are set based on motor control experience and can ensure basic sliding mode control functions. In the multi-scale search space, only the four dimensions corresponding to the stability parameter set are optimized. In the individual position vector of the mixed whale population, the four dimensions of the stability parameter set participate in the position update of the wide-area exploration mode and the local development mode, while the four dimensions of the anti-bounce parameter set remain fixed and do not participate in the update. In the first stage of optimization, the fitness function focuses on evaluating the system's dynamic response performance and stability indices, including speed tracking error, rise time, and overshoot. The first-stage convergence condition is set as follows: within 30 consecutive iterations, the improvement in the globally optimal fitness value corresponding to the stability parameter set is less than 0.01, or the maximum number of iterations in the previous stage (200 iterations) is reached. When the first-stage convergence condition is met, the stability parameter set is considered to have reached a relatively optimal state, providing stable basic dynamic performance for the system, thus completing the first stage of optimization. The first-stage optimization, by fixing the anti-chattering parameter set, reduces the dimensional complexity of the optimization problem, enabling the algorithm to efficiently search for the optimal combination of stability parameters in 4-dimensional space.

[0056] It should be noted that the fitness function in the first-stage optimization process focuses on evaluating dynamic response performance and stability indicators. It is implemented by adjusting the weights based on the stage-aware integrated fitness function described in step 3, and does not disable the automatic stage identification and switching mechanism in step 3. Specifically, in each fitness evaluation of the first-stage optimization, the current response stage identification based on the rate of change of speed error described in step 3.1 is still performed. It still automatically switches between the dynamic adjustment stage and the steady-state maintenance stage and dynamically adjusts the weight coefficients. However, based on the stage-aware integrated fitness value calculation described in step 3.4, the basic weight coefficient w3 of the speed error term and the basic value of the overshoot penalty coefficient λ are further increased for the first-stage optimization. This ensures that regardless of whether the system is in the dynamic adjustment stage or the steady-state maintenance stage, the fitness function always pays more attention to dynamic response performance indicators (such as rise time, overshoot, and speed tracking error) than to steady-state performance indicators (such as torque pulsation and flux pulsation), thereby achieving the goal of prioritizing the optimization of stability parameter groups in the first stage to establish the basic dynamic performance of the system.

[0057] Step 8.3: In the later stage of iterative evolution, inherit the results of the stability parameter set obtained in the first stage of optimization, unlock the anti-bounce parameter set, and perform a second stage of fine-grained optimization evolution using the anti-bounce parameter set. Specifically, this includes: after the convergence condition of the first stage is met, enter the later stage of iterative evolution and execute the second stage of optimization, which starts from the 201st generation iteration and continues until the maximum number of iterations reaches 500 generations. Inherit the optimal values ​​of the stability parameter set obtained in the first stage of optimization, and fix the four parameters of the stability parameter set to the parameter values ​​corresponding to the globally optimal individual at the end of the first stage, for example, Kp=35.2, Ki=120.5, kT=450.3, kψ=380.7. Unlock the four parameters of the anti-bounce parameter set, allowing these parameters to be optimized in a multi-scale search space. The four dimensions of the anti-bounce parameter set in the individual position vector of the mixed whale population begin to participate in the position update of the wide-area exploration mode and the local exploitation mode, while the four dimensions of the stability parameter set remain fixed and are no longer updated. In the second-stage optimization process, the fitness function focuses on evaluating the system's steady-state performance and chatter suppression effect, including torque pulsation, flux linkage pulsation, and the high-frequency chattering amplitude of the control signal. Since the stability parameter set has been determined through the first-stage optimization, the system possesses good basic stability and dynamic response. The second stage can then focus on optimizing the anti-chattering parameters based on the stable system, achieving smoothness and accuracy in sliding mode control. The second-stage optimization is also conducted in 4-dimensional space. By inheriting the stability parameter results from the first stage, it avoids parameter coupling interference that might occur with simultaneous optimization in 8-dimensional space, achieving decoupled, staged optimization of stability optimization and chatter suppression. At the end of the second stage, the output is a complete 8-dimensional optimal control parameter set containing both the stability parameter set and the anti-chattering parameter set.

[0058] It should be noted that the fitness function in the second-stage optimization process focuses on evaluating steady-state performance and chatter suppression effect. It is also implemented by adjusting the weights based on the stage-aware integrated fitness function described in step 3, and does not disable the automatic stage identification and switching mechanism in step 3.

[0059] In this embodiment of the invention, by dividing the set of parameters to be optimized into a stability parameter group and an anti-bounce parameter group, fixing the anti-bounce parameter group in the early stage of iterative evolution, and only performing the first-stage optimization on the stability parameter group until the convergence condition is met, and inheriting the results of the first stage in the later stage of iterative evolution, fixing the stability parameter group and unlocking the anti-bounce parameter group to perform the second-stage fine-grained optimization, the technical problems of existing parameter optimization methods, such as simultaneous full-space optimization of all parameters, high-dimensional parameter coupling leading to low search efficiency and difficulty in convergence, and lack of parameter physical function grouping and staged optimization strategy which cannot effectively decouple the optimization process of different functional parameters, are overcome. Thus, the technical effect of reducing the dimensional complexity of the optimization problem through a staged parameter optimization strategy, first optimizing the stability parameter group to establish the basic performance of the system and then optimizing the anti-bounce parameter group to refine the control quality, effectively decoupling the parameter optimization process of stability optimization and bounce suppression, and significantly improving the search efficiency and convergence reliability of the hybrid whale optimization algorithm in multi-parameter optimization of permanent magnet synchronous motors is achieved.

[0060] In a preferred embodiment of the present invention, the above-mentioned output of the target control parameter set corresponding to the globally optimal position vector, and the configuration of the target control parameter set into the permanent magnet synchronous motor sliding mode direct torque control system to drive the motor operation, includes: Step 9.1: From the global optimal position vector output by the final iteration, extract the optimal values ​​of the stability parameter set and the anti-bounce parameter set to form the target control parameter set. Specifically, this includes: when the hybrid whale optimization algorithm completes the maximum number of iterations (500 generations) or meets other termination conditions, extract the individual with the smallest stage-aware integrated fitness value in the population as the global optimal individual, and read its position vector Xbest. final This 8-dimensional vector contains the optimized control parameters. Based on the parameter grouping definition, the optimal values ​​of the stability parameter group are extracted from the global optimal position vector: the outer loop proportional gain of the rotational speed Kp. opt Speed ​​outer loop integral gain Ki opt Torque inner loop linear gain kT opt and the linear gain kψ of the inner loop of the magnetic flux linkage opt These four parameters correspond to the first four dimensions or a specific dimension of the position vector (determined by the encoding method). The optimal values ​​of the anti-bounce parameter set are derived: the torque inner circulation surface coefficient cT. opt Torque inner loop switching gain εT opt , the inner slip coefficient of the magnetic flux linkage cψ opt and the inner loop switching gain εψ of the flux linkage opt These four parameters correspond to the remaining dimensions of the position vector. The eight optimal parameter values ​​obtained are used to construct the target control parameter set, which is recorded as the parameter vector X. opt =[Kp opt Ki opt,cT opt ,εT opt ,kT opt ,cψ opt ,εψ opt ,kψ opt This parameter set represents the optimal control strategy after global optimization using the hybrid whale optimization algorithm, enabling the permanent magnet synchronous motor sliding mode direct torque control system to achieve the best overall performance. Simultaneously, the stage-aware comprehensive fitness value F corresponding to the optimal parameter set is recorded. opt , as a quantitative evaluation indicator of optimization effect.

[0061] Step 9.2 involves configuring the outer loop gain of the speed control parameter set to the outer loop proportional-integral controller, and configuring the sliding surface coefficient, switching gain, and linear gain to the integral sliding surface and exponential reaching law of the flux sliding controller and torque sliding controller, respectively. Specifically, this includes: based on the target control parameter set X... opt The parameter configuration operation of the sliding mode direct torque control system for the permanent magnet synchronous motor is performed. For the speed outer loop proportional-integral controller, the optimized proportional gain Kp is... opt Configured as the proportional gain of the PI controller, the integral gain Ki opt The integral coefficient configured for the PI controller is updated in the following formula for calculating the output of the outer loop speed controller: Tref = Kp opt ·eω+Ki opt ·∫eω dt, where eω is the rotational speed error and Tref is the output reference torque command. For the flux linkage sliding mode controller, the optimized flux linkage inner loop sliding surface coefficient cψ is used. opt In the integral sliding surface design, the sliding surface equation is updated to: sψ=eψ+cψ_opt·∫eψ dt, where eψ is the flux linkage error; the switching gain εψ is then applied. opt and linear gain kψ opt When incorporated into the exponential reaching law, the reaching law equation is updated to: dsψ / dt = -εψ opt ·sgn(sψ)-kψ opt ·sψ, the flux linkage sliding mode controller outputs the d-axis reference voltage ud* based on the updated sliding surface and reaching law. For the torque sliding mode controller, the same configuration method is used, with the torque inner loop sliding surface coefficient cT... opt Switching gain εT opt and linear gain kT opt Assigned to the torque sliding surface and the reaching law respectively: sT=eT+cT opt ·∫eT dt,dsT / dt=-εT opt ·sgn(sT)-kT opt• sT, the torque sliding mode controller outputs the q-axis reference voltage uq*. Through parameter configuration, the three controllers of the permanent magnet synchronous motor sliding mode direct torque control system (speed outer loop PI controller, flux sliding mode controller, and torque sliding mode controller) all adopt the optimized target control parameter set to ensure that the overall control strategy reaches the optimal level.

[0062] Step 9.3: Based on the configured speed outer-loop proportional-integral controller, flux linkage sliding mode controller, and torque sliding mode controller, a stator voltage reference command is generated and output as a drive signal via the space vector pulse width modulation module to control the operation of the permanent magnet synchronous motor. Specifically, the permanent magnet synchronous motor sliding mode direct torque control system begins real-time operation based on the configured target control parameter set. The speed outer-loop proportional-integral controller receives feedback signals of the reference speed ωref and the actual speed ω, calculates the speed error eω=ωref-ω, and based on the configured Kp... opt with Ki opt Output reference torque command Tref. The flux linkage sliding mode controller receives feedback signals of the reference flux linkage amplitude ψref and the actual flux linkage amplitude ψs, calculates the flux linkage error eψ=ψref-ψs, and bases it on the configured sliding surface coefficient cψ. opt With the approach law parameter εψ opt 、kψ opt The d-axis reference voltage ud* is calculated using a sliding mode control law. The torque sliding mode controller receives the feedback signals of the reference torque command Tref output from the outer speed loop and the actual electromagnetic torque Te, and calculates the torque error eT=Tref-Te based on the configured sliding surface coefficient cT. opt With the approach law parameter εT opt kT opt The q-axis reference voltage uq* is calculated and output using a sliding mode control law. The reference voltages ud* and uq* in the dq rotating coordinate system are input to the coordinate inverse transformation module. After Park and Clarke inverse transformations, reference voltages ua*, ub*, and uc* in the three-phase stationary coordinate system are generated as the stator voltage reference command. The stator voltage reference command is input to the space vector pulse width modulation (SVPWM) module. The SVPWM module calculates the corresponding switching state sequence and duration based on the reference voltage vector, generates a PWM drive signal, and outputs it to the inverter's power switching devices. This controls the inverter to output the desired stator voltage, driving the permanent magnet synchronous motor to operate according to the optimized control strategy. During motor operation, feedback signals such as actual speed, electromagnetic torque, and stator flux linkage are collected to form a closed-loop control in real time, ensuring that the motor achieves optimized control performance with low torque ripple, high flux linkage tracking accuracy, and fast dynamic response under the target control parameter set.

[0063] In this embodiment of the invention, the optimal values ​​of the stability parameter set and the anti-chattering parameter set are extracted from the globally optimal position vector output by the final iteration to form the target control parameter set. The speed outer loop gain of the target control parameter set is configured to the speed outer loop PI controller, and the sliding mode surface coefficient and the reaching law parameter are configured to the flux linkage and torque sliding mode controller. Based on the configured controller, a stator voltage reference command is generated and the drive signal is output through the SVPWM module to control the motor operation. Therefore, this invention overcomes the technical problems of existing parameter optimization methods, such as the inability to directly apply the optimization results to the actual control system, the lack of a complete implementation plan from optimized parameters to controller configuration, and the cumbersome manual parameter adjustment that makes it difficult to achieve multi-objective comprehensive optimization. This invention achieves seamless configuration and application of optimized parameters to the control system, enables the permanent magnet synchronous motor sliding mode direct torque control system to automatically achieve optimal control performance based on the target control parameter set, eliminates the cumbersome process of manual parameter adjustment, and ensures that the system can adaptively obtain the best control parameters under different motor models and operating conditions, significantly improving the performance and intelligence level of the permanent magnet synchronous motor control system.

[0064] Please see Figure 2 The embodiments of the present invention also propose a sliding mode direct torque control system for permanent magnet synchronous motors based on a hybrid whale optimization algorithm, the system comprising: Main control unit, inverter, signal acquisition unit and permanent magnet synchronous motor; The signal acquisition unit is used to acquire real-time operating data of the permanent magnet synchronous motor. The main control unit is connected to the signal acquisition unit and the inverter. The main control unit includes a parameter optimization subunit and a direct torque control subunit. The parameter optimization subunit includes: The space construction module is used to obtain the set of parameters to be optimized for the sliding mode direct torque control system of permanent magnet synchronous motor, to evaluate the sensitivity of each parameter dimension in the set of parameters to be optimized, and to decouple and map each parameter dimension based on the evaluation results to construct a multi-scale search space. The model simulation module is used to initialize a hybrid whale population in a multi-scale search space. For each individual in the hybrid whale population, the control parameter set is reconstructed based on the individual's position vector in the multi-scale search space and input into a preset motor control model for operation, and dynamic response data during operation is collected. The fitness assessment module is used to extract characteristic indicators of the system at different response stages based on dynamic response data, and to dynamically adjust the fitness evaluation strategy in combination with the characteristic indicators to calculate the individual's stage-aware comprehensive fitness value. The transition probability calculation module is used to introduce an evolutionary stagnation factor during the iterative evolution of a hybrid whale population, and to jointly calculate the nonlinear dynamic transition probability based on the evolutionary stagnation factor and the current iterative state. The wide-area exploration module is used to control an individual to enter wide-area exploration mode when the random judgment value is less than the nonlinear dynamic transformation probability. It updates the individual's position vector based on the beluga whale swimming mechanism and combined with the dimension scaling factor of the multi-scale search space. The local development module is used to control an individual to enter local development mode when the random judgment value is greater than or equal to the nonlinear dynamic transformation probability. It updates the individual's position vector based on the whale spiral attack mechanism and combined with the dynamic convergence factor. The whale fall reset module is used to identify inferior individuals in the population based on the phase-aware comprehensive fitness value after each location update, perform a targeted perturbation whale fall mechanism on the inferior individuals to generate reset individuals, and replace the corresponding inferior individuals with the reset individuals. The optimization output module is used to control the repeated iterative evolution process of the above modules until the termination condition is met, and outputs the target control parameter set corresponding to the global optimal position vector; The direct torque control subunit is used to receive the target control parameter set output by the optimization output module to configure the controller parameters, and to generate a space vector pulse width modulation signal by combining the real-time operating data input by the signal acquisition unit, and output the space vector pulse width modulation signal to the inverter.

[0065] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0066] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0067] It should be noted that all formulas in this manual are calculated by removing dimensions and taking their numerical values. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0068] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A sliding mode direct torque control method for permanent magnet synchronous motors based on a hybrid whale optimization algorithm, characterized in that, include: A set of parameters to be optimized for the sliding mode direct torque control system of a permanent magnet synchronous motor is obtained. Sensitivity evaluation is performed on each parameter dimension in the set of parameters to be optimized. Based on the evaluation results, the parameter dimensions are decoupled and mapped to construct a multi-scale search space. The set of parameters to be optimized includes the outer loop control parameters of speed and the inner loop sliding mode control parameters of torque and flux linkage. A hybrid whale population is initialized in the multi-scale search space. For each individual in the hybrid whale population, a set of control parameters is reconstructed based on the individual's position vector in the multi-scale search space and input into a preset motor control model for operation. Dynamic response data during the operation process is collected. Based on the dynamic response data, feature indicators of the system at different response stages are extracted, and the fitness evaluation strategy is dynamically adjusted in combination with the feature indicators to calculate the individual's stage-perception comprehensive fitness value. In the iterative evolution of the hybrid whale population, an evolutionary stagnation factor is introduced. Based on this stagnation factor and the current iterative state, a nonlinear dynamic transition probability is jointly calculated, including: Record the number of times the global optimal fitness value of the hybrid whale population has not been improved within a consecutive preset number of generations, and use the ratio of the number of times the fitness value has not been improved to the consecutive preset number of generations as the evolutionary stagnation factor; Obtain the iteration ratio of the current iteration number to the maximum iteration number, and calculate the basic transition probability based on the iteration ratio; The evolutionary stagnation factor is used as a stagnation penalty term, and the stagnation penalty term is superimposed on the basic transition probability to generate the nonlinear dynamic transition probability; When the random determination value is less than the nonlinear dynamic conversion probability, the individual is controlled to enter the wide-area exploration mode, and the position vector of the individual is updated based on the beluga whale swimming mechanism and combined with the dimension scaling factor of the multi-scale search space. When the random judgment value is greater than or equal to the nonlinear dynamic conversion probability, the individual is controlled to enter the local development mode, and the position vector of the individual is updated based on the whale spiral attack mechanism and combined with the dynamic convergence factor. After each location update, inferior individuals in the population are identified based on the stage-aware comprehensive fitness value. A directional perturbation whale fall mechanism is performed on the inferior individuals to generate reset individuals, and the reset individuals replace the corresponding inferior individuals. Repeat the above iterative evolution process until the termination condition is met, output the target control parameter set corresponding to the global optimal position vector, and configure the target control parameter set into the permanent magnet synchronous motor sliding mode direct torque control system to drive the motor to run; The process of repeating the above iterative evolution process until the termination condition is met also includes: The set of parameters to be optimized is divided into a stability parameter group and an anti-bounce parameter group. The stability parameter group includes the outer loop gain and the inner loop linear gain of the rotational speed, and the anti-bounce parameter group includes the sliding surface coefficient and the switching gain. In the early stage of iterative evolution, the variable values ​​of the anti-bounce parameter group are fixed, and the stability parameter group is optimized and evolved only in the multi-scale search space until the fitness value of the stability parameter group satisfies the first stage convergence condition. In the later stage of iterative evolution, the results of the stability parameter set obtained in the first stage of optimization are inherited, the anti-bounce parameter set is unlocked, and the anti-bounce parameter set is subjected to a second stage of fine-grained optimization evolution.

2. The method according to claim 1, characterized in that, The process involves sensitivity evaluation of each parameter dimension in the set of parameters to be optimized, decoupling and mapping each parameter dimension based on the evaluation results, and constructing a multi-scale search space, including: Preset perturbations are applied to the proportional gain and integral gain of the outer speed loop, as well as the sliding surface coefficient, switching gain and linear gain of the inner torque loop and inner flux loop in the parameter set to be optimized, respectively, and the rate of change of the output variable of the motor control model is collected as a sensitivity index. Based on the sensitivity index, the set of parameters to be optimized is divided into a high-sensitivity parameter combination and a low-sensitivity parameter combination. A first search step size interval is configured for the high-sensitivity parameter combination, and a second search step size interval is configured for the low-sensitivity parameter combination, wherein the upper limit of the step size of the first search step size interval is smaller than the upper limit of the step size of the second search step size interval. The high-sensitivity parameter combination and the low-sensitivity parameter combination are mapped to different dimensional intervals of the same search space to generate the multi-scale search space, and dimensional scaling factors corresponding to the first search step size interval or the second search step size interval are assigned to different dimensional intervals in the multi-scale search space.

3. The method according to claim 1, characterized in that, Based on the dynamic response data, the system extracts characteristic indicators at different response stages, and dynamically adjusts the fitness evaluation strategy in conjunction with these characteristic indicators to calculate the individual's stage-aware comprehensive fitness value, including: The current response stage is identified based on the rate of change of rotational speed error in the dynamic response data, and the current response stage includes a dynamic adjustment stage or a steady-state maintenance stage. During the dynamic adjustment phase, overshoot and rise time characteristics are extracted as dominant indicators, and the dynamic weight coefficients of the speed error term and overshoot penalty term in the fitness function are increased. During the steady-state maintenance phase, torque ripple characteristics and flux ripple characteristics are extracted as dominant indicators, and the dynamic weight coefficients of torque error term and flux error term in the fitness function are increased. Based on the dynamic weight coefficients corresponding to the current response stage, the time-multiplied absolute error integral index is weighted and calculated, and the stage perception comprehensive fitness value is generated by combining the overshoot penalty term.

4. The method according to claim 2, characterized in that, When the random determination value is less than the nonlinear dynamic transformation probability, the individual is controlled to enter a wide-area exploration mode. The individual's position vector is updated based on the beluga whale swimming mechanism and combined with the dimensional scaling factor of the multi-scale search space, including: When the random determination value is less than the nonlinear dynamic transformation probability, a reference individual position is randomly selected from the current population. Obtain the dimension scaling factor corresponding to the dimension interval where the individual is located, and use the dimension scaling factor as the adaptive step size adjustment coefficient; Based on the reference individual position, the individual's current position, and the adaptive step size adjustment coefficient, the individual's displacement in the multi-scale search space is calculated. The displacement is superimposed on the current position of the individual to generate the updated position vector of the individual in the wide-area exploration mode.

5. The method according to claim 1, characterized in that, When the random determination value is greater than or equal to the nonlinear dynamic transformation probability, the individual is controlled to enter a local development mode, and the position vector of the individual is updated based on the whale spiral attack mechanism and combined with the dynamic convergence factor, including: When the random determination value is greater than or equal to the nonlinear dynamic transformation probability, the global optimal individual position of the current population is obtained; Calculate the spatial distance between the current position of the individual and the position of the globally optimal individual, and construct a dynamic convergence factor that decreases with each iteration based on the current number of iterations; Based on the dynamic convergence factor and the logarithmic spiral constant, the spiral encirclement displacement of the individual toward the position of the globally optimal individual is calculated. The spiral encircling displacement is combined with the globally optimal individual position to generate the updated position vector of the individual in the local development mode.

6. The method according to claim 1, characterized in that, The step of identifying inferior individuals in the population based on the stage-aware integrated fitness value after each location update, and performing a directional perturbation whale fall mechanism on the inferior individuals to generate reset individuals includes: After each location update, the individuals in the population are sorted according to the overall fitness value of the stage perception, and the individuals with fitness values ​​at the bottom of the preset proportion are identified as inferior individuals. For the inferior individual, obtain the current global best individual position and the historical best position of the inferior individual; Within the oriented neighborhood formed by the historical best position of the inferior individual and the current global best individual position, a Gaussian perturbation operator is introduced to generate a perturbation position. The perturbation position is checked and corrected for boundary violations. The corrected perturbation position is used as the reset individual so that the reset individual retains the historical evolution information of the inferior individual and performs local optimization in the optimal neighborhood.

7. The method according to claim 1, characterized in that, The process of outputting the target control parameter set corresponding to the globally optimal position vector and configuring the target control parameter set into the permanent magnet synchronous motor sliding mode direct torque control system to drive the motor includes: From the global optimal position vector output by the final iteration, the optimal values ​​of the stability parameter set and the anti-bounce parameter set are extracted to form the target control parameter set; The outer loop gain of the target control parameter set is configured to the outer loop proportional-integral controller of the speed, and the sliding surface coefficient, switching gain and linear gain are respectively configured to the integral sliding surface and exponential approach law of the flux sliding controller and the torque sliding controller; Based on the configured speed outer loop proportional-integral controller, flux sliding mode controller, and torque sliding mode controller, a stator voltage reference command is generated, and a drive signal is output via the space vector pulse width modulation module to control the operation of the permanent magnet synchronous motor.

8. A sliding mode direct torque control system for a permanent magnet synchronous motor based on a hybrid whale optimization algorithm, used to implement the method described in any one of claims 1 to 7, characterized in that, include: Main control unit, inverter, signal acquisition unit and permanent magnet synchronous motor; The signal acquisition unit is used to acquire the real-time operating data of the permanent magnet synchronous motor; The main control unit is connected to the signal acquisition unit and the inverter. The main control unit includes a parameter optimization subunit and a direct torque control subunit. The parameter optimization subunit includes: The space construction module is used to obtain the set of parameters to be optimized in the sliding mode direct torque control system of the permanent magnet synchronous motor, perform sensitivity evaluation on each parameter dimension in the set of parameters to be optimized, and decouple and map each parameter dimension based on the evaluation results to construct a multi-scale search space. The model simulation module is used to initialize the hybrid whale population in the multi-scale search space, and for each individual in the hybrid whale population, to reconstruct the control parameter set based on the position vector of the individual in the multi-scale search space, and input it into the preset motor control model for operation, and collect dynamic response data during the operation process. The fitness assessment module is used to extract feature indicators of the system at different response stages based on the dynamic response data, and dynamically adjust the fitness evaluation strategy in combination with the feature indicators to calculate the individual's stage-aware comprehensive fitness value. The transition probability calculation module is used to introduce an evolution stagnation factor during the iterative evolution of the hybrid whale population, and to jointly calculate the nonlinear dynamic transition probability based on the evolution stagnation factor and the current iterative state. The wide-area exploration module is used to control the individual to enter the wide-area exploration mode when the random judgment value is less than the nonlinear dynamic conversion probability, and to update the position vector of the individual based on the beluga whale swimming mechanism and combined with the dimension scaling factor of the multi-scale search space. The local development module is used to control the individual to enter the local development mode when the random judgment value is greater than or equal to the nonlinear dynamic transformation probability, and to update the position vector of the individual based on the whale spiral attack mechanism and combined with the dynamic convergence factor. The whale fall reset module is used to identify inferior individuals in the population based on the stage perception comprehensive fitness value after each location update, perform a directional perturbation whale fall mechanism on the inferior individuals to generate reset individuals, and replace the corresponding inferior individuals with the reset individuals. The optimization output module is used to control the repeated iterative evolution process of the above modules until the termination condition is met, and outputs the target control parameter set corresponding to the global optimal position vector; The direct torque control subunit is used to receive the target control parameter set output by the optimization output module to configure the controller parameters, and generate a space vector pulse width modulation signal by combining the real-time operating data input by the signal acquisition unit, and output the space vector pulse width modulation signal to the inverter.

Images