Torque optimal distribution method of dual-motor driving system for 1ms high-precision control period

Abstract

The invention discloses a torque optimization distribution method of a dual-motor driving system for a 1ms high-precision control period, which comprises the following steps of: 1, constructing a real-time dual-motor torque distribution mechanism based on an interpolation model and an improved quantum particle swarm optimization (QPSO) algorithm, updating an individual optimal position and global optimum according to the fitness of each particle, outputting front and rear motor torques corresponding to the optimal comprehensive efficiency; 2, on the basis of the real vehicle speed and the target vehicle speed, particle search boundaries are limited on the basis of an improved quantum particle swarm optimization (QPSO) algorithm, quantum behaviors are evolved, a multi-dimensional composite fitness function is introduced, and the optimal compensation torque and the optimal convergence curve are output; according to the method, the bound state model in quantum mechanics is introduced, the particle position evolution behavior is described by the probability distribution of the wave function, and the capability of jumping out of the local extremal region of the particles is enhanced through the quantum tunneling effect, so that the overall global optimization performance is improved.

Description

Technical Field

[0001] This invention belongs to the field of motor drive optimization, and in particular relates to a torque optimization allocation method for a dual-motor drive system with a high-precision control cycle of 1ms. Background Technology

[0002] In intelligent control tasks with high dynamic response requirements, the real-time performance of the algorithm directly determines the executability and control stability of the system. Current mainstream intelligent optimization algorithms, such as genetic algorithms (GA) and standard particle swarm optimization (PSO), although strong in global optimization capabilities, suffer from difficulties in meeting real-time response constraints in high-frequency control loops due to their reliance on iterative population evolution strategies, redundant fitness evaluation, and lack of convergence acceleration mechanisms. This is especially true in control cycles of 10ms and below, where a significant response bottleneck exists.

[0003] Real-time bottleneck manifestations:

[0004] High computational complexity: Traditional PSO or GA typically requires processing populations of tens to hundreds of particles / individuals, and each iteration requires multiple computational steps such as position update, fitness evaluation, sorting, and optimal individual selection. The overall complexity increases linearly or even reaches O(Nd) (where N is the population size and dd is the dimension of the solution space). When executed on an embedded platform, it is very easy to trigger a computing bottleneck.

[0005] Convergence speed is lower than control cycle: Even with early convergence strategies or heuristic initialization, traditional methods still require 5 to 10 iterations to obtain an acceptable solution, which is impractical in hard real-time systems (such as 1ms-5ms cycles).

[0006] Unable to fully utilize parallel processing: Most existing embedded controllers are limited by hardware architecture and power consumption constraints, and cannot support GPU or multi-core parallel acceleration. Traditional optimization algorithms are highly dependent on parallel resources, which further limits their real-time performance.

[0007] The algorithm structure is non-recursive: Traditional algorithms such as PSO and GA do not have time-domain recursiveness, and the current control quantity cannot be directly derived from the previous solution result. It is necessary to iterate again, which increases redundant calculations.

[0008] Lack of model simplification mechanism: The failure to effectively utilize the dynamic characteristics of the system (such as state prediction and constraint boundary evolution) to prune the solution space results in repeated searches in the global space in each round of optimization, which is inefficient.

[0009] Fitness function evaluation is time-consuming: In multi-objective constrained optimization tasks, the fitness function often involves weighted calculation of multiple sub-objectives, physical model calls, or simulation interface evaluation, which further slows down the algorithm execution speed.

[0010] In summary, existing intelligent optimization methods suffer from significant response lag and computational bottlenecks in real-time industrial applications with control cycles of less than 10ms. The root cause lies in the non-recursive nature of the algorithm structure, high population redundancy, and the lack of incrementality in the search method. To meet the stringent requirements of next-generation intelligent electric drive control, industrial robots, and automation systems for millisecond-level real-time optimization, rapid decision generation, and low computational burden, it is urgent to construct a highly efficient online optimization framework with lightweight structure, model embedding capabilities, and time-recursive properties to achieve closed-loop, closed-time execution of control in high-frequency dynamic scenarios. Summary of the Invention

[0011] Objective: This invention aims to provide a torque optimization allocation method for dual-motor drive systems with a high-precision control cycle of 1ms. It introduces the bound-state model from quantum mechanics, using the probability distribution of the wave function to characterize the particle position evolution behavior. The quantum tunneling effect enhances the particle's ability to escape local extrema, thereby improving overall global optimization performance. Based on this, an adaptive mechanism combining a dynamic contraction factor and a boundary penalty function is constructed to ensure that the particle balances search efficiency and boundary feasibility during convergence, thus achieving stable and feasible torque allocation solution space navigation under nonlinear constraints. Experimental results show that the improved QPSO algorithm exhibits significant advantages over the traditional PSO algorithm in terms of convergence speed, solution accuracy, and robustness.

[0012] Technical solution: The present invention provides a torque optimization allocation method for a dual-motor drive system with a high-precision control cycle of 1ms, comprising the following steps:

[0013] Step 1: Construct a real-time dual-motor torque distribution mechanism based on the interpolation model and the improved quantum particle swarm optimization algorithm QPSO. Update the individual optimal position and global optimal position according to the fitness of each particle, and output the front and rear motor torques corresponding to the optimal overall efficiency.

[0014] Step 2: Based on the actual vehicle speed and the target vehicle speed, the particle search boundary is constrained by the improved quantum particle swarm optimization algorithm (QPSO), the quantum behavior is evolved, a multi-dimensional composite fitness function is introduced, and the optimal compensation torque and the optimal convergence curve are output.

[0015] Furthermore, step 1 specifically includes the following steps:

[0016] Step 1.1: Construct the interpolation model and preprocess the efficiency function;

[0017] The interpolation model is defined as follows:

[0018] η i =f i (T i ,ω i ), i = 1, 2

[0019] Where, η i T represents the efficiency of the i-th motor. i and ω i f represents its torque and speed, respectively. i It is an interpolation function;

[0020] Step 1.2: Initialize the particle swarm. Within the set dual-motor torque constraint boundary, generate the initial particle swarm using a uniform distribution and introduce the total target torque constraint condition.

[0021] Step 1.3: Construct a fitness function with the goal of maximizing the overall energy efficiency of the system;

[0022] Step 1.4: In each iteration, update the fitness of each particle. If the current particle position is better than its historical best value, then update its individual best value p. i Simultaneously, a global search space analysis is performed on the entire population to calculate the optimal center position P for all individual particles. mean ;

[0023] Step 1.5: Implement global jump search of particles using the quantum position update model in the improved QPSO;

[0024] Step 1.6: Based on the dual strategy of boundary truncation method and nonlinear penalty function method, correct or penalize particles that do not meet the following constraints during the iteration process;

[0025] Step 1.7: When the maximum number of iterations or the convergence accuracy threshold is reached, select the solution with the highest fitness in the current particle swarm as the final optimal torque allocation output value, and send the optimal torque allocation scheme to the drive controller in real time as the reference output for the next control cycle.

[0026] Furthermore, step 1.2 specifically involves: within the defined dual-motor torque constraint boundary, generating an initial particle swarm using a uniform distribution, where each particle represents a feasible combination of motor torques (T1, T2); simultaneously setting an initial convergence factor β0 and an optimal individual position p. i And the global optimum g of the population;

[0027] Introduce the total target torque constraint:

[0028] T total =T1+T2

[0029] This ensures that all generated particles meet the vehicle's power requirements.

[0030] Furthermore, step 1.3 specifically involves: taking the maximization of the system's overall energy efficiency as the optimization objective, and using a weighted combination of the efficiencies of the two motors as the fitness function:

[0031] F(T1,T2)=w1·η1(T1)+w2·η2(T2)

[0032] Where w1 and w2 are the weighting coefficients for the efficiency of the front motor and the rear motor, respectively, used to enhance the actual physical meaning of the optimization direction, η1 is the efficiency of the front motor, and η2 is the efficiency of the rear motor.

[0033] Furthermore, step 1.5 specifically involves: employing the quantum position update model in the improved QPSO to achieve a global jump search for the particle, whose position evolution is controlled by the following formula:

[0034]

[0035] Among them, P i It is the individual's optimal position, P mean The center of the optimal value for all particles is u, which is a uniformly distributed random variable, and β is a dynamic contraction factor that adaptively decays with the iteration process and is used to balance exploration and convergence.

[0036] Furthermore, step 1.6 specifically involves: using a dual strategy of boundary truncation and nonlinear penalty function, correcting or penalizing particles that do not meet the following constraints during the iteration process:

[0037] T min ≤T i ≤T max T1 + T2 = T total

[0038] The form of the penalty function is:

[0039] F penalized =F-λ·|T1+T2-T total |

[0040] Where λ is the penalty coefficient and F is the fitness function, which represents the evaluation index used to measure the quality of the solution during the optimization process. It is adaptively increased to strengthen the constraint convergence speed.

[0041] Furthermore, step 2 specifically includes the following steps:

[0042] Step 2.1: Initialize the improved Quantum Particle Swarm Optimization (QPSO) algorithm;

[0043] Algorithm control parameter settings: Set the population size, maximum number of iterations, and quantum behavior dynamic shrinkage factor β to balance search capability and convergence speed;

[0044] Physical boundary constraint definition: The upper and lower boundaries of the torque are set to ensure that the optimization results conform to the actual physical capabilities of the motor;

[0045] Memory and data structure optimization: The pre-allocation structure is used to improve the computational efficiency and enable fast storage and access of convergence curves.

[0046] Step 2.2: Initialize the particle swarm and construct the boundary constraint mechanism;

[0047] Step 2.3: Construct a quantum behavior evolution mechanism by introducing random perturbation and global mean guidance;

[0048] Step 2.4: Construct a multi-objective fitness function that considers speed error, torque stability, and speed change smoothness;

[0049] Step 2.5: Employ a two-condition termination mechanism, evaluate the impact of the quantum factor on convergence speed and search capability based on parameter sensitivity analysis, and verify robustness.

[0050] Furthermore, step 2.2 specifically involves: Particle swarm initialization: initializing the solution vectors using a uniform distribution.

[0051] X = (ub - lb)·rand(pop, dim) + lb

[0052] Where X is the initial solution vector of the particle; ub is the upper boundary constraint of the solution vector; lb is the lower boundary constraint of the solution vector; rand(pop,dim) is the function to generate uniformly distributed random numbers; pop is the population size, i.e. the number of particles; and dim is the number of variables contained in the solution vector of each particle.

[0053] Boundary constraint handling: Introducing a vectorized boundary correction function:

[0054] X correct =min(max(X,lb),ub)

[0055] Among them, X corrent This is the current particle position matrix.

[0056] Individual optimal value update mechanism: Local fitness criterion is used for individual optimal value update.

[0057] If f(x) new ) <f(pBest i If pBest is updated, then pBest will be updated. i =x new .

[0058] Furthermore, step 2.3 specifically involves: global mean calculation: calculating the mean of the best historical solutions of the current iteration population.

[0059]

[0060] Where, m bestpBest is the mean of the global optimal historical solutions, used to represent the center position of the current overall optimal solution in the population. i is the historical best position (personal best) of the i-th particle, that is, the best solution vector found by the particle so far.

[0061] Position evolution equation:

[0062]

[0063] Where r∈{0,1}, u~U(0,1) are random perturbations;

[0064] Two-modal perturbation mechanism: via (-1) r Achieve directional perturbation of the search space.

[0065] Furthermore, step 2.4 specifically involves constructing a multi-objective fitness function that considers speed error, torque stability, and speed change smoothness, in the following form:

[0066]

[0067] Among them, v ref (t) represents the desired velocity at time t; v real (t) represents the actual speed at time t; λ1 is the torque stability weighting coefficient, used to balance the importance of speed error and torque performance; T comp To compensate for the amplitude of the torque; λ2 is the speed change smoothness weighting coefficient, used to balance the smoothness of speed error and speed change rate; Δv real (t) represents the rate of change of the actual velocity.

[0068] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages:

[0069] 1. This invention achieves optimized allocation of the total target torque of two motors by introducing structural improvements to the Quantum-behaved Particle Swarm Optimization (QPSO) algorithm. Improvements include: a dynamic inertia weight adjustment mechanism, enhanced fitness function structure, and a search space compression strategy to improve the algorithm's convergence speed and stability under complex operating conditions. This strategy ensures that the total system torque meets the drive requirements while comprehensively considering the efficiency characteristic curves of the two motors, achieving high stability and high overall efficiency in torque allocation, thereby effectively reducing energy consumption and minimizing motor load fluctuations.

[0070] 2. This invention addresses the vehicle speed deviation problem caused by load disturbances, environmental changes, or system response lag during system operation by designing a real-time torque compensation strategy. This strategy uses the dynamic error between the target vehicle speed and the current actual vehicle speed as input and rapidly calculates the optimal compensation torque based on an improved QPSO algorithm. By introducing an error tracking function, a torque adjustment sensitivity coefficient, and an adaptive step size update rule, the error under closed-loop speed control is minimized (approaching zero), thereby improving the dynamic response capability and speed control accuracy of the entire vehicle system under rapidly changing operating conditions.

[0071] 3. This invention supports 1ms-level high-frequency control cycles, meeting the real-time control requirements of industrial applications; it balances torque stability and motor efficiency, improving system operating efficiency; it introduces a dynamic compensation mechanism to enhance the system's adaptability to disturbances and hysteresis, maintaining high vehicle speed following accuracy; the algorithm structure is lightweight, suitable for embedded controller deployment. It is applicable to various multi-motor collaborative applications requiring high-speed dynamic response, including electric vehicles, dual-motor servo control systems, and intelligent robots, demonstrating good engineering practicality and promotional value.

[0072] 4. This invention integrates efficiency optimization and disturbance compensation control modules based on interpolation modeling within a millisecond-level control cycle. By improving the quantum particle swarm algorithm design, it achieves dynamic distribution and error compensation of dual motor torque under all operating conditions. Its overall control response time is stably maintained at the level of 1ms, which is far superior to deep strategy methods that require multi-layer neural network inference processes.

[0073] 5. Unlike the "black box output" of end-to-end reinforcement learning networks, this invention explicitly embeds multi-dimensional control objectives such as motor efficiency, speed error, and speed fluctuation into the fitness function, which has good physical interpretability and engineering parameter tuning capabilities, and can quickly adjust and optimize strategies according to different driving scenarios.

[0074] 6. Thanks to the built-in perturbation search and boundary shrinkage mechanism of the particle swarm algorithm, this invention can still quickly recover control performance through adaptive iteration when faced with sudden perturbations or inputs outside the operating conditions. It has stronger fault tolerance and robustness, is suitable for control scenarios in complex environments, and avoids the failure of reinforcement learning strategies in the case of "no samples seen".

[0075] 7. This invention achieves high-speed optimization through a simplified particle swarm controller, with low algorithm computation and easy deployment on existing embedded platforms in vehicles, without the need for high-performance AI chips or additional inference platforms, thus possessing extremely high engineering promotion value. Attached Figure Description

[0076] Figure 1 Optimization flowchart for improving real-time torque distribution of dual motors in QPSO.

[0077] Figure 2 To improve the QPSO dual-motor torque dynamic compensation control strategy diagram.

[0078] Figure 3 To improve the computation time diagram of the QPSO algorithm.

[0079] Figure 4 The energy consumption curves for the three strategies under three typical operating conditions are shown: (a) NEDC; (b) CHTC-B; (c) CHTC-C. Detailed Implementation

[0080] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0081] This invention belongs to the field of motor control and intelligent optimization, specifically involving a real-time torque optimization and allocation method for a dual-motor drive system with a high-precision control cycle of 1ms. It aims to solve the problems of insufficient real-time performance and low allocation efficiency of traditional methods in high-speed dynamic response scenarios.

[0082] The method mainly includes two core modules:

[0083] (1) Real-time dual-motor torque distribution strategy based on improved QPSO:

[0084] This module optimizes the allocation of the total target torque between two motors by introducing structural improvements to the Quantum-behaved Particle Swarm Optimization (QPSO) algorithm. Improvements include a dynamic inertia weight adjustment mechanism, enhanced fitness function structure, and a search space compression strategy to improve the algorithm's convergence speed and stability under complex operating conditions. This strategy ensures that the total system torque meets the drive requirements while comprehensively considering the efficiency characteristic curves of the two motors, achieving high stability and high overall efficiency in torque allocation, thereby effectively reducing energy consumption and minimizing motor load fluctuations.

[0085] (2) Real-time torque dynamic compensation mechanism based on improved QPSO:

[0086] To address vehicle speed deviation caused by load disturbances, environmental changes, or system response lag during system operation, a real-time torque compensation strategy is designed. This strategy uses the dynamic error between the target vehicle speed and the current actual vehicle speed as input and rapidly calculates the optimal compensation torque based on an improved QPSO algorithm. By introducing an error tracking function, a torque adjustment sensitivity coefficient, and an adaptive step size update rule, the error under closed-loop speed control is minimized (approaching zero), thereby improving the dynamic response capability and speed control accuracy of the entire vehicle system under rapidly changing operating conditions.

[0087] In summary, the dual-motor torque real-time optimization allocation method proposed in this invention has the following advantages:

[0088] ① Supports 1ms-level high-frequency control cycles to meet industrial-grade real-time control requirements;

[0089] ②Balancing torque stability with motor efficiency improves system operating efficiency;

[0090] ③ Introduce a dynamic compensation mechanism to enhance the system's adaptability to disturbances and lag, and maintain high-speed following accuracy;

[0091] ④ The algorithm structure is lightweight and suitable for deployment in embedded controllers.

[0092] This method is applicable to various multi-motor collaborative application scenarios that require high-speed dynamic response, including electric vehicles, dual-motor servo control systems, and intelligent robots, and has good engineering practicality and promotion value.

[0093] The core idea of ​​the quantum behavior model introduced by QPSO is that a particle exists in a probability distribution state, and its position is determined by the following formula:

[0094]

[0095] in, mbest is a linear combination of the individual's historical best position and the group's best position. t Let u be the average of the best positions in the history of all particles, u~U(0,1) be uniformly distributed random variables, and β be the convergence factor.

[0096] This mechanism allows particles to "jump" to new solution space regions with a certain probability, significantly enhancing the global exploration capability and the ability to escape local optima during the search process. Simultaneously, by guiding the average optimal position, the algorithm maintains a certain degree of convergence and stability. Based on this, multiple enhancement mechanisms are introduced, combining the nonlinear efficiency model of the electric drive system, real-time requirements, and constraints, to construct a dual-motor torque optimization allocation process that combines global search capability with high real-time performance.

[0097] Example

[0098] This invention can provide the most efficient dual-motor torque distribution combination within 1ms under different torque requirements, achieving a comprehensive dual-motor working efficiency of 93.31307%.

[0099] The algorithm of this invention underwent a continuous stress test for 288 hours (1,036,800 seconds). During this test period, the average computation time was 0.78281 ms, and the proportion of computation times consistently controlled within 1 ms was greater than 99%, fully demonstrating that the method possesses industrial-grade real-time performance and stability, and is suitable for multi-motor systems with 1ms-level control requirements. The computation time is as follows... Figure 3 As shown.

[0100] To evaluate the energy consumption performance of different control strategies under various driving conditions, this invention compares the energy consumption data of a traditional torque-sharing strategy, a 10ms step-size strategy based on particle swarm optimization (PSO), and a 1ms step-size strategy based on improved QPSO under three typical driving conditions: NEDC, CHTC-B, and CHTC-C. The energy consumption curves are shown below. Figure 4 As shown in the figure. The results show that the proposed improved QPSO-1ms step size strategy exhibits significant energy consumption optimization advantages under all three operating conditions.

[0101] Specifically, under NEDC conditions, the total energy consumption of the improved QPSO strategy is 39976.49, a reduction of approximately 11.37% compared to the traditional strategy (45097.75) and approximately 11.99% compared to the PSO strategy (45182.70). Under CHTC-B conditions, the energy consumption of this strategy further decreases to 20442.50, a reduction of 19.58% compared to the traditional strategy (25423.94) and a significant reduction of 24.66% compared to the PSO strategy (27130.97). Under CHTC-C conditions, the energy consumption of the improved QPSO strategy is 65423.27, a reduction of 6.69% and 14.57% compared to the traditional strategy (69596.02) and the PSO strategy (76487.05), respectively.

[0102] In summary, the improved QPSO strategy, with its finer control step size and superior global search capability, achieves more efficient energy utilization under various operating conditions, demonstrating good generalization ability and engineering application potential, and possesses practical value for widespread application in dual-motor drive systems.

[0103] Compared with existing technologies (deep reinforcement learning strategies), the present invention has the following advantages, as shown in Table 1.

[0104] Table 1

[0105]

Claims

1. A torque optimal distribution method for a dual-motor drive system oriented to a 1 ms high-precision control period, characterized in that, Comprising the following steps: Step 1, based on the interpolation model and improved quantum particle swarm optimization algorithm QPSO, a real-time dual-motor torque distribution mechanism is constructed, and the individual optimal position and global optimum are updated according to the fitness of each particle, and the optimal comprehensive efficiency corresponding to the front and rear motor torque is output; Step 2, based on the real vehicle speed and target vehicle speed, based on the improved quantum particle swarm optimization algorithm QPSO, the particle search boundary is limited, the quantum behavior is evolved, a multi-dimensional compound fitness function is introduced, and the optimal compensation torque and optimal convergence curve are output.

2. The torque optimal distribution method for dual-motor drive system with 1 ms high-precision control period according to claim 1, characterized in that, Step 1 specifically includes the following steps: Step 1.1, construct an interpolation model and preprocess the efficiency function; The interpolation model is defined as follows: η i = f i (T i , ω i ), i = 1, 2 wherein η i represents the efficiency of the i-th motor, T i and ω i represent the torque and the rotational speed thereof, respectively, f i is an interpolation function; Step 1.2, initialize the particle swarm, generate the initial particle swarm within the set dual-motor torque constraint boundary, and introduce the total target torque constraint condition; Step 1.3, construct the fitness function with the optimization goal of maximizing the system total energy efficiency; Step 1.4, update the fitness of each particle in each iteration, if the current particle position is better than its historical optimal value, update its individual optimal p i ; at the same time, the global search space analysis is carried out on the whole population, and the center position P mean ; of the individual optimal of the whole particle is calculated Step 1.5, implement global jump search of particles by using the quantum position update model in the improved QPSO; Step 1.6, according to the double strategy of boundary truncation method and nonlinear penalty function method, correct or punish the particles that do not meet the following constraints in the iteration process; Step 1.7, when the maximum iteration number or convergence accuracy threshold is reached, select the solution with the highest fitness in the current particle swarm as the final optimal torque distribution output value, and send the optimal torque distribution scheme to the drive controller in real time as the reference output of the next control period.

3. The torque optimal distribution method for dual-motor drive system with 1 ms high-precision control period according to claim 2, characterized in that, Step 1.2 is specifically: within the set double-motor torque constraint boundary, initial particle swarm is generated by uniform distribution, each particle represents a set of feasible motor torque combination (T1, T2); at the same time, set initial convergence factor β0, individual optimal position p i and population global optimum g; Introduce the total target torque constraint condition: T total = T1 + T2 Make the generated particles meet the vehicle power demand.

4. The torque optimal distribution method for dual-motor drive system with 1 ms high-precision control period according to claim 3, characterized in that, Step 1.3 specifically: with the optimization goal of maximizing the system total energy efficiency, the fitness function adopts the weighted combination form of two-motor efficiency: F(T1,T2)=w1·η1(T1)+w2·η2(T2) Wherein, w1 and w2 are the weight coefficients of the front motor and rear motor efficiency, respectively, used to enhance the actual physical meaning of the optimization direction, η1 is the front motor efficiency, and η2 is the rear motor efficiency.

5. The torque optimal distribution method for dual-motor drive system with 1 ms high-precision control period according to claim 4, characterized in that, Step 1.5 specifically: the global jump search of particles is realized by using the quantum position update model in the improved QPSO, and the position evolution is controlled by the following formula: where P i is the individual optimal position, P mean is the center of all particle optimal values, u is a uniformly distributed random variable, and β is a dynamic contraction factor that is self-adaptively decayed during the iteration process to balance exploration and convergence.

6. The torque optimal distribution method for dual-motor drive system with 1 ms high-precision control period according to claim 5, characterized in that, Step 1.6 specifically: according to the double strategy of boundary truncation method and nonlinear penalty function method, correct or punish the particles that do not meet the following constraints in the iteration process: T min ≤T i ≤T max ,T1+T2=T total The form of the penalty function is: F penalized = F - λ · |T1+T2-T total | Wherein, λ is the penalty coefficient, F is the fitness function, which represents the evaluation index for measuring the pros and cons of the solution in the optimization process, and is adaptively increased to strengthen the constraint convergence speed.

7. The torque optimal distribution method for dual-motor drive system with 1 ms high-precision control period according to claim 1, characterized in that, Step 2 specifically includes the following steps: Step 2.1, initialize the improved quantum particle swarm optimization algorithm QPSO; Algorithm control parameter setting: set the population size, maximum iteration number and quantum behavior dynamic contraction factor β to balance the search ability and convergence speed; Physical boundary constraint definition: set the upper and lower boundaries of the torque optimization results to conform to the actual physical ability of the motor; Memory and data structure optimization: pre-allocate structure to improve operation efficiency and realize fast storage and access of convergence curve; Step 2.2, initialize the particle swarm and construct the boundary constraint mechanism; Step 2.3, construct quantum behavior evolution mechanism, introduce random disturbance and global mean guidance; Step 2.4, construct multi-objective fitness function considering speed error, torque stability and speed change smoothness; Step 2.5, adopt double condition termination mechanism, evaluate the influence of quantum factor on convergence speed and search ability according to parameter sensitivity analysis, and verify the robustness.

8. The torque optimal distribution method for dual-motor drive system with 1 ms high-precision control period according to claim 7, characterized in that, Step 2.2 is as follows: particle swarm initialization: uniform distribution is used for solution vector initialization: X = (ub-lb)·rand(pop,dim)+lb Where, X is the initial solution vector of the particle; ub is the upper boundary constraint of the solution vector; lb is the lower boundary constraint of the solution vector; rand(pop,dim) is a function for generating uniformly distributed random numbers; pop is the population size, that is, the number of particles; dim is the number of variables included in each particle solution vector; Boundary constraint processing: introduce vectorized boundary correction function: X corrent = min(max(X, lb), ub) where X corrent is the current particle position matrix; Individual optimal value updating mechanism: adopt local fitness criterion for individual optimal updating: If f(x new ) < f(pBest i ), then update pBest i = x new .

9. The torque optimal distribution method for dual-motor drive system with 1 ms high-precision control period according to claim 7, characterized in that, Step 2.3 is as follows: global mean calculation: calculate the mean value of the optimal historical solution of the current iteration population: where m best is the mean of the global best solution history, used to represent the center position of the current population's overall best solution; pBest i is the historical best position of the i-th particle, i.e., the best solution vector found so far by the particle Position evolution equation: Where r ∈ {0, 1}, u ~ U(0, 1) is random disturbance; Dual-mode perturbation mechanism: by (-1) r Search space direction perturbation is implemented.

10. The torque optimal distribution method for dual-motor drive system with 1 ms high-precision control period according to claim 7, characterized in that, Step 2.4 is as follows: construct multi-objective fitness function considering speed error, torque stability and speed change smoothness, which is as follows: where v ref (t) is the value of the desired speed at time t; v real (t) is the value of the actual speed at time t; λ1 is a torque stability weight coefficient for balancing the importance of the speed error and the torque performance; T comp is the amplitude of the compensation torque; λ2 is a speed change smoothness weight coefficient for balancing the smoothness of the speed error and the speed change rate; Δv real (t) is the rate of change of the actual speed.

Images