Permanent magnet synchronous motor predictive control method and system based on sampling frequency optimization

Abstract

The application discloses a permanent magnet synchronous motor predictive control method and system based on sampling frequency optimization, relates to the technical field of permanent magnet synchronous motors, and comprises the following steps: determining an adaptive sampling frequency according to a current operation scene of the permanent magnet synchronous motor, and setting an optimization time window; obtaining historical monitoring data in a previous optimization time window, predicting a state trajectory of the permanent magnet synchronous motor in the current optimization time window according to the historical monitoring data, and calculating voltage vector sequences and action times that need to be output in a current sampling period with the minimum current tracking error at the end of the optimization time window as the target; generating pulse width modulation waveforms according to the voltage vector sequences and the action times, and outputting the pulse width modulation waveforms to an inverter to drive the permanent magnet synchronous motor. The technical problems of the prior art, such as difficulty in adapting to variable operation conditions of the permanent magnet synchronous motor, single-period optimization, control quantity fluctuation, strong model dependency and insufficient robustness, are solved.

Description

Technical Field

[0001] This application relates to the field of permanent magnet synchronous motor technology, specifically to a predictive control method and system for permanent magnet synchronous motors based on sampling frequency optimization. Background Technology

[0002] With the rapid development of industrial automation and intelligent manufacturing technologies, permanent magnet synchronous motors have been widely used in electric vehicles, CNC machine tools, aerospace and home appliances due to their excellent characteristics such as high power density, high efficiency and wide speed range. In the control strategy of permanent magnet synchronous motors, model predictive control has gradually become a research hotspot for high-performance motor drive systems due to its advantages such as intuitive concept, easy handling of multivariable constraints and fast dynamic response.

[0003] However, traditional model predictive control methods typically employ a fixed sampling frequency in practical applications, failing to fully consider the dynamic characteristics of permanent magnet synchronous motors (PMSMs) under different operating scenarios. Under low-speed, light-load conditions, the motor's electrical time constant is relatively large, and current changes are relatively slow. Excessively high sampling frequencies not only waste computational resources but also introduce unnecessary measurement noise. Conversely, under high-speed, heavy-load, or sudden-load conditions, the motor's dynamic response accelerates, and a fixed sampling frequency struggles to capture rapidly changing current states, leading to decreased prediction accuracy and deteriorated control performance.

[0004] On the other hand, traditional predictive control often takes the current error within a single sampling period as the optimization target, lacking a global consideration of the state evolution trend over multiple future sampling periods. This can easily lead to short-sighted optimization problems, causing the control quantity to fluctuate drastically between adjacent periods, resulting in large current ripple and torque pulsation.

[0005] Furthermore, existing state prediction technologies largely rely on mathematical models of motors. In the event of parameter mismatch or external disturbances, model mismatch errors accumulate and amplify, affecting the robustness of predictive control. Summary of the Invention

[0006] This application provides a predictive control method and system for permanent magnet synchronous motors based on sampling frequency optimization. This solves the technical problems in the prior art where fixed sampling frequency is difficult to adapt to the changing operating conditions of permanent magnet synchronous motors, single-cycle optimization leads to fluctuations in control quantity, and the model has strong dependence but insufficient robustness.

[0007] The technical solution to the above-mentioned technical problems in this application is as follows: In a first aspect, this application provides a predictive control method for a permanent magnet synchronous motor based on sampling frequency optimization, the method comprising: The appropriate sampling frequency is determined based on the current operating scenario of the permanent magnet synchronous motor, and an optimized time window is set based on the appropriate sampling frequency. Within each sampling period, historical monitoring data from the previous optimized time window is acquired. Based on the historical monitoring data, the state trajectory of the permanent magnet synchronous motor within the current optimized time window is predicted. With the goal of minimizing the current tracking error at the end of the optimized time window, the voltage vector sequence and duration to be output within the current sampling period are calculated based on the state trajectory. A pulse width modulation waveform is generated based on the voltage vector sequence and the duration of action, and output to the inverter to drive the permanent magnet synchronous motor.

[0008] Secondly, this application provides a predictive control system for a permanent magnet synchronous motor based on sampling frequency optimization, including: The sampling frequency determination module is used to determine the appropriate sampling frequency based on the current operating scenario of the permanent magnet synchronous motor, and to set an optimization time window based on the appropriate sampling frequency. The state trajectory prediction module is used to acquire historical monitoring data from the previous optimized time window in each sampling period, predict the state trajectory of the permanent magnet synchronous motor in the current optimized time window based on the historical monitoring data, and calculate the voltage vector sequence and action time to be output in the current sampling period based on the state trajectory with the goal of minimizing the current tracking error at the end of the optimized time window. The synchronous motor drive module is used to generate a pulse width modulation waveform based on the voltage vector sequence and the duration of action, and output it to the inverter to drive the permanent magnet synchronous motor.

[0009] This application provides one or more technical solutions, which have at least the following technical effects or advantages: This application provides a predictive control method and system for permanent magnet synchronous motors based on sampling frequency optimization. First, an appropriate sampling frequency is determined according to the current operating scenario of the permanent magnet synchronous motor, achieving dynamic matching between the sampling frequency and the motor's dynamic characteristics. The sampling frequency is reduced under low-speed, light-load conditions to decrease computational burden and measurement noise, while it is increased under high-speed, heavy-load, or sudden load changes to enhance state capture capabilities, thereby improving computational resource utilization efficiency while maintaining control accuracy. Second, a long short-time memory (LSTM) network is used to construct a state prediction model, utilizing historical monitoring data sequences to predict the state trajectory within the future optimization time window, enhancing the robustness of state prediction under parameter mismatch and external disturbances. Third, with the goal of minimizing the current tracking error at the end of the optimization time window, a value function is constructed by combining the instantaneous tracking errors at each sampling moment within the window, achieving multi-cycle global optimization. This avoids the short-sightedness problem caused by traditional single-cycle optimization and effectively suppresses drastic fluctuations in the control quantity. Finally, by optimizing the target of minimizing the effective value of current ripple within the time window, rationally allocating the zero vector action time, and adopting a symmetrical waveform generation strategy to generate pulse width modulation waveforms, the current ripple and torque pulsation are further reduced, thereby improving the operating quality and control performance of the permanent magnet synchronous motor.

[0010] Through the above technical solutions, this application achieves the beneficial effects of adaptive adjustment of sampling frequency according to operating conditions, multi-cycle global optimization to reduce current ripple, and data-driven prediction to improve control robustness. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a flowchart illustrating the predictive control method for permanent magnet synchronous motors based on sampling frequency optimization provided in an embodiment of this application. Figure 2 This is a schematic diagram of the structure of a predictive control system for a permanent magnet synchronous motor based on sampling frequency optimization provided in an embodiment of this application.

[0013] The components represented by each number in the attached diagram are explained below: Sampling frequency determination module 11, state trajectory prediction module 12, synchronous motor drive module 13. Detailed Implementation

[0014] This application provides a predictive control method and system for permanent magnet synchronous motors based on sampling frequency optimization. This method addresses the technical problems of existing fixed sampling frequencies being unable to adapt to the changing operating conditions of permanent magnet synchronous motors, single-cycle optimization leading to control quantity fluctuations, and strong model dependence with insufficient robustness.

[0015] Example 1, as Figure 1 As shown, this application provides a predictive control method for permanent magnet synchronous motors based on sampling frequency optimization, including: S10: Determine the appropriate sampling frequency based on the current operating scenario of the permanent magnet synchronous motor, and set an optimized time window based on the appropriate sampling frequency; In this embodiment, the current operating scenario includes the motor speed, load torque, and operating condition type. The determination of the sampling frequency takes into account the balance between the motor's dynamic response characteristics and computational resource constraints, and then the optimization time window is set according to the appropriate sampling frequency.

[0016] Specifically, step S10 in the method includes: Obtain the current operating scenario of the permanent magnet synchronous motor, wherein the operating scenario includes at least low-speed light-load condition, rated condition, high-speed heavy-load condition and sudden change condition; The appropriate sampling frequency is determined based on the current operating scenario. The appropriate sampling frequency is equal to a preset baseline sampling frequency multiplied by a dynamic adjustment factor, where the preset baseline sampling frequency is 1.25 seconds / time. When the permanent magnet synchronous motor operates under low-speed, light-load conditions, the dynamic adjustment factor is set to two. When the permanent magnet synchronous motor operates under rated conditions, the dynamic adjustment factor is set to four. When the permanent magnet synchronous motor operates under high-speed, heavy-load conditions or sudden change conditions, the dynamic adjustment factor is set to eight. The optimized time window is set according to the adapted sampling frequency, and the length of the optimized time window is set to the sum of a preset number of sampling periods, wherein the preset number is an integer greater than or equal to 5, and the sampling period is the time span between adjacent sampling points.

[0017] In this embodiment, firstly, by monitoring the motor speed and load torque in real time and combining them with a preset operating condition discrimination threshold, the current operating condition type is automatically identified.

[0018] For low-speed, light-load conditions, where the motor speed is less than 30% of the rated speed and the load torque is less than 20% of the rated torque, the electrical time constant is relatively large and the current changes gradually. Using a lower sampling frequency, i.e., twice the reference frequency, can effectively reduce the processor's computational load and AD conversion noise interference. For rated conditions, where both the motor speed and load torque are near their rated values, using four times the reference frequency achieves a balance between control performance and computational overhead. For high-speed, heavy-load conditions, where the motor speed exceeds 80% of the rated speed or the load torque exceeds 70% of the rated torque, as well as sudden change conditions where the load torque change rate exceeds a preset threshold, using eight times the reference frequency ensures accurate capture of rapid dynamic processes.

[0019] Secondly, the length of the optimization time window is set to the sum of multiple sampling periods. If the window is too short, it will be difficult to reflect the advantages of multi-period optimization, while if the window is too long, it will lead to the accumulation of prediction errors and an increased burden on online computation.

[0020] For example, taking high-speed heavy-load conditions as an example, the adaptive sampling frequency is 10 kHz, the sampling period is 100 microseconds, and the optimization time window length is 500 microseconds, which is sufficient to cover the typical range of the motor stator electrical time constant, so that the optimization target can fully consider the state evolution trend in multiple future control cycles.

[0021] S20: In each sampling period, acquire historical monitoring data from the previous optimized time window, predict the state trajectory of the permanent magnet synchronous motor in the current optimized time window based on the historical monitoring data, and calculate the voltage vector sequence and action time to be output in the current sampling period based on the state trajectory with the goal of minimizing the current tracking error at the end of the optimized time window. In this embodiment, within each sampling period, historical monitoring data within the previous optimization time window and the actual applied voltage vector sequence within the previous period are first acquired. Then, a state prediction model is constructed based on a long short-term memory network, and the historical monitoring data is used as the input sequence to predict the stator current prediction value at each sampling moment within the current optimization time window, thus forming a continuous state trajectory.

[0022] Furthermore, with the goal of minimizing the current tracking error at the end of the optimized time window, the voltage vector sequence to be output and the duration of action within the current sampling period are calculated based on the generated state trajectory.

[0023] The state trajectory of the permanent magnet synchronous motor within the current optimized time window, predicted based on the historical monitoring data, includes: Obtain historical monitoring data within the previous optimized time window, wherein the historical monitoring data includes direct-axis current values ​​and quadrature-axis current values ​​collected at each sampling time within the previous optimized time window; A state prediction model is constructed based on a long short-term memory network, wherein the state prediction model is used to output a state data sequence within a future time window based on the input historical monitoring data sequence; The historical monitoring data are arranged in chronological order to form an input sequence, and the input sequence is fed into the state prediction model for forward calculation to predict the state data sequence within the current optimization time window as the state trajectory.

[0024] In this embodiment of the application, firstly, the historical monitoring data includes the direct-axis current value and quadrature-axis current value at each sampling moment within the previous optimized time window. The direct-axis current value and quadrature-axis current value are collected in real time by current sensors and obtained through coordinate transformation, which can fully reflect the electromagnetic state evolution process of the motor within the previous time window.

[0025] Secondly, Long Short-Term Memory (LSTM) networks selectively retain and forget historical information through gating mechanisms, enabling them to capture long-term dependencies between state variables of permanent magnet synchronous motors (PMSMs). Specifically, historical monitoring data is arranged chronologically to form an input sequence, with the dimension of the input sequence corresponding to the number of sampling points in the historical monitoring data. Each sampling point contains two feature dimensions: direct-axis current value and quadrature-axis current value. The input sequence is fed into the state prediction model for forward computation. The encoder processes each sampling point in the input sequence sequentially, updating the cell state and hidden state, and finally outputting an encoded vector. The decoder uses the encoded vector as initial conditions, combined with the prediction output of the previous time step, to progressively generate the current prediction values ​​for each sampling time step within the current optimization time window, forming a continuous state trajectory.

[0026] Furthermore, a state prediction model is constructed based on Long Short-Term Memory (LSTM) networks, including: Acquire the monitoring data sequence of the permanent magnet synchronous motor during its historical operating period, and divide the monitoring data sequence into multiple training sample pairs, each training sample pair containing an input sequence and a target output sequence; Construct a long short-term memory network model, wherein the long short-term memory network model includes an input layer, at least one long short-term memory hidden layer, and an output layer; The input sequences from the training sample pairs are sequentially input into the Long Short-Term Memory (LSTM) network model. The hidden state and cell state are updated time-by-time through the calculation of the forget gate, input gate, output gate, and cell state in the LSM hidden layer. The updated hidden states are mapped to a predicted state sequence through the output layer, and the error between the predicted state sequence and the target output sequence is calculated. With the goal of minimizing the error, the network parameters of the Long Short-Term Memory Network model are updated using the backpropagation algorithm until a preset convergence condition is met. The trained Long Short-Term Memory Network model is then used as the state prediction model.

[0027] In this embodiment, firstly, a monitoring data sequence of the permanent magnet synchronous motor during its historical operating period is acquired. This monitoring data sequence covers the entire operating range of the motor, including direct-axis current and quadrature-axis current values ​​under different speeds, load torques, and parameter perturbation conditions. The monitoring data sequence is then divided into multiple training sample pairs using a sliding window method. The length of the input sequence corresponds to the number of sampling points in the previous optimization time window, and the length of the target output sequence corresponds to the number of sampling points in the current optimization time window. Adjacent sample pairs overlap by a preset proportion to ensure data utilization efficiency.

[0028] Secondly, a Long Short-Term Memory (LSTM) network model is constructed. The input layer receives a two-dimensional current feature vector, namely the direct-axis current value and the quadrature-axis current value. The LSM hidden layer adopts a multi-layer stacked structure, with each layer containing multiple memory units. The forget gate controls the retention of historical information, the input gate adjusts the injection ratio of new information, and the output gate determines the output of the hidden state at the current moment. The cell state serves as a long-term memory carrier throughout the entire time series processing. The output layer maps the final hidden state to a two-dimensional current prediction vector.

[0029] Furthermore, during model training, the input sequence of the training sample pairs is fed into the Long Short-Term Memory (LSTM) network model time-by-time. At each time step, the forget gate calculates the forgetting weight based on the current input and the hidden state of the previous time step, determining the historical information to be discarded in the cell state. The input gate calculates the update weights of candidate cell states and generates new candidate information by combining them with the current input. The cell state is updated according to the forgetting weight and the input weight, achieving selective retention and refresh of long-term memory. The output gate calculates the output weight based on the updated cell state and the current input, generating the hidden state at the current time step. The hidden states at each time step are then passed to the output layer to obtain the predicted state sequence.

[0030] Furthermore, the root mean square error between the predicted state sequence and the target output sequence is calculated as the loss function. An adaptive moment estimation optimization algorithm is used to update the network parameters. The learning rate is dynamically adjusted according to the training process, and gradient clipping is introduced to prevent gradient explosion. An early stopping mechanism is used to monitor the validation set error during training. Training is terminated when the validation set error no longer decreases for several consecutive iterations. The network parameters at this time are taken as the optimal parameters to obtain the trained state prediction model.

[0031] For example, a state prediction model is constructed based on a Long Short-Term Memory (LSTM) network. The input layer has a dimension of 2, corresponding to the direct-axis current value and the quadrature-axis current value. The hidden layer adopts a two-layer stacked structure, with each layer containing 128 memory units. The output layer has a dimension of 2, predicting the current value within the future optimization time window. The learning rate is initialized to 0.001, the batch size is set to 64, the number of training iterations is 2000 rounds, and early stopping is triggered when the validation set error does not improve for 50 consecutive rounds. Finally, the root mean square error of the current prediction on the test set is less than 0.05 amperes.

[0032] Furthermore, with the goal of minimizing the current tracking error at the end of the optimized time window, the voltage vector sequence to be output and its duration within the current sampling period are calculated based on the state trajectory, including: Extract the predicted current value at the end of the optimized time window and the predicted current value at each sampling time within the optimized time window from the state trajectory. A value function is constructed based on the endpoint tracking error between the predicted current value and the given current value at the end of the optimized time window, and the instantaneous tracking error between the predicted current value and the given current value at each sampling time within the optimized time window. By enumerating candidate voltage vector sequences and solving for the duration of each vector, the optimal voltage vector sequence and its corresponding duration that minimize the value function are found. The optimal voltage vector sequence and its duration obtained from the solution are used as the voltage vector sequence and duration to be output within the current sampling period.

[0033] In this embodiment, firstly, the predicted current value at the end of the optimized time window is extracted from the state trajectory as the endpoint state, and at the same time, the predicted current value at each sampling time within the window is extracted as the instantaneous state. The given current value is calculated by the speed loop controller based on the speed deviation and includes two components: the direct-axis current given value and the quadrature-axis current given value.

[0034] Secondly, a value function is constructed. The endpoint tracking error is weighted using a quadratic weight matrix, emphasizing the steady-state accuracy requirement at the end of the window. The instantaneous tracking error is introduced with a time decay factor, giving higher weight to errors closer to the current time, while taking into account the rapid response characteristics in the dynamic process. The specific form of the value function is the approximation of the weighted sum of squares of the endpoint tracking error plus the integral of the weighted sum of squares of the instantaneous tracking error. The weight coefficients are adaptively adjusted according to the operating condition. Under high-speed heavy-load conditions, the endpoint weight is appropriately increased to suppress overshoot, while under low-speed light-load conditions, the instantaneous weight is increased to improve smoothness.

[0035] Furthermore, the generation of candidate voltage vector sequences is based on eight basic switching state combinations of a two-level inverter for a permanent magnet synchronous motor, including six active vectors and two zero vectors. For each candidate vector sequence, the volt-second balance principle is used to solve for the duration of each vector. That is, within each sampling period, the duration of adjacent active and zero vectors is allocated so that the average value of the synthesized voltage vector is equal to the desired voltage vector.

[0036] Specifically, based on the endpoint tracking error between the predicted current value and the given current value at the end of the optimized time window, and the instantaneous tracking error between the predicted current value and the given current value at each sampling time within the optimized time window, a value function is constructed, including: The sum of squares of the endpoint tracking errors is used as the first term of the value function, and the first term is multiplied by the first weighting coefficient to obtain the first weighted value; The sum of the squares of the instantaneous tracking errors at each sampling moment within the optimization time window is taken as the second term of the value function, and the second term is multiplied by the second weight coefficient to obtain the second weighted value, wherein the sum of the first weight coefficient and the second weight coefficient is 1; The first weighted value is added to the second weighted value to obtain the value of the value function.

[0037] In this embodiment, firstly, the endpoint tracking error reflects the deviation between the motor current and the expected given value at the end of the optimization time window. Taking the sum of squares of the endpoint tracking error as the first term of the value function can effectively amplify the influence of larger errors and prompt the optimization process to prioritize the elimination of steady-state deviations. The first weighting coefficient is used to adjust the relative importance of endpoint accuracy in the overall optimization objective, and its value ranges from 0 to 1.

[0038] Secondly, the instantaneous tracking error represents the cumulative dynamic deviation between the current and the given value at each sampling moment within the optimization time window. The sum of the squares of the instantaneous tracking errors at each moment is used as the second term of the value function, which can constrain the trajectory smoothness throughout the dynamic process and avoid the situation of drastic current fluctuations or excessive torque pulsation. The second weighting coefficient and the first weighting coefficient are complementary, and their sum is fixed at 1.

[0039] Furthermore, in practical applications, the specific values ​​of the first and second weighting coefficients are dynamically adjusted according to the current operating conditions. For example, for high-speed heavy-load conditions, the first weighting coefficient is set to 0.7, and the second weighting coefficient is set to 0.3 accordingly, prioritizing the steady-state accuracy at the endpoint to suppress speed overshoot; for rated conditions, both are set to 0.5 to achieve a balance between steady-state accuracy and dynamic smoothness; for low-speed light-load conditions, the first weighting coefficient is reduced to 0.3, and the second weighting coefficient is increased to 0.7, focusing on improving the smoothness of the current response to reduce torque ripple.

[0040] Furthermore, by enumerating candidate voltage vector sequences and solving for the action time of each vector, the optimal voltage vector sequence and its corresponding action time that minimize the value function are found, including: All effective voltage vectors that the inverter can output are combined in pairs to generate multiple candidate vector combinations. Each candidate vector combination contains a first candidate effective vector and a second candidate effective vector. A zero vector is assigned to each candidate vector combination to form a complete candidate voltage vector sequence, wherein the candidate voltage vector sequence is arranged in the order of the first candidate effective vector, the second candidate effective vector, and the zero vector; For each candidate voltage vector sequence, the action time of the first candidate effective vector, the action time of the second candidate effective vector, and the total action time of the zero vector that minimize the value function are calculated, where the sum of the three action times is equal to the duration of the sampling period; For each candidate voltage vector sequence, the obtained action time is substituted into the discrete mathematical model of the permanent magnet synchronous motor to predict the current trajectory under the action of the current candidate voltage vector sequence and calculate the actual value function value. The candidate voltage vector sequence and its duration that minimize the actual value function are selected as the optimal voltage vector sequence and its corresponding duration.

[0041] In this embodiment, when the permanent magnet synchronous motor is powered by a two-level voltage source inverter, the switching state combinations of the three-phase bridge arms of the inverter generate eight basic voltage vectors, including six effective vectors and two zero vectors. The six effective vectors divide the voltage space plane into six sectors, and each effective vector has a fixed amplitude and phase angle. By combining the six effective vectors in pairs, thirty candidate vector combinations can be generated. Each combination contains a first candidate effective vector and a second candidate effective vector, and the two effective vectors are usually vectors from adjacent sectors or vectors from opposite sectors to ensure the continuous adjustability of the synthesized voltage vector.

[0042] Secondly, a zero vector is assigned to each candidate vector combination to form a complete candidate voltage vector sequence. The selection of the zero vector includes two redundant states: all three-phase upper bridge arms are fully conducting or all three-phase lower bridge arms are fully conducting. The voltage vectors generated by the two states are both zero vectors, but the switching loss characteristics are different. The candidate voltage vector sequence is arranged in the order of the first candidate effective vector, the second candidate effective vector, and the zero vector. This arrangement conforms to the time sequence requirements of space vector modulation, which facilitates the subsequent calculation of the action time and the generation of the pulse sequence.

[0043] Furthermore, for each candidate voltage vector sequence, an optimization problem is established with the goal of minimizing the value function. Let the sampling period be Ts, the action time of the first candidate effective vector be t1, the action time of the second candidate effective vector be t2, and the total action time of the zero vector be t0. Then, the action time must satisfy the equality constraint t1 + t2 + t0 = Ts, and the inequality constraints t1 ≥ 0, t2 ≥ 0, and t0 ≥ 0. The discrete current equation of the permanent magnet synchronous motor in the rotating coordinate system is embedded as an equality constraint in the optimization problem, describing the change law of the current state under the action of a specific voltage vector, including motor parameters such as stator resistance, direct-axis inductance, and quadrature-axis inductance.

[0044] Specifically, in the solution process, a strategy combining analytical and numerical methods is adopted. For a given candidate voltage vector sequence, the instantaneous tracking error term in the value function is first ignored, and only the endpoint tracking error is considered to derive an analytical expression for the action time. Then, the analytical solution is used as the initial value, and a sequential quadratic programming algorithm is used for numerical optimization. The action time is iteratively corrected to minimize the instantaneous tracking error. This hybrid solution strategy significantly reduces the online computational burden while ensuring computational accuracy. The solution time for a single candidate sequence can be controlled in the microsecond range.

[0045] Furthermore, the obtained action time is substituted into the discrete mathematical model of the permanent magnet synchronous motor to perform recursive prediction of the current trajectory. The discrete mathematical model takes the measured current value at the current sampling time as the initial state and calculates the predicted current value at each time step by step in the order of the action time t1 of the first candidate effective vector, the action time t2 of the second candidate effective vector, and the action time t0 of the zero vector. The predicted current trajectory is compared with the given current value, the endpoint tracking error and the instantaneous tracking error are calculated, and then the actual value function value corresponding to the candidate voltage vector sequence is obtained.

[0046] Traverse all candidate voltage vector sequences and calculate their actual value function value. Select the candidate sequence that minimizes the value function value as the optimal voltage vector sequence by comparison. If multiple candidate sequences have similar value function values, prioritize the sequence with fewer switching times to reduce the inverter's switching losses. The final optimal voltage vector sequence and its duration are the switching control commands that the inverter needs to execute in the current sampling period.

[0047] Furthermore, the optimal voltage vector sequence and its duration obtained from the solution are used as the voltage vector sequence and duration to be output within the current sampling period, including: With the goal of minimizing the effective value of current ripple within the optimization time window, the allocation ratio of the action time of the first zero vector and the action time of the second zero vector is determined, and the total action time of the zero vector is allocated as the action time of the first zero vector and the action time of the second zero vector, wherein the first zero vector is the zero vector corresponding to the turn-off of all switches, and the second zero vector is the zero vector corresponding to the turn-on of all switches. The allocated first zero vector action time and second zero vector action time are combined with the action times of the first effective vector and second effective vector to form the complete voltage vector sequence and action time that needs to be output within the current sampling period.

[0048] In this embodiment, the zero vector allocation strategy affects the switching loss distribution and current ripple characteristics of the inverter. The first zero vector corresponds to the state where all three-phase upper bridge arms are off and all three-phase lower bridge arms are on, and the second zero vector corresponds to the state where all three-phase upper bridge arms are on and all three-phase lower bridge arms are off. In both zero vector states, the motor terminal voltage is zero, but the conduction loss and switching loss of the power devices are different.

[0049] First, to minimize the effective value of the current ripple within the optimization time window, a sub-optimization problem is established with the zero vector allocation ratio as the optimization variable. The calculation of the effective value of the current ripple is based on a discrete mathematical model. During the zero vector's action, since the terminal voltage is zero, the rate of change of the current is determined only by the back electromotive force and the voltage drop across the resistor, and the ripple amplitude is positively correlated with the duration of the zero vector's action. By analyzing the trajectory of the current vector during the first and second zero vector actions, an analytical expression for the effective value of the ripple is derived. This expression exhibits convex function characteristics with respect to the zero vector allocation ratio, and a unique optimal solution exists.

[0050] In the specific solution, based on the current value, speed value and rotor position angle at the current sampling moment, the slope of the current change under the two zero vector actions is calculated, and then the optimal distribution ratio that minimizes the effective value of ripple is determined. This ratio changes dynamically with the operating conditions. Under high-speed conditions, the back electromotive force is large and the current changes drastically during the zero vector period. Appropriately shortening the total action time of the zero vector or adjusting the distribution ratio can effectively suppress ripple. Under low-speed conditions, a longer action time of the zero vector is allowed without significantly increasing ripple.

[0051] Secondly, the total action time of the zero vector is decomposed into the first zero vector action time t01 and the second zero vector action time t02 according to the optimal allocation ratio, satisfying that t01 plus t02 equals t0. Combining the previously solved first effective vector action time t1 and second effective vector action time t2, a complete seven-segment or five-segment space vector modulation sequence is constructed. The segment sequence is arranged in the order of t01 / 2, t1, t2, t02, t2, t1, t01 / 2. This symmetrical structure can effectively reduce the harmonic content. The five-segment sequence sets t01 or t02 to zero, reducing the number of switching operations but slightly increasing the ripple, which is suitable for occasions where the switching frequency is limited.

[0052] Finally, the generated voltage vector sequence and its duration are output to the inverter drive circuit as a pulse width modulation signal to control the on and off of the power switches, achieving precise current control of the permanent magnet synchronous motor. This optimization process is repeated in each sampling period to form a closed-loop predictive control, enabling the motor current to quickly track the given value while maintaining low current ripple and switching losses.

[0053] S30: Generate a pulse width modulation waveform based on the voltage vector sequence and the duration of action, and output it to the inverter to drive the permanent magnet synchronous motor.

[0054] In this embodiment, the generation of the pulse width modulation waveform is based on the principle of space vector modulation. The voltage vector sequence and its duration obtained by optimization are converted into switching control signals for the three-phase bridge arms of the inverter and output to the inverter to drive the permanent magnet synchronous motor.

[0055] Specifically, step S30 in the method includes: Each switching cycle is divided into a first half cycle and a second half cycle. The fixed waveform transmission sequence of the first half cycle is set to output the first zero vector, the first effective vector, the second effective vector, and the second zero vector in sequence. The fixed waveform transmission sequence of the second half cycle is set to output the second zero vector, the second effective vector, the first effective vector, and the first zero vector in sequence. Within the current sampling period, according to the fixed waveform generation sequence, a corresponding pulse width modulation waveform is generated based on the voltage vector sequence and the duration of action; and the generated pulse width modulation waveform is output to the inverter to control the switching transistors in the inverter to drive the permanent magnet synchronous motor.

[0056] In this embodiment, a symmetrical seven-segment space vector modulation strategy is first used to generate a pulse width modulation waveform. This strategy effectively reduces the harmonic content of the output voltage and the current ripple of the motor by dividing each switching cycle into a first half-cycle and a second half-cycle and arranging them in a mirror image with the midpoint as the axis of symmetry. The waveform output sequence of the first half-cycle is the first zero vector, the first effective vector, the second effective vector, and the second zero vector, respectively. In the second half-cycle, the second zero vector, the second effective vector, the first effective vector, and the first zero vector are output in the reverse order, forming a seven-segment symmetrical structure of t01 / 2, t1, t2, t02, t2, t1, and t01 / 2.

[0057] Specifically, in the first half of the cycle, the action time of the first zero vector is t01 / 2, the action time of the first effective vector is t1, the action time of the second effective vector is t2, and the action time of the second zero vector is t02; in the second half of the cycle, the action time of each vector is equal to that of the first half of the cycle, but the order of arrangement is completely reversed.

[0058] Furthermore, within the current sampling period, based on the action time of each vector obtained from the aforementioned optimization solution, the time node of each switching moment is calculated, and the pulse width modulation signal of the corresponding three-phase bridge arm is generated.

[0059] For example, taking phase A as an example, by comparing the conduction state of the upper bridge arm switch of phase A with the triangular carrier signal during each vector action period, a corresponding PWM pulse sequence is generated. The pulse generation of phases B and C follows the same principle, but the phases differ by 120 electrical degrees. After dead-time insertion and drive level conversion, the pulse signals of each phase are output to the gate drive circuit of the power switch of the inverter to control the conduction and turn-off of the insulated gate bipolar transistor or metal-oxide-semiconductor field-effect transistor.

[0060] In summary, compared with existing technologies, this application reduces dynamic response latency to within one sampling cycle by increasing the sampling frequency to a multiple of the switching frequency and performing rolling optimization in each sampling cycle; it significantly improves steady-state accuracy and reduces current ripple and torque pulsation by setting an optimization time window covering multiple sampling cycles and constructing a value function that includes endpoint error and full-process ripple penalty; it achieves balanced utilization of computing resources and reduces peak load by predicting state trajectories through long short-time memory networks and distributing computational tasks across sampling cycles; and it enhances robustness to changes in operating conditions and parameter drift through zero-vector allocation online optimization and adaptive parameter correction.

[0061] In summary, the embodiments of this application have at least the following technical effects: This application provides a predictive control method for permanent magnet synchronous motors (PMSMs) based on sampling frequency optimization. First, an appropriate sampling frequency is determined according to the current operating scenario of the PMSM, achieving dynamic matching between the sampling frequency and the motor's dynamic characteristics. The sampling frequency is reduced under low-speed, light-load conditions to decrease computational burden and measurement noise, while it is increased under high-speed, heavy-load, or sudden load changes to enhance state capture capabilities, thereby improving computational resource utilization efficiency while maintaining control accuracy. Second, a long short-time memory (LSTM) network is used to construct a state prediction model, utilizing historical monitoring data sequences to predict the state trajectory within the future optimization time window, enhancing the robustness of state prediction under parameter mismatch and external disturbances. Third, with the goal of minimizing the current tracking error at the end of the optimization time window, a value function is constructed by combining the instantaneous tracking errors at each sampling moment within the window, achieving multi-cycle global optimization. This avoids the short-sightedness problem caused by traditional single-cycle optimization and effectively suppresses drastic fluctuations in the control quantity. Finally, by optimizing the target of minimizing the effective value of current ripple within the time window, rationally allocating the zero vector action time, and adopting a symmetrical waveform generation strategy to generate pulse width modulation waveforms, the current ripple and torque pulsation are further reduced, thereby improving the operating quality and control performance of the permanent magnet synchronous motor.

[0062] Through the above technical solutions, this application achieves the beneficial effects of adaptive adjustment of sampling frequency according to operating conditions, multi-cycle global optimization to reduce current ripple, and data-driven prediction to improve control robustness.

[0063] Example 2, as Figure 2 As shown, based on the same inventive concept as the sampling frequency-optimized predictive control method for permanent magnet synchronous motors provided in Embodiment 1, this application also provides a sampling frequency-optimized predictive control system for permanent magnet synchronous motors, including: The sampling frequency determination module 11 is used to determine the appropriate sampling frequency based on the current operating scenario of the permanent magnet synchronous motor, and to set an optimized time window based on the appropriate sampling frequency. The state trajectory prediction module 12 is used to acquire historical monitoring data in the previous optimization time window in each sampling period, predict the state trajectory of the permanent magnet synchronous motor in the current optimization time window based on the historical monitoring data, and calculate the voltage vector sequence and action time to be output in the current sampling period based on the state trajectory with the goal of minimizing the current tracking error at the end of the optimization time window. The synchronous motor drive module 13 is used to generate a pulse width modulation waveform according to the voltage vector sequence and the action time, and output it to the inverter to drive the permanent magnet synchronous motor.

[0064] In one embodiment, the sampling frequency determination module 11 is specifically used for: Obtain the current operating scenario of the permanent magnet synchronous motor, wherein the operating scenario includes at least low-speed light-load condition, rated condition, high-speed heavy-load condition and sudden change condition; The appropriate sampling frequency is determined based on the current operating scenario. The appropriate sampling frequency is equal to a preset baseline sampling frequency multiplied by a dynamic adjustment factor, where the preset baseline sampling frequency is 1.25 seconds / time. When the permanent magnet synchronous motor operates under low-speed, light-load conditions, the dynamic adjustment factor is set to two. When the permanent magnet synchronous motor operates under rated conditions, the dynamic adjustment factor is set to four. When the permanent magnet synchronous motor operates under high-speed, heavy-load conditions or sudden change conditions, the dynamic adjustment factor is set to eight. The optimized time window is set according to the adapted sampling frequency, and the length of the optimized time window is set to the sum of a preset number of sampling periods, wherein the preset number is an integer greater than or equal to 5, and the sampling period is the time span between adjacent sampling points.

[0065] Furthermore, in one embodiment of the application, predicting the state trajectory of the permanent magnet synchronous motor within the current optimized time window based on the historical monitoring data includes: Obtain historical monitoring data within the previous optimized time window, wherein the historical monitoring data includes direct-axis current values ​​and quadrature-axis current values ​​collected at each sampling time within the previous optimized time window; A state prediction model is constructed based on a long short-term memory network, wherein the state prediction model is used to output a state data sequence within a future time window based on the input historical monitoring data sequence; The historical monitoring data are arranged in chronological order to form an input sequence, and the input sequence is fed into the state prediction model for forward calculation to predict the state data sequence within the current optimization time window as the state trajectory.

[0066] Furthermore, a state prediction model is constructed based on Long Short-Term Memory (LSTM) networks, including: Acquire the monitoring data sequence of the permanent magnet synchronous motor during its historical operating period, and divide the monitoring data sequence into multiple training sample pairs, each training sample pair containing an input sequence and a target output sequence; Construct a long short-term memory network model, wherein the long short-term memory network model includes an input layer, at least one long short-term memory hidden layer, and an output layer; The input sequences from the training sample pairs are sequentially input into the Long Short-Term Memory (LSTM) network model. The hidden state and cell state are updated time-by-time through the calculation of the forget gate, input gate, output gate, and cell state in the LSM hidden layer. The updated hidden states are mapped to a predicted state sequence through the output layer, and the error between the predicted state sequence and the target output sequence is calculated. With the goal of minimizing the error, the network parameters of the Long Short-Term Memory Network model are updated using the backpropagation algorithm until a preset convergence condition is met. The trained Long Short-Term Memory Network model is then used as the state prediction model.

[0067] Furthermore, in one embodiment, with the goal of minimizing the current tracking error at the end of the optimized time window, the voltage vector sequence to be output and its duration within the current sampling period are calculated based on the state trajectory, including: Extract the predicted current value at the end of the optimized time window and the predicted current value at each sampling time within the optimized time window from the state trajectory. A value function is constructed based on the endpoint tracking error between the predicted current value and the given current value at the end of the optimized time window, and the instantaneous tracking error between the predicted current value and the given current value at each sampling time within the optimized time window. By enumerating candidate voltage vector sequences and solving for the duration of each vector, the optimal voltage vector sequence and its corresponding duration that minimize the value function are found. The optimal voltage vector sequence and its duration obtained from the solution are used as the voltage vector sequence and duration to be output within the current sampling period.

[0068] Furthermore, based on the endpoint tracking error between the predicted current value and the given current value at the end of the optimized time window, and the instantaneous tracking error between the predicted current value and the given current value at each sampling moment within the optimized time window, a value function is constructed, including: The sum of squares of the endpoint tracking errors is used as the first term of the value function, and the first term is multiplied by the first weighting coefficient to obtain the first weighted value; The sum of the squares of the instantaneous tracking errors at each sampling moment within the optimization time window is taken as the second term of the value function, and the second term is multiplied by the second weight coefficient to obtain the second weighted value, wherein the sum of the first weight coefficient and the second weight coefficient is 1; The first weighted value is added to the second weighted value to obtain the value of the value function.

[0069] Furthermore, by enumerating candidate voltage vector sequences and solving for the action time of each vector, the optimal voltage vector sequence and its corresponding action time that minimize the value function are found, including: All effective voltage vectors that the inverter can output are combined in pairs to generate multiple candidate vector combinations. Each candidate vector combination contains a first candidate effective vector and a second candidate effective vector. A zero vector is assigned to each candidate vector combination to form a complete candidate voltage vector sequence, wherein the candidate voltage vector sequence is arranged in the order of the first candidate effective vector, the second candidate effective vector, and the zero vector; For each candidate voltage vector sequence, the action time of the first candidate effective vector, the action time of the second candidate effective vector, and the total action time of the zero vector that minimize the value function are calculated, where the sum of the three action times is equal to the duration of the sampling period; For each candidate voltage vector sequence, the obtained action time is substituted into the discrete mathematical model of the permanent magnet synchronous motor to predict the current trajectory under the action of the current candidate voltage vector sequence and calculate the actual value function value. The candidate voltage vector sequence and its duration that minimize the actual value function are selected as the optimal voltage vector sequence and its corresponding duration.

[0070] Furthermore, the optimal voltage vector sequence and its duration obtained from the solution are used as the voltage vector sequence and duration to be output within the current sampling period, including: With the goal of minimizing the effective value of current ripple within the optimization time window, the allocation ratio of the action time of the first zero vector and the action time of the second zero vector is determined, and the total action time of the zero vector is allocated as the action time of the first zero vector and the action time of the second zero vector, wherein the first zero vector is the zero vector corresponding to the turn-off of all switches, and the second zero vector is the zero vector corresponding to the turn-on of all switches. The allocated first zero vector action time and second zero vector action time are combined with the action times of the first effective vector and second effective vector to form the complete voltage vector sequence and action time that needs to be output within the current sampling period.

[0071] In one embodiment, the synchronous motor drive module 13 is specifically used for: Each switching cycle is divided into a first half cycle and a second half cycle. The fixed waveform transmission sequence of the first half cycle is set to output the first zero vector, the first effective vector, the second effective vector, and the second zero vector in sequence. The fixed waveform transmission sequence of the second half cycle is set to output the second zero vector, the second effective vector, the first effective vector, and the first zero vector in sequence. Within the current sampling period, according to the fixed waveform generation sequence, a corresponding pulse width modulation waveform is generated based on the voltage vector sequence and the duration of action; and the generated pulse width modulation waveform is output to the inverter to control the switching transistors in the inverter to drive the permanent magnet synchronous motor.

Claims

1. A method for predictive control of a permanent magnet synchronous motor based on sampling frequency optimization, characterized in that, include: The appropriate sampling frequency is determined based on the current operating scenario of the permanent magnet synchronous motor, and an optimized time window is set based on the appropriate sampling frequency, including: The optimized time window is set according to the adaptive sampling frequency, and the length of the optimized time window is set to the sum of a preset number of sampling periods, wherein the preset number is an integer greater than or equal to 5, and the sampling period is the time span between adjacent sampling points. Within each sampling period, historical monitoring data from the previous optimized time window is acquired. Based on this historical monitoring data, the state trajectory of the permanent magnet synchronous motor within the current optimized time window is predicted. With the goal of minimizing the current tracking error at the end of the optimized time window, the voltage vector sequence and its duration to be output within the current sampling period are calculated based on the state trajectory, including: Extract the predicted current value at the end of the optimized time window and the predicted current value at each sampling time within the optimized time window from the state trajectory. A value function is constructed based on the endpoint tracking error between the predicted current value and the given current value at the end of the optimized time window, and the instantaneous tracking error between the predicted current value and the given current value at each sampling time within the optimized time window. By enumerating candidate voltage vector sequences and solving for the duration of each vector, the optimal voltage vector sequence and its corresponding duration that minimize the value function are found. The optimal voltage vector sequence and its duration obtained from the solution are used as the voltage vector sequence and duration to be output in the current sampling period. A pulse width modulation waveform is generated based on the voltage vector sequence and the duration of action, and output to the inverter to drive the permanent magnet synchronous motor.

2. The predictive control method for permanent magnet synchronous motors based on sampling frequency optimization according to claim 1, characterized in that, The appropriate sampling frequency is determined based on the current operating scenario of the permanent magnet synchronous motor, including: Obtain the current operating scenario of the permanent magnet synchronous motor, wherein the operating scenario includes at least low-speed light-load condition, rated condition, high-speed heavy-load condition and sudden change condition; The appropriate sampling frequency is determined based on the current operating scenario. The appropriate sampling frequency is equal to a preset reference sampling frequency multiplied by a dynamic adjustment factor, where the preset reference sampling frequency is 1.25 seconds / time. When the permanent magnet synchronous motor operates under low-speed and light-load conditions, the dynamic adjustment factor is set to two. When the permanent magnet synchronous motor operates under rated conditions, the dynamic adjustment factor is set to four. When the permanent magnet synchronous motor operates under high-speed, heavy-load conditions or sudden change conditions, the dynamic adjustment factor is set to eight.

3. The predictive control method for permanent magnet synchronous motors based on sampling frequency optimization according to claim 1, characterized in that, Based on the historical monitoring data, the state trajectory of the permanent magnet synchronous motor within the current optimized time window is predicted, including: Obtain historical monitoring data within the previous optimized time window, wherein the historical monitoring data includes direct-axis current values ​​and quadrature-axis current values ​​collected at each sampling time within the previous optimized time window; A state prediction model is constructed based on a long short-term memory network, wherein the state prediction model is used to output a state data sequence within a future time window based on the input historical monitoring data sequence; The historical monitoring data are arranged in chronological order to form an input sequence, and the input sequence is fed into the state prediction model for forward calculation to predict the state data sequence within the current optimization time window as the state trajectory.

4. The predictive control method for permanent magnet synchronous motors based on sampling frequency optimization according to claim 3, characterized in that, A state prediction model based on Long Short-Term Memory (LSTM) networks is constructed, including: Acquire the monitoring data sequence of the permanent magnet synchronous motor during its historical operating period, and divide the monitoring data sequence into multiple training sample pairs, each training sample pair containing an input sequence and a target output sequence; Construct a long short-term memory network model, wherein the long short-term memory network model includes an input layer, at least one long short-term memory hidden layer, and an output layer; The input sequences from the training sample pairs are sequentially input into the Long Short-Term Memory (LSTM) network model. The hidden state and cell state are updated time-by-time through the calculation of the forget gate, input gate, output gate, and cell state in the LSM hidden layer. The updated hidden states are mapped to a predicted state sequence through the output layer, and the error between the predicted state sequence and the target output sequence is calculated. With the goal of minimizing the error, the network parameters of the Long Short-Term Memory Network model are updated using the backpropagation algorithm until a preset convergence condition is met. The trained Long Short-Term Memory Network model is then used as the state prediction model.

5. The predictive control method for permanent magnet synchronous motors based on sampling frequency optimization according to claim 1, characterized in that, Based on the endpoint tracking error between the predicted current value and the given current value at the end of the optimized time window, and the instantaneous tracking error between the predicted current value and the given current value at each sampling time within the optimized time window, a value function is constructed, including: The sum of squares of the endpoint tracking errors is used as the first term of the value function, and the first term is multiplied by the first weighting coefficient to obtain the first weighted value; The sum of the squares of the instantaneous tracking errors at each sampling moment within the optimization time window is taken as the second term of the value function, and the second term is multiplied by the second weight coefficient to obtain the second weighted value, wherein the sum of the first weight coefficient and the second weight coefficient is 1; The first weighted value is added to the second weighted value to obtain the value of the value function.

6. The predictive control method for permanent magnet synchronous motors based on sampling frequency optimization according to claim 1, characterized in that, By enumerating candidate voltage vector sequences and solving for the duration of each vector, the optimal voltage vector sequence and its corresponding duration that minimize the value function are sought, including: All effective voltage vectors that the inverter can output are combined in pairs to generate multiple candidate vector combinations. Each candidate vector combination contains a first candidate effective vector and a second candidate effective vector. A zero vector is assigned to each candidate vector combination to form a complete candidate voltage vector sequence, wherein the candidate voltage vector sequence is arranged in the order of the first candidate effective vector, the second candidate effective vector, and the zero vector; For each candidate voltage vector sequence, the action time of the first candidate effective vector, the action time of the second candidate effective vector, and the total action time of the zero vector that minimize the value function are calculated, where the sum of the three action times is equal to the duration of the sampling period; For each candidate voltage vector sequence, the obtained action time is substituted into the discrete mathematical model of the permanent magnet synchronous motor to predict the current trajectory under the action of the current candidate voltage vector sequence and calculate the actual value function value. The candidate voltage vector sequence and its duration that minimize the actual value function are selected as the optimal voltage vector sequence and its corresponding duration.

7. The predictive control method for permanent magnet synchronous motors based on sampling frequency optimization according to claim 1, characterized in that, The optimal voltage vector sequence and its duration obtained from the solution are used as the voltage vector sequence and duration to be output within the current sampling period, including: With the goal of minimizing the effective value of current ripple within the optimization time window, the allocation ratio of the action time of the first zero vector and the action time of the second zero vector is determined, and the total action time of the zero vector is allocated as the action time of the first zero vector and the action time of the second zero vector, wherein the first zero vector is the zero vector corresponding to the turn-off of all switches, and the second zero vector is the zero vector corresponding to the turn-on of all switches. The allocated first zero vector action time and second zero vector action time are combined with the action times of the first effective vector and second effective vector to form the complete voltage vector sequence and action time that needs to be output within the current sampling period.

8. The predictive control method for permanent magnet synchronous motors based on sampling frequency optimization according to claim 1, characterized in that, A pulse width modulation waveform is generated based on the voltage vector sequence and the duration of action, and output to the inverter to drive the permanent magnet synchronous motor, including: Each switching cycle is divided into a first half cycle and a second half cycle. The fixed waveform transmission sequence of the first half cycle is set to output the first zero vector, the first effective vector, the second effective vector, and the second zero vector in sequence. The fixed waveform transmission sequence of the second half cycle is set to output the second zero vector, the second effective vector, the first effective vector, and the first zero vector in sequence. Within the current sampling period, according to the fixed waveform generation sequence, a corresponding pulse width modulation waveform is generated based on the voltage vector sequence and the duration of action; and the generated pulse width modulation waveform is output to the inverter to control the switching transistors in the inverter to drive the permanent magnet synchronous motor.

9. A predictive control system for permanent magnet synchronous motors based on sampling frequency optimization, characterized in that, The method for performing the predictive control of a permanent magnet synchronous motor based on sampling frequency optimization as described in any one of claims 1-8 includes: The sampling frequency determination module is used to determine the appropriate sampling frequency based on the current operating scenario of the permanent magnet synchronous motor, and to set an optimization time window based on the appropriate sampling frequency. The state trajectory prediction module is used to acquire historical monitoring data from the previous optimized time window in each sampling period, predict the state trajectory of the permanent magnet synchronous motor in the current optimized time window based on the historical monitoring data, and calculate the voltage vector sequence and action time to be output in the current sampling period based on the state trajectory with the goal of minimizing the current tracking error at the end of the optimized time window. The synchronous motor drive module is used to generate a pulse width modulation waveform based on the voltage vector sequence and the duration of action, and output it to the inverter to drive the permanent magnet synchronous motor.

Images