Dead-time compensation method, apparatus, device, and storage medium

Abstract

The application discloses a dead-time compensation method and device, equipment and a storage medium, and relates to the technical field of motor drive control. The method comprises the following steps: acquiring a current working condition characteristic vector and a traditional compensation voltage; passing the current working condition characteristic vector through a neural network to obtain residual harmonic coefficients; obtaining a residual compensation voltage according to the residual harmonic coefficients; and obtaining a dead-time compensation voltage according to the traditional compensation voltage and the residual compensation voltage, and performing dead-time compensation. The application reduces the complexity of the neural network, and realizes self-adaptive dead-time compensation in all working conditions while retaining the stability of traditional compensation.

Description

Technical Field

[0001] This application relates to the field of motor drive control technology, and in particular to a dead zone compensation method, device, equipment and storage medium. Background Technology

[0002] Permanent magnet synchronous motor drive systems commonly employ space vector pulse width modulation (SVM) to control inverter switching. However, the inverter is affected by non-ideal factors such as dead time and voltage drop during power device operation, resulting in a deviation between the actual output voltage and the ideal modulation voltage. This leads to phase current distortion and increased electromagnetic torque ripple. To suppress this dead-time effect, a compensation voltage is typically applied in the dq-axis coordinate system, known as dead-time compensation technology.

[0003] Among existing dead zone compensation schemes, traditional fixed parameter compensation methods have simple structures and good stability, but the compensation rules are fixed and it is difficult to maintain the optimal compensation effect under all operating conditions. Although the method based on neural networks to directly output the complete compensation voltage has a certain degree of adaptability, the model has high complexity, large output dimension, and high training difficulty, which is not conducive to the deployment of embedded controllers.

[0004] Therefore, how to achieve adaptive dead zone compensation under all operating conditions with low model complexity while preserving the stability of traditional compensation is a problem that urgently needs to be solved. Summary of the Invention

[0005] The main objective of this application is to provide a dead zone compensation method, apparatus, device, and storage medium, aiming to solve the technical problem of how to achieve adaptive dead zone compensation under all operating conditions with low model complexity while retaining the stability of traditional compensation.

[0006] To achieve the above objectives, this application provides a dead zone compensation method, comprising: Obtain the feature vector of the current operating condition and the traditional compensation voltage; The residual harmonic coefficients are obtained by passing the current operating condition feature vector through a neural network. The residual compensation voltage is obtained based on the residual harmonic coefficients. Based on the traditional compensation voltage and the residual compensation voltage, the dead zone compensation voltage is obtained, and dead zone compensation is performed.

[0007] In one embodiment, before obtaining the current operating condition feature vector and the traditional compensation voltage, the following steps are included: Obtain the operating characteristic quantities of the current working condition; Based on the operating characteristics of the current working condition, the traditional compensation voltage is obtained; The traditional compensation voltage is integrated with the operating characteristics of the current operating condition to obtain the current operating condition feature vector.

[0008] In one embodiment, the neural network is obtained through offline training, the offline training process including: Under multiple preset operating conditions, the goal is to minimize the peak-to-peak value of electromagnetic torque, and the optimal residual harmonic coefficient is searched as the training label. The neural network is trained using the actual operating characteristics and the actual traditional compensation voltage under each working condition as input samples and the corresponding optimal residual harmonic coefficient as the output label.

[0009] In one embodiment, obtaining the residual harmonic coefficients by passing the current operating condition feature vector through a neural network includes: The current operating condition feature vector is passed through the hidden layer of the neural network to obtain the hidden feature vector; The hidden feature vector is mapped through the output layer of the neural network to obtain the residual harmonic coefficients.

[0010] In one embodiment, obtaining the residual compensation voltage based on the residual harmonic coefficients includes: Multiplying the residual harmonic coefficients by a sinusoidal basis function that is six times the rotor electrical angle yields the d-axis residual compensation voltage. The residual harmonic coefficient is multiplied by the cosine basis function of six times the rotor electrical angle to obtain the q-axis residual compensation voltage.

[0011] In one embodiment, obtaining the dead-zone compensation voltage based on the conventional compensation voltage and the residual compensation voltage includes: The residual compensation voltage is limited to obtain the limited residual compensation voltage. The dead zone compensation voltage is obtained by superimposing the traditional compensation voltage with the residual compensation voltage after limiting.

[0012] In one embodiment, the method further includes: Get the mode switching variable; Based on the value of the mode switching variable, the system switches between no-compensation mode, traditional compensation mode, and hybrid compensation mode; wherein, in the hybrid compensation mode, the dead zone compensation voltage is obtained by superimposing the traditional compensation voltage and the residual compensation voltage.

[0013] In addition, to achieve the above objectives, this application also provides a dead-zone compensation device, comprising: The acquisition module is used to acquire the current operating condition feature vector and the traditional compensation voltage; The neural network module is used to obtain the residual harmonic coefficients by passing the current operating condition feature vector through a neural network. The residual compensation module is used to obtain the residual compensation voltage based on the residual harmonic coefficients. The result module is used to obtain the dead zone compensation voltage based on the traditional compensation voltage and the residual compensation voltage, and to perform dead zone compensation.

[0014] In addition, to achieve the above objectives, this application also proposes a dead-zone compensation device, the device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the dead-zone compensation method as described above.

[0015] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the dead-zone compensation method described above.

[0016] This application provides a dead-zone compensation method. By employing a technique that uses both the traditional compensation voltage and the current operating condition feature vector as input, and then using a neural network to output only the residual harmonic coefficients, synthesizing the residual compensation voltage from these coefficients and superimposing it with the traditional compensation voltage, the neural network does not need to directly output the complete compensation voltage. It only needs to learn the residual part outside of the traditional compensation, significantly reducing the network's output dimensionality and learning difficulty. This solves the technical problem of how to achieve full-condition adaptive dead-zone compensation with low model complexity while preserving the stability of traditional compensation. Attached Figure Description

[0017] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the first embodiment of dead zone compensation in this application; Figure 2 This is a flowchart illustrating the second embodiment of the dead zone compensation in this application; Figure 3 This is a flowchart illustrating the third embodiment of dead zone compensation in this application; Figure 4 This is a structural block diagram of the first embodiment of the dead zone compensation device of this application; Figure 5 This is a schematic diagram of the dead zone compensation device for the hardware operating environment involved in the embodiments of this application.

[0020] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0021] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0022] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0023] The main solution of this application is as follows: First, obtain the current operating condition feature vector and the traditional compensation voltage; pass the current operating condition feature vector through a neural network to obtain the residual harmonic coefficient; obtain the residual compensation voltage based on the residual harmonic coefficient; obtain the dead zone compensation voltage based on the traditional compensation voltage and the residual compensation voltage, and perform dead zone compensation.

[0024] In this embodiment, for ease of description, the following description will focus on identifying dead zone compensation as the main execution subject.

[0025] While traditional fixed-parameter compensation methods are simple in structure and stable, their compensation rules are rigid, making it difficult to maintain optimal compensation performance across all operating conditions. Methods based on neural networks that directly output complete compensation voltages, while possessing some adaptive capability, suffer from high model complexity and large output dimensions, hindering deployment in embedded controllers. Therefore, achieving adaptive dead-zone compensation across all operating conditions with low model complexity while preserving the stability of traditional compensation methods is a pressing issue that needs to be addressed.

[0026] This application uses traditional compensation as a foundation and neural network prediction of residual coefficients as an enhanced branch. It only needs to output low-dimensional residual harmonic coefficients instead of complete compensation voltage, which reduces the complexity of neural network and achieves full-condition adaptive dead-zone compensation while retaining the stability of traditional compensation.

[0027] It should be noted that the executing entity in this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, or mobile phone, or an electronic device or dead-zone compensation program capable of performing the above functions. The following description uses a computing device as an example to illustrate this embodiment and the subsequent embodiments.

[0028] Based on this, this application proposes a dead-zone compensation method according to a first embodiment. Please refer to [link / reference]. Figure 1 The dead zone compensation method includes steps S10 to S40: Step S10: Obtain the current operating condition feature vector and the traditional compensation voltage.

[0029] Understandably, this solution corrects residuals based on a traditional compensation base, rather than predicting the complete compensation voltage from scratch. Therefore, two types of information need to be acquired simultaneously: first, a condition feature vector reflecting the motor's current operating state, used to determine the extent of the dead-zone effect under the current conditions; and second, the traditional compensation voltage, used to clarify how much compensation has been provided by the fixed-parameter compensation. Only by acquiring both types of information can the subsequent neural network accurately determine how much uncovered residual remains in the traditional compensation under the current conditions.

[0030] It should be noted that the current operating condition feature vector refers to a multi-dimensional array composed of multiple real-time operating parameters. For example, six parameters—d-axis current, q-axis current, electrical angular velocity, electromagnetic torque, load torque, and dead-zone error equivalent amplitude—are combined in a preset order to form a 6-dimensional vector. Among them, the d-axis and q-axis currents reflect the current state of the motor, the electrical angular velocity reflects the speed condition, the electromagnetic torque and load torque reflect the load condition, and the dead-zone error equivalent amplitude reflects the nonlinearity of the inverter. The traditional compensation voltage refers to the d-axis and q-axis compensation voltages calculated using a fixed formula based on the dead-zone error equivalent amplitude, the traditional compensation proportional coefficient, and a sine / cosine function of six times the rotor electrical angle. It represents the fixed compensation amount of the traditional method.

[0031] Understandably, by simultaneously acquiring the operating condition feature vector and the traditional compensation voltage, a complete decision-making basis is provided for the subsequent neural network. This allows the network to know the current operating condition and the degree of base compensation during inference, thus only needing to focus on the residual part not covered by traditional compensation, reducing the learning burden and output dimensionality of the network. At the same time, this step ensures that the base function of traditional compensation is fully utilized, and architecturally ensures that the hybrid compensation effect is not inferior to the effect of traditional compensation acting alone.

[0032] refer to Figure 2 In one feasible implementation, steps A11 to A13 are included before step S10: Step A11: Obtain the operating characteristic quantities of the current operating condition.

[0033] Understandably, the strength of the dead-zone effect is not constant, but changes significantly with variations in operating parameters such as motor speed, load torque, current magnitude, and the inverter's own dead-zone equivalent amplitude. To achieve adaptive dead-zone compensation under all operating conditions, it is first necessary to accurately perceive the motor's current operating state. Therefore, it is necessary to collect key parameters reflecting the motor's real-time operating state at the very beginning of the compensation process, providing a data foundation for subsequently judging the degree of dead-zone impact under the current operating conditions and calculating the compensation amount.

[0034] It should be noted that the operating characteristic quantities under the current working condition refer to multiple physical quantities that can characterize the motor's operating state and are collected in real time from the motor drive system. Taking this scheme as an example, these include: d-axis current and q-axis current (reflecting the excitation and torque components of the motor in the synchronous rotating coordinate system), electrical angular velocity (reflecting the speed of motor rotation, in rad / s, for example, the electrical angular velocity corresponding to the rated speed is approximately 628 rad / s), electromagnetic torque (reflecting the instantaneous torque output by the motor, in N·m), load torque (reflecting the externally applied resistance torque, in N·m), and equivalent amplitude of dead-zone error (reflecting the equivalent voltage amplitude of the combined effect of the inverter's dead time and the voltage drop of the power devices, in V, for example, a typical value of 10V or 20V). These characteristic quantities are collected and calculated in real time by hardware such as current sensors and encoders.

[0035] Understandably, by acquiring the aforementioned operational characteristics, the entire compensation process gains condition awareness, enabling subsequent traditional compensation calculations and neural network residual predictions to make decisions based on real-world operating conditions. This avoids the undercompensation or overcompensation caused by the inability of traditional fixed-parameter schemes to perceive changes in operating conditions. It also provides sufficient input information for lightweight neural networks, allowing them to output adaptive residual harmonic coefficients for different operating conditions, thereby achieving full-condition compensation optimization while ensuring real-time performance.

[0036] Step A12: Obtain the conventional compensation voltage based on the operating characteristics of the current operating condition.

[0037] Understandably, traditional fixed-parameter dead-zone compensation methods, based on physical models of the dead-zone effect, offer advantages such as simple structure, computational stability, and high engineering reliability. Completely abandoning traditional compensation and directly using neural networks to predict the complete compensation voltage from zero would not only place an excessive learning burden on the network and increase model complexity, but also waste mature and long-validated prior knowledge of the physical model. Therefore, this step in the hybrid compensation architecture first uses traditional methods to generate a basic compensation quantity, providing a stable and reliable compensation foundation for the entire system. This allows subsequent neural network residual prediction branches to focus only on the residual portions not covered by traditional compensation, without having to repeatedly learn the compensation components already addressed by traditional methods.

[0038] It should be noted that traditional compensation voltage refers to the d-axis and q-axis compensation voltages calculated using a fixed-parameter formula based on the dead-zone effect physical model. The equivalent amplitude of the dead-zone error refers to unifying non-ideal factors such as inverter dead time, power device conduction voltage drop, and switching delay into an equivalent voltage amplitude. For example, under certain inverter parameters, this equivalent amplitude can be 10V or 20V. The traditional compensation ratio coefficient is a preset positive number less than 1, such as 0.85, used for conservative compensation to avoid overcompensation. The specific calculation method for the traditional compensation voltage is as follows: multiply the equivalent amplitude of the dead-zone error by the traditional compensation ratio coefficient, then multiply by the cosine and sine functions of six times the rotor electrical angle respectively, and invert the results to obtain the traditional compensation amounts for the d-axis and q-axis.

[0039] The formula for calculating the traditional compensation voltage is as follows:

[0040] in, This is the traditional compensation amount for the d-axis; This is the traditional compensation amount for the q-axis; This refers to the traditional compensation ratio coefficient; This is the equivalent amplitude of the dead zone error; The rotor electrical angle is given.

[0041] Understandably, the conventional compensation voltage generated in this step provides a deterministic baseline output for the entire compensation system. Its calculation process does not rely on online training or iterative optimization, ensuring real-time performance and stability. Simultaneously, this conventional compensation voltage will be used as part of the current operating condition feature vector in subsequent steps, inputting into a lightweight neural network. This allows the network to know exactly how much compensation has been done during inference, thus only needing to learn the residual mapping relationships beyond conventional compensation. This design reduces the network's learning objective from outputting the complete compensation voltage waveform to outputting four residual harmonic coefficients, significantly reducing the network's output dimensionality and training difficulty. At the same time, the architecture ensures that the final hybrid compensation effect is not inferior to the effect of conventional compensation acting alone.

[0042] Step A13: Integrate the traditional compensation voltage with the operating characteristic quantity of the current operating condition to obtain the current operating condition feature vector.

[0043] Understandably, step A12 has generated the traditional compensation voltage, and step A11 has acquired various feature quantities reflecting the motor's operating state. However, at this point, the two types of information are separate: the traditional compensation voltage exists independently, and the operating feature quantities lack crucial information about the extent of traditional compensation. If the operating feature quantities are directly input into the neural network, the network will be unable to distinguish which errors have been covered by traditional compensation and which are residual errors, requiring it to learn the complete compensation mapping from scratch, resulting in excessive network complexity. Therefore, this step integrates the two into a unified feature vector, enabling the neural network to simultaneously obtain information about both the current operating condition and the traditional compensation voltage, thereby focusing the learning objective on the residual portion beyond traditional compensation.

[0044] It should be noted that the current operating condition feature vector refers to a multi-dimensional array formed by concatenating the operating feature quantities obtained in step A11 and the traditional compensation voltage generated in step A12 in a preset order. For example, the operating feature quantities include six parameters: d-axis current, q-axis current, electrical angular velocity, electromagnetic torque, load torque, and equivalent amplitude of dead zone error. The traditional compensation voltage includes two parameters: d-axis traditional compensation quantity and q-axis traditional compensation quantity. After integration, an 8-dimensional current operating condition feature vector is formed. The values ​​for each dimension are scalars obtained through real-time acquisition or calculation. For example, at a certain moment, the d-axis current is 0.1A, the q-axis current is 5.2A, the electric angular velocity is 314rad / s, the electromagnetic torque is 10N·m, the load torque is 10N·m, the equivalent amplitude of the dead zone error is 20V, the traditional compensation amount for the d-axis is -17V, and the traditional compensation amount for the q-axis is 0V. Then the current operating condition feature vector is [0.1, 5.2, 314, 10, 10, 20, -17, 0].

[0045] Understandably, by integrating the operational status information and the compensated information into a single feature vector, a structured and complete input is provided for the subsequent lightweight neural network. This eliminates the need for the network to infer the current compensated amount, avoiding the problem of the network repeatedly learning from previously solved compensations, and significantly reducing output dimensionality and model complexity. Furthermore, the unified feature vector format facilitates data organization and computational pipelined execution during online inference, enabling efficient deployment in embedded controllers.

[0046] Step S20: The current operating condition feature vector is processed through a neural network to obtain the residual harmonic coefficients.

[0047] Understandably, while traditional fixed-parameter compensation provides a stable compensation base, its compensation amount is fixed under given dead-zone equivalent amplitude and proportional coefficient, and cannot adaptively adjust to fluctuations in operating conditions such as speed and load. Therefore, when deviating from the design operating conditions, there will inevitably be residuals of insufficient or excessive compensation. If a neural network were to directly output the complete compensation voltage to correct this residual, the network would need to learn a mapping relationship covering everything from zero to the complete compensation amount, resulting in high output dimensionality, high training difficulty, and difficulty in deployment on embedded controllers. Therefore, this step uses a neural network to predict only four low-dimensional residual harmonic coefficients to describe the residual portion not covered by traditional compensation, rather than outputting the complete compensation voltage waveform. Since the input feature vector already contains the traditional compensation voltage, the network explicitly knows the amount of compensation already provided by the base, and its learning objective is limited to determining how much traditional compensation is still lacking under the current operating conditions.

[0048] It should be noted that a neural network refers to a multi-layered perceptual structure with an input layer, hidden layers, and an output layer. For example, the input layer receives an 8-dimensional feature vector of the current operating condition, the hidden layer contains 16 neurons and performs nonlinear feature extraction through an activation function, and the output layer generates four numerical values. The residual harmonic coefficients refer to these four numerical values ​​output by the neural network, named the d-axis sine coefficient, d-axis cosine coefficient, q-axis sine coefficient, and q-axis cosine coefficient, respectively. These four coefficients are not directly used as the compensation voltage output, but rather as weights for the sine and cosine basis functions when subsequently synthesizing the residual compensation voltage. For example, if the network output is [0.5, -0.3, 0.2, 0.1] under a certain operating condition, then the coefficient of the d-axis sine term is 0.5, the coefficient of the d-axis cosine term is -0.3, the coefficient of the q-axis sine term is 0.2, and the coefficient of the q-axis cosine term is 0.1. This means that under the current operating condition, the traditional compensation needs to supplement the d-axis sine component with 0.5 times the amplitude of the basis function, and the d-axis cosine component needs to be corrected by 0.3 times the amplitude of the basis function.

[0049] Understandably, by outputting only four residual harmonic coefficients instead of the complete compensation voltage through a neural network, the output dimension is reduced from a time-varying waveform related to electrical angle to four scalar coefficients related to operating conditions. This significantly reduces the dimensionality of the network's learning objective and the number of model parameters. This allows the neural network to be implemented with a minimal three-layer structure, requiring only one forward propagation during online inference, resulting in extremely low computational cost and easy deployment in embedded motor controllers. Furthermore, since the input feature vector already contains the traditional compensation voltage, the network only needs to learn the residual mapping instead of the complete mapping, leading to a clear learning objective, fast convergence speed, and high prediction accuracy.

[0050] In one feasible implementation, the neural network in step S20 is obtained through offline training. The offline training process includes: under multiple preset operating conditions, with the goal of minimizing the peak-to-peak value of electromagnetic torque, and searching for the optimal residual harmonic coefficient as the training label; using the actual operating characteristic quantities and actual conventional compensation voltage of each operating condition as input samples, and the corresponding optimal residual harmonic coefficient as the output label, to train the neural network.

[0051] Understandably, the neural network in step S20 needs to be able to output residual harmonic coefficients based on the operating condition feature vector. However, this network does not possess this mapping capability before deployment, and a correspondence between the operating condition features and the optimal residual coefficients needs to be established in advance. If online training is used, the computing power and storage resources of the embedded controller cannot support the iterative optimization process, and online exploration may cause unstable compensation outputs or even system oscillations. Therefore, this step adopts a strategy of separating offline training and online inference: before actual operation, multiple representative operating conditions are traversed on a simulation environment or experimental platform, and the optimal residual coefficients that minimize electromagnetic torque ripple under each operating condition are found through optimization search. Then, these residual coefficients are used as labels to train the network, enabling the network to learn the mapping law between the operating condition features and the optimal residual coefficients.

[0052] It should be noted that the preset operating conditions refer to combinations of operating conditions artificially set during the offline training phase, covering common operating ranges of the motor. For example, the equivalent amplitude of dead zone error is set to 10V and 20V respectively, and the load torque is set to 10N·m and 15N·m respectively. These are combined in pairs to form four representative operating conditions, covering typical scenarios such as light load and heavy load, weak dead zone effect and strong dead zone effect. The peak-to-peak value of electromagnetic torque refers to the difference between the maximum and minimum values ​​of electromagnetic torque within one electrical cycle. For example, under a certain operating condition, the peak-to-peak value of electromagnetic torque of the traditional compensation scheme is 0.359N·m. The smaller this value, the smaller the torque ripple and the better the dead zone compensation effect. The optimal residual harmonic coefficients refer to a set of residual harmonic coefficients that minimize the peak-to-peak value of electromagnetic torque under a given operating condition, found by using algorithms such as traversal search or particle swarm optimization, with the objective function of minimizing the peak-to-peak value of electromagnetic torque. These residual harmonic coefficients serve as the ideal output labels under that operating condition.

[0053] Understandably, by decoupling the computationally intensive learning components of the network from the embedded controller through offline training, online deployment only requires loading the pre-trained model parameters for lightweight forward inference, significantly reducing online computing power requirements. Using the minimization of peak-to-peak electromagnetic torque as the optimization objective, the network directly optimizes the final compensation effect, avoiding indirect optimization biases from intermediate variables and ensuring direct alignment between training labels and actual control objectives. Simultaneously, by searching for optimal labels and constructing training sets under multiple representative operating conditions, the network learns a complete mapping pattern covering different dead zone intensities and load levels during training, guaranteeing generalization ability across various operating conditions during online inference. After training, the network parameters can be exported as an array file directly accessible to the embedded controller, facilitating deployment without the need for online training data collection.

[0054] Step S30: Obtain the residual compensation voltage based on the residual harmonic coefficients.

[0055] Understandably, the residual harmonic coefficients output in step S20 are only four abstract values, representing the correction weights for the sine and cosine components of the d-axis and q-axis, respectively. These coefficients are not voltage signals that can be directly applied to the motor control loop; they need to be converted into physically meaningful dq-axis compensation voltages to actually correct the errors caused by the inverter dead-zone effect. Furthermore, the dead-zone effect in the dq-axis coordinate system mainly manifests as a pulsating component at six times the fundamental frequency, a physical law already adopted in the traditional compensation in step A12. Therefore, this step, also based on the sine and cosine basis functions at six times the electrical angle, calculates and converts the four coefficients into corresponding residual compensation voltages, ensuring that the residual correction remains consistent with the physical characteristics of the dead-zone effect.

[0056] It should be noted that the residual compensation voltage refers to the d-axis and q-axis compensation voltage components, which are driven by the four residual harmonic coefficients output by the neural network and synthesized through a sine and cosine basis function of six times the rotor electrical angle. The synthesis method is as follows: the d-axis residual compensation voltage equals the d-axis sine coefficient multiplied by the sine of six times the electrical angle, plus the d-axis cosine coefficient multiplied by the cosine of six times the electrical angle; the q-axis residual compensation voltage is obtained similarly by superimposing the q-axis sine and cosine coefficients multiplied by their respective basis functions.

[0057] Understandably, this step, through basis function synthesis, transforms four low-dimensional coefficients into a residual compensation voltage that changes in real time with the electrical angle, achieving a transition from a static mapping from operating conditions to coefficients to a dynamic generation of voltage waveforms from coefficients. Since the basis functions themselves have already embedded the physical prior of the dead-zone effect and sixth-harmonic pulsations, the neural network does not need to learn the shape of the voltage waveform; it only needs to learn the intensity of each frequency component under each operating condition, further reducing the learning burden on the network. Simultaneously, the operation of multiplying and superimposing the coefficients with the basis functions is extremely simple, involving only sine and cosine lookup tables and multiplication-addition operations, resulting in very low online computational overhead, making it suitable for real-time execution in embedded controllers. The amplitude of the residual compensation voltage is entirely determined by the four coefficients, and the range of coefficient values ​​has been effectively constrained during offline training, ensuring that the residual correction amount is naturally kept within a reasonable range, avoiding abnormal output.

[0058] In one feasible implementation, step S30 includes multiplying the residual harmonic coefficient by a sine basis function of six times the rotor electrical angle to obtain the d-axis residual compensation voltage; and multiplying the residual harmonic coefficient by a cosine basis function of six times the rotor electrical angle to obtain the q-axis residual compensation voltage.

[0059] It is understandable that the residual harmonic coefficients output in step S20 are four scalar values ​​related to the operating conditions, which do not possess time-varying characteristics and cannot be directly injected into the motor control loop as compensation voltage. However, the error voltage caused by the dead-zone effect in the dq-axis coordinate system exhibits a clear pulsation characteristic of six times the fundamental frequency, a physical law already applied in the traditional compensation in step A12. Therefore, this step utilizes the sine and cosine basis functions of six times the rotor electrical angle as the waveform carrier of the residual compensation voltage, converting the four coefficients output in step S20 into a time-varying voltage signal that changes in real time with the electrical angle. The fundamental reason for using sine and cosine basis functions instead of other waveform functions is that the essence of the dead-zone error voltage is the sixth harmonic component generated by the three-phase inverter switching dead zone after coordinate transformation. The sine and cosine basis functions perfectly match the frequency and phase of this harmonic component, enabling accurate fitting of the residual with minimal parameters.

[0060] It should be noted that six times the rotor electrical angle refers to the rotor electrical angle acquired in real time by the encoder. Multiply by 6 to get the angle value Its period is one-sixth of the fundamental electrical period. The sinusoidal basis function is a sinusoidal function sin(θ) with an independent variable of six times the electrical angle. The cosine basis function refers to the cosine function cos(θ) with six times the electrical angle as the independent variable. Together, they constitute the orthogonal basis for the synthesis of residual compensation voltages. The d-axis residual compensation voltage refers to the residual correction voltage component applied along the d-axis, which is composed of the d-axis sine coefficient and sin(...). The product of ) and the coefficient of the d-axis cosine term with cos( The product of ) is obtained by superimposing the q-axis residual compensation voltage. Similarly, it is obtained by combining the q-axis sine coefficient with sin( The product of ) and the coefficient of the q-axis cosine term and cos( The product of ) is obtained by superimposing.

[0061] The calculation formulas for the d-axis residual compensation voltage and the q-axis residual compensation voltage are as follows:

[0062] in, This is the d-axis residual compensation voltage; This is the q-axis residual compensation voltage; The coefficients of the d-axis sine term; The coefficients of the d-axis cosine term; The coefficients of the q-axis cosine term; The coefficients of the q-axis sine term; It is six times the rotor electrical angle.

[0063] Understandably, converting four static coefficients related to the operating conditions into compensation voltages that dynamically change with the electrical angle allows the abstract values ​​output by the neural network to be translated into physical voltage signals that can be directly injected into the control loop. Using a sine and cosine basis function of six times the electrical angle as the waveform carrier for the residual compensation voltage fully utilizes the physical laws of the sixth harmonic of the dead-zone effect. The neural network only needs to learn the intensity coefficients of each frequency component, without needing to learn the shape or frequency characteristics of the voltage waveform, further reducing the learning burden on the network. The calculation of the sine and cosine functions can be quickly obtained through offline table creation or real-time table lookup. Combined with two multiplications and one addition, the single-axis residual voltage calculation can be completed, resulting in minimal online computation, perfectly meeting the real-time requirements of embedded controllers. Furthermore, since the range of the sine and cosine functions is always within the [-1, 1] interval, the instantaneous amplitude of the residual compensation voltage is limited by the values ​​of the four coefficients, facilitating subsequent amplitude limiting processing.

[0064] Step S40: Based on the conventional compensation voltage and the residual compensation voltage, obtain the dead zone compensation voltage and perform dead zone compensation.

[0065] It should be noted that dead-zone compensation voltage refers to the d-axis and q-axis compensation voltages ultimately injected into the motor vector control loop. It is obtained by adding the traditional compensation voltage and the residual compensation voltage on the corresponding axes, respectively. That is, the d-axis dead-zone compensation voltage equals the sum of the traditional d-axis compensation and the residual d-axis compensation voltage, and the q-axis dead-zone compensation voltage equals the sum of the traditional q-axis compensation and the residual q-axis compensation voltage. Dead-zone compensation involves superimposing these dead-zone compensation voltages onto the output of the current loop PI controller. This is then combined with the dq-axis voltage command generated by the current loop and driven by the inverse Park converter and SVPWM module, thus pre-compensating the error voltage caused by the dead-zone effect in the inverter output voltage. For example, if at a certain moment the traditional d-axis compensation is -15V and the residual d-axis compensation voltage is 0.3V, the superimposed d-axis dead-zone compensation voltage will be -14.7V, which will be injected into the current loop output.

[0066] Understandably, by superimposing the outputs, the traditional compensation base and the residual correction branch are coordinated, enabling the final dead-zone compensation voltage to possess both the stability of the traditional method and the adaptive capability of the data-driven method. Since the residual compensation voltage is an incremental correction based on the traditional compensation, its amplitude is typically much smaller than the traditional compensation voltage. Even if the residual prediction deviates under abnormal operating conditions, the superimposed output still prioritizes the traditional compensation, preventing significant deviations in the compensation voltage or system instability. The dead-zone compensation voltage is directly superimposed onto the current loop output, requiring no modification to any part of the existing vector control framework, achieving integrated deployment of hybrid compensation with minimal modification cost. Field tests have verified that under multiple representative operating conditions, the peak-to-peak value of the electromagnetic torque after hybrid compensation is reduced compared to the traditional compensation scheme, with the highest improvement being significant, demonstrating the effectiveness of the superimposed output.

[0067] refer to Figure 3 In one feasible implementation, step S40 may include steps B11-B12: Step B11: Limit the residual compensation voltage to obtain the limited residual compensation voltage.

[0068] Understandably, the residual compensation voltage is synthesized from the residual harmonic coefficients output by the neural network using sine and cosine basis functions. However, as a data-driven model, the neural network may output abnormal coefficient values ​​deviating from a reasonable range under extreme operating conditions not covered by training data or under sensor noise interference, thus synthesizing a residual compensation voltage with excessive amplitude. If the unconstrained residual compensation voltage is directly superimposed on the traditional compensation voltage and injected into the control loop, the excessive correction amount may cause instantaneous changes in the dq axis voltage command, leading to motor current oscillations or even system instability. Therefore, after the residual compensation voltage is synthesized but before it is superimposed on the traditional compensation voltage, a limiting circuit is set to restrict the amplitude of the residual compensation voltage within a preset safe range.

[0069] It should be noted that amplitude limiting refers to clamping the instantaneous value or amplitude of the residual compensation voltage. When the residual compensation voltage exceeds the preset upper limit, it is forcibly set to the upper limit; when it falls below the preset lower limit, it is forcibly set to the lower limit. Between the upper and lower limits, the original value is maintained. The amplitude-limited residual compensation voltage is the residual compensation voltage component after the above clamping process, and its amplitude is constrained within the preset allowable range. The preset upper and lower limits can be determined comprehensively based on the motor's rated voltage, the amplitude of the traditional compensation voltage, and the system's tolerance. For example, setting the upper and lower limits to 20% of the traditional compensation voltage amplitude, when the traditional compensation voltage amplitude is approximately 17V, the residual compensation voltage is limited to between -3.4V and +3.4V, ensuring that the residual correction amount is always much smaller than the traditional compensation base magnitude.

[0070] Understandably, by using amplitude limiting, the system retains the ability of the residual compensation voltage to finely correct traditional compensation while cutting off the risk path of abnormal coefficient values ​​propagating to the control loop, thus providing a safety barrier for the hybrid compensation system. Since the amplitude limiting only affects the residual compensation voltage and does not affect the normal output of the traditional compensation voltage, even under extreme conditions where amplitude limiting is triggered, the system can still maintain basic dead-zone suppression by relying on the traditional compensation base, preventing complete compensation failure. Simultaneously, the amplitude limiting threshold setting also conversely constrains the effective range of the neural network output coefficients, allowing the network to learn within a safe region during offline training by appropriately selecting the amplitude limiting value, further improving the model's engineering reliability.

[0071] Step B12: Superimpose the conventional compensation voltage with the residual compensation voltage after limiting to obtain the dead zone compensation voltage.

[0072] Understandably, the preceding steps have generated two independent compensation voltage components: the traditional compensation voltage provides a fixed-base compensation based on the dead-zone effect physical model, while the limited residual compensation voltage provides an adaptive correction amount subject to safety constraints. These two components address different aspects of dead-zone compensation: traditional compensation addresses the fixed, general error portion, while residual correction addresses the deviation portion that varies with operating conditions; they are functionally complementary. Without superposition, traditional compensation alone cannot adapt to changes in operating conditions, and residual correction alone loses the stable base of the physical model. Therefore, this step adds the two components on their corresponding axes to form a complete dead-zone compensation voltage, ensuring that the final output possesses both stability and adaptability.

[0073] It should be noted that "superposition" refers to adding the traditional d-axis compensation value to the d-axis residual compensation voltage to obtain the d-axis dead-zone compensation voltage, and adding the traditional q-axis compensation value to the q-axis residual compensation voltage to obtain the q-axis dead-zone compensation voltage. The dead-zone compensation voltage refers to the sum of the d-axis and q-axis compensation voltages ultimately injected into the motor vector control loop to offset the inverter dead-zone effect error voltage. This dead-zone compensation voltage will be directly superimposed on the d-axis and q-axis voltage command output terminals of the current loop PI controller, participating in subsequent coordinate transformation and SVPWM modulation.

[0074] The formula for calculating the dead zone compensation voltage is as follows:

[0075] in, This is the d-axis dead zone compensation voltage; This is the q-axis dead zone compensation voltage; This is the d-axis residual compensation voltage; This is the q-axis residual compensation voltage; This is the traditional compensation amount for the d-axis; This is the traditional compensation amount for the q-axis.

[0076] Understandably, by integrating the prior physical model of traditional compensation with the adaptive capability of neural networks into a unified compensation output, adaptive suppression of dead-zone error across all operating conditions is achieved while preserving the stability and interpretability of traditional compensation. Since the amplitude of the residual compensation voltage after limiting is constrained within a safe range and is typically much smaller than the amplitude of the traditional compensation voltage, the superimposed dead-zone compensation voltage always has the traditional compensation as the dominant component. The residual correction is only slightly adjusted based on the traditional compensation, ensuring that the hybrid compensation effect is not inferior to the effect of traditional compensation acting alone. Experimental verification shows that under operating conditions with an equivalent dead-zone error amplitude of 20V and a load torque of 10N·m, the hybrid compensation scheme with superimposed output significantly reduces the peak-to-peak value of electromagnetic torque compared to the traditional compensation scheme, effectively suppressing torque ripple caused by the dead-zone effect.

[0077] In one feasible implementation, the dead zone compensation method further includes: obtaining a mode switching variable; switching between a no-compensation mode, a traditional compensation mode, and a hybrid compensation mode according to the value of the mode switching variable; wherein, in the hybrid compensation mode, the dead zone compensation voltage is obtained by superimposing the traditional compensation voltage and the residual compensation voltage.

[0078] Understandably, the hybrid compensation method proposed in this solution involves three different compensation strategies: the no-compensation mode corresponds to the original vector control, the traditional compensation mode corresponds to fixed-parameter dead-zone compensation, and the hybrid compensation mode corresponds to the superposition output of traditional compensation and neural network residual correction. In practical engineering applications, different operating scenarios may require different compensation strategies. For example, during the system startup phase or the transition phase before the neural network has completed initialization, it is necessary to revert to the traditional compensation mode to ensure stable system operation. During the performance comparison test phase, it is necessary to fairly compare the compensation effects of the three modes within the same system framework. If each compensation mode requires a separate control program, it not only increases software maintenance costs but also fails to guarantee the consistency of operating conditions during the comparison test. Therefore, this step sets a unified mode switching variable to integrate the three compensation modes into the same control framework, and seamless switching between modes can be achieved by changing the variable value.

[0079] It should be noted that the mode switching variable is a preset integer variable used to indicate the compensation mode currently used by the system. Different values ​​correspond to different compensation strategies. For example, when the mode switching variable is 0, the system enters the no-compensation mode, outputting no compensation voltage, and the inverter operates only according to the original voltage command output by vector control. When the mode switching variable is 1, the system enters the traditional compensation mode, only superimposing the generated traditional compensation voltage onto the current loop output, and the residual prediction branch does not work. When the mode switching variable is 2, the system enters the hybrid compensation mode, superimposing the traditional compensation voltage and the limited residual compensation voltage onto the current loop output. The value of the mode switching variable can be set by the host computer command, the controller's internal logic, or preset operating condition judgment conditions.

[0080] Understandably, by using a unified mode switching variable, this solution integrates the three compensation strategies into a single control framework and software architecture, avoiding the development of multiple independent programs for different compensation modes and significantly reducing software maintenance complexity and engineering deployment costs. In performance verification and comparative testing, without changing hardware connections and main control parameters, the comparison of the three schemes—no compensation, traditional compensation, and hybrid compensation—can be quickly completed through single-variable switching, ensuring consistency in test conditions and making the attribution analysis of performance improvements more accurate and reliable. In engineering applications, this switching mechanism also provides fault rollback capability: when the neural network inference module malfunctions, the system can automatically or manually switch back to the traditional compensation mode, utilizing the traditional compensation base to continue maintaining basic dead-zone suppression functionality, thus improving the system's engineering reliability and fault tolerance.

[0081] This application also provides a dead zone compensation device, please refer to... Figure 4 The dead zone compensation device includes: The acquisition module 10 is used to acquire the current operating condition feature vector and the traditional compensation voltage; Neural network module 20 is used to obtain residual harmonic coefficients by passing the current operating condition feature vector through a neural network; The residual compensation module 30 is used to obtain the residual compensation voltage based on the residual harmonic coefficients. The result module 40 is used to obtain the dead zone compensation voltage based on the traditional compensation voltage and the residual compensation voltage, and to perform dead zone compensation.

[0082] The dead-zone compensation device provided in this application, employing the dead-zone compensation method described in the above embodiments, solves the technical problem of how to achieve full-condition adaptive dead-zone compensation with low model complexity while maintaining the stability of traditional compensation. Compared with the prior art, the beneficial effects of the dead-zone compensation device provided in this application are the same as those of the dead-zone compensation method provided in the above embodiments, and other technical features in the dead-zone compensation device are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0083] This application provides a dead-zone compensation device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, which are executed by the at least one processor to enable the at least one processor to perform the dead-zone compensation method in Embodiment 1 above.

[0084] The following is for reference. Figure 5 The diagram illustrates a structural schematic suitable for implementing the dead-zone compensation device in the embodiments of this application. The dead-zone compensation device in the embodiments of this application may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 5 The dead zone compensation device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0085] like Figure 5As shown, the dead-zone compensation device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage device 1003 into a random access memory (RAM) 1004. The RAM 1004 also stores various programs and data required for the operation of the dead-zone compensation device. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to the I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the dead-zone compensation device to communicate wirelessly or wiredly with other devices to exchange data. Although dead-zone compensation devices with various systems are shown in the figures, it should be understood that implementation or possession of all the systems shown is not required. More or fewer systems may be implemented alternatively.

[0086] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0087] The dead-zone compensation device provided in this application, employing the dead-zone compensation method described in the above embodiments, solves the technical problem of how to achieve full-condition adaptive dead-zone compensation with low model complexity while maintaining the stability of traditional compensation. Compared with the prior art, the beneficial effects of the dead-zone compensation device provided in this application are the same as those of the dead-zone compensation method provided in the above embodiments, and other technical features of this dead-zone compensation device are the same as those disclosed in the previous embodiment method, and will not be repeated here.

[0088] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0089] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0090] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, the computer-readable program instructions being used to execute the dead-zone compensation method in the above embodiments.

[0091] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0092] The aforementioned computer-readable storage medium may be included in the dead-zone compensation device; or it may exist independently and not assembled into the dead-zone compensation device.

[0093] The aforementioned computer-readable storage medium carries one or more programs, which, when executed by the dead-zone compensation device, cause the dead-zone compensation device to: Obtain two-dimensional spatial information of rock cuttings fragmentation; Based on the two-dimensional spatial information of the rock fragments, a candidate connection relationship diagram is obtained by mapping. The candidate connection graph is merged to obtain an independent fragment set; Based on the aforementioned set of independent fragments, the area and morphological parameters of the rock debris fragments are obtained; The degree of rock fragmentation is obtained based on the area and morphological parameters of the fragmented rock material.

[0094] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0095] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0096] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0097] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the above-described dead-zone compensation method. This solves the technical problem of how to achieve full-condition adaptive dead-zone compensation with low model complexity while preserving the stability of traditional compensation methods. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the dead-zone compensation method provided in the above embodiments, and will not be repeated here.

[0098] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent scope of this application.

Claims

1. A dead zone compensation method, characterized in that, The method includes: Obtain the feature vector of the current operating condition and the traditional compensation voltage; The residual harmonic coefficients are obtained by passing the current operating condition feature vector through a neural network. The residual compensation voltage is obtained based on the residual harmonic coefficients. Based on the traditional compensation voltage and the residual compensation voltage, the dead zone compensation voltage is obtained, and dead zone compensation is performed.

2. The dead zone compensation method according to claim 1, characterized in that, Before obtaining the current operating condition feature vector and the traditional compensation voltage, the following steps are included: Obtain the operating characteristic quantities of the current working condition; Based on the operating characteristics of the current working condition, the traditional compensation voltage is obtained; The traditional compensation voltage is integrated with the operating characteristics of the current operating condition to obtain the current operating condition feature vector.

3. The dead zone compensation method according to claim 1, characterized in that, The neural network is obtained through offline training, and the offline training process includes: Under multiple preset operating conditions, the goal is to minimize the peak-to-peak value of electromagnetic torque, and the optimal residual harmonic coefficient is searched as the training label. The neural network is trained using the actual operating characteristics and the actual traditional compensation voltage under each working condition as input samples and the corresponding optimal residual harmonic coefficient as the output label.

4. The dead zone compensation method according to claim 1, characterized in that, The step of obtaining the residual harmonic coefficients by passing the current operating condition feature vector through a neural network includes: The current operating condition feature vector is passed through the hidden layer of the neural network to obtain the hidden feature vector; The hidden feature vector is mapped through the output layer of the neural network to obtain the residual harmonic coefficients.

5. The dead zone compensation method according to claim 1, characterized in that, The step of obtaining the residual compensation voltage based on the residual harmonic coefficient includes: Multiplying the residual harmonic coefficients by a sinusoidal basis function that is six times the rotor electrical angle yields the d-axis residual compensation voltage. The residual harmonic coefficient is multiplied by the cosine basis function of six times the rotor electrical angle to obtain the q-axis residual compensation voltage.

6. The dead zone compensation method according to claim 1, characterized in that, The step of obtaining the dead zone compensation voltage based on the conventional compensation voltage and the residual compensation voltage includes: The residual compensation voltage is limited to obtain the limited residual compensation voltage. The dead zone compensation voltage is obtained by superimposing the traditional compensation voltage with the residual compensation voltage after limiting.

7. The dead zone compensation method according to claim 1, characterized in that, The method further includes: Get the mode switching variable; Based on the value of the mode switching variable, the system switches between no-compensation mode, traditional compensation mode, and hybrid compensation mode; wherein, in the hybrid compensation mode, the dead zone compensation voltage is obtained by superimposing the traditional compensation voltage and the residual compensation voltage.

8. A dead-zone compensation device, characterized in that, include: The acquisition module is used to acquire the operating characteristic quantities of the current working condition; The traditional compensation module is used to obtain the traditional compensation voltage based on the operating characteristics of the current working condition; An integration module is used to integrate the traditional compensation voltage with the operating characteristics of the current operating condition to obtain the current operating condition feature vector. The neural network module is used to obtain the residual harmonic coefficients by passing the current operating condition feature vector through a neural network. The residual compensation module is used to obtain the residual compensation voltage based on the residual harmonic coefficients. The result module is used to obtain the dead zone compensation voltage based on the traditional compensation voltage and the residual compensation voltage, and to perform dead zone compensation.

9. A dead zone compensation device, characterized in that, The device includes: a memory, a processor, and a dead-zone compensation program stored on the memory and capable of running on the processor, the dead-zone compensation program being configured to implement the steps of the dead-zone compensation method as described in any one of claims 1 to 7.

10. A storage medium, characterized in that, The storage medium stores a dead-zone compensation program, which, when executed by a processor, implements the steps of the dead-zone compensation method as described in any one of claims 1 to 7.

Images