Phase current reconstruction method for high-modulation-ratio three-resistance sampling of permanent magnet synchronous motor

Abstract

The invention discloses a phase current reconstruction method for high-modulation-ratio three-resistance sampling of a permanent magnet synchronous motor. The phase current reconstruction method comprises the following steps: S1, obtaining a preliminary sampling frame; s2, generating boundary input data; s3, initializing a rule weight parameter set; s4, obtaining a weighted candidate current interval set; s5, performing cross check on the weighted candidate current interval set and the output of the comparator array, eliminating inference branches which do not conform to the polarity and amplitude limit of hardware, obtaining a candidate current interval set subjected to boundary reduction, and fusing the candidate current interval set with a reconstructed three-phase current vector of a previous pulse width modulation period to generate a reconstructed three-phase current vector of a current period; and S6, periodically and circularly executing the steps S1 to S5, and completing real-time closed-loop control of the phase current reconstruction method for high-modulation-ratio three-resistance sampling of the permanent magnet synchronous motor after the error of the reconstructed three-phase current vector is continuously kept to be smaller than a preset value under the high-modulation-ratio driving condition. According to the invention, steady-state fitting in the current estimation process is realized, and abrupt change points are prevented from generating abnormal output.

Description

Technical Field

[0001] The present invention relates to the technical field of permanent magnet synchronous motors, and in particular to a phase current reconstruction method for high modulation ratio three-resistance sampling of permanent magnet synchronous motors. Background Art

[0002] With the development of high-performance servo drive systems, permanent magnet synchronous motors are widely used in electric vehicles, elevators, and motor controller fields due to their high power density, excellent control performance, and energy-saving advantages. In practical applications, to achieve high-precision closed-loop control of the motor operating state, it is crucial to obtain accurate three-phase current information. Most traditional current sampling methods use Hall current sensors or shunt resistors to sample the phase current signals, and the analog-to-digital converter is used to obtain the values for the controller. However, under high modulation ratio driving conditions, that is, the operating state where the pulse width modulation signal is close to the full duty cycle, the sampling window is extremely short or some phase currents are in the dead zone shielding area, resulting in a significant decrease in the integrity and measurability of the three-phase current signals, directly affecting the accuracy of current detection.

[0003] Currently, the mainstream phase current reconstruction methods mostly rely on mathematical modeling and the derivation of space vector relationships. Under the architecture of the two-resistor method or the one-resistor method, current compensation and reconstruction are performed through the logical relationship between the space voltage vector and the current pulse width modulation sector number. However, it is usually assumed that the current change is linearly controllable, ignoring the disturbance effects of dead time, voltage utilization rate, and inductance characteristics on the current change boundary under high modulation ratio, resulting in an overly large or severely distorted inference interval under high modulation ratio, making it difficult to ensure real-time control accuracy.

[0004] In addition, although some improved algorithms using symbolic logic or fuzzy inference methods theoretically enhance the adaptability and robustness of the model, most of them still do not fully consider the interference factors of sampling saturation and inconsistent boundary polarities in hardware conditions, and lack a dynamic evaluation mechanism for the credibility of candidate inference intervals. Therefore, in practical control, problems such as cumulative current identification deviation, modulation instability, and even driving misoperation are likely to occur.

[0005] In summary, the existing technologies have significant defects in the three-resistance sampling reconstruction problem of permanent magnet synchronous motors under high modulation ratio: on the one hand, the traditional reconstruction methods have imperfect modeling of boundary constraint conditions and an overly wide inference range; on the other hand, the existing logical inference methods are still insufficient in dynamic adaptability and credibility evaluation mechanisms, making it difficult to meet the requirements of high-precision and strong-robustness current identification in actual industrial control systems, severely restricting the application effect of permanent magnet synchronous motors in high-performance scenarios. There is an urgent need to propose a more logically complete and real-time adaptable reconstruction method for complex constraint conditions of incomplete sampling and inconsistent polarity determination. Summary of the Invention

[0006] An object of the present invention is to propose a method for reconstructing phase current with three - resistor sampling for a permanent - magnet synchronous motor with a high modulation ratio. The present invention realizes the steady - state fitting of the current estimation process and avoids abnormal outputs at mutation points.

[0007] A method for reconstructing phase current with three - resistor sampling for a permanent - magnet synchronous motor with a high modulation ratio according to an embodiment of the present invention includes the following steps:

[0008] S1. Synchronously sample the three - phase current through three sampling resistors to obtain a preliminary sampling frame;

[0009] S2. Within the same pulse - width modulation period, read the current pulse - width modulation sector number, DC bus voltage utilization rate, sampling saturation flag, and the determination results of the comparator array on the polarities and amplitude thresholds of the three - phase sampling signals to generate boundary input data;

[0010] S3. Input the preliminary sampling frame and the boundary input data into the symbolic logic reasoning module, perform rule matching using the static boundary rule library, output the first candidate current interval set, and simultaneously initialize the rule weight parameter set;

[0011] S4. Call the dynamic weight adaptive layer in the symbolic logic reasoning module, update the rule weight parameter set according to the residual error between the reconstruction result of the previous cycle and the current preliminary sampling frame, and perform re - reasoning on the first candidate current interval set to obtain a weighted candidate current interval set;

[0012] S5. Cross - check the weighted candidate current interval set with the output of the comparator array, eliminate the inference branches that do not meet the hardware polarity and amplitude limits, obtain the candidate current interval set after boundary reduction, and fuse it with the reconstructed three - phase current vector of the previous pulse - width modulation cycle to generate the reconstructed three - phase current vector of the current cycle;

[0013] S6. Periodically and cyclically execute steps S1 to S5. After the reconstructed three - phase current vector continuously maintains an error less than a preset value under high - modulation - ratio driving conditions, the real - time closed - loop control of the method for reconstructing phase current with three - resistor sampling for a permanent - magnet synchronous motor with a high modulation ratio is completed.

[0014] Optionally, S1 includes the following steps:

[0015] S11. Initialize the pulse - width modulation control of the three - phase inverter of the permanent - magnet synchronous motor, and set the modulation ratio threshold γ th , when the modulation ratio γ ≥ γ th , enable the high - modulation - ratio observation mode. The modulation ratio γ is defined as:

[0016]

[0017] Among them, is the maximum value of the output voltage of the three-phase inverter, V dc is the DC bus voltage;

[0018] S12. In the high modulation ratio observation mode, three low-value sampling resistors R a , R b , R c are respectively connected in series to the motor A-phase, B-phase, and C-phase branches, and the instantaneous voltages across each sampling resistor are synchronously sampled through a high-precision analog-to-digital conversion module ;

[0019] S13. According to Ohm's law, the system divides the voltages across the three low-value sampling resistors by their corresponding resistance values respectively to calculate the observed values of the three-phase instantaneous currents. The observed values of the instantaneous currents are respectively for the A-phase, for the B-phase,

[0020] S14. The observed values of the three-phase instantaneous currents collected constitute a preliminary sampling frame within the current pulse width modulation period where t k represents the time stamp of the current sampling moment.

[0021] Optionally, the S2 includes the following steps:

[0022] S21. Read the sector number σ k of the current PWM modulation within the current pulse width modulation period T k ∈ {1, 2, 3, 4, 5, 6} to indicate the region to which the inverter output voltage vector belongs in the six-sector spatial division. The sector number has a direct correspondence with the current vector position;

[0023] S22. Synchronously obtain the DC bus voltage utilization rate μ k of the current period. The DC bus voltage utilization rate reflects the voltage modulation depth and dead zone masking degree:

[0024]

[0025] where represents the amplitude of the average output voltage of the three phases of the inverter within the kth modulation period, V dc represents the bus voltage;

[0026] S23. Identify the sampling states of the three-phase channels during the sampling process of the previous period, and record the sampling saturation flag set where indicates whether the input of the analog-to-digital converter is oversaturated in the x-phase during the kth period. 1 indicates oversaturation, 0 indicates normal, and x ∈ {a, b, c};

[0027] S24. Synchronously extract the polarity state of the three-phase instantaneous current signal and the judgment result of the amplitude threshold in the current cycle to obtain the polarity discrimination code stream Wherein:

[0028]

[0029] Wherein, represents the observed value of the instantaneous current of the x-phase in the current cycle, V th+ and V th- are the positive and negative threshold levels, which are used to delimit the boundary conditions of the current signal;

[0030] S25. Integrate the PWM sector number σ k of the current cycle, the bus voltage utilization rate μ k the sampling saturation flag set Θ k and the polarity discrimination code stream Π k to form the boundary input data set B k .

[0031] Optionally, the S3 includes the following steps:

[0032] S31. Input the preliminary sampling frame F init and the boundary input data set B k into the extended symbol logic reasoning module. The extended symbol logic reasoning module includes a hypergraph static boundary rule library R = {R1, R2,..., R m};

[0033] S32. According to the PWM sector number σ k of the current pulse width modulation cycle, the DC bus voltage utilization rate μ k and the stator phase inductance L s of the permanent magnet synchronous motor, construct a sector-voltage gradient parameter set in the symbol logic reasoning module. The sector-voltage gradient parameter set is used to characterize the maximum voltage change trend of each phase current under the given sector and bus conditions. The voltage gradient magnitude of each phase current is proportional to the unit voltage direction symbol of the phase in the current PWM sector number and is affected by the inductance characteristics;

[0034]

[0035] Wherein, is the unit voltage direction coefficient of the x-phase in the PWM sector;

[0036] S33. Combine the sector-voltage gradient parameter set with the preset dead time Δt dz, calculate the maximum disturbance amplitude of the three-phase current occurring within the current modulation period to form a current boundary change vector. The current boundary change vector describes the maximum increase and decrease amplitude of the three-phase current caused by the change in the drive voltage during the dead-time, and is respectively defined as the for phase A for phase B

[0037] S34. Based on the observed values of the three-phase instantaneous current and the corresponding current boundary change vector construct the first candidate current interval set for the current period The candidate interval of the candidate current for phase A is formed by extending the current instantaneous value of phase A upward and downward by a maximum disturbance amplitude Similarly for phases B and C, the corresponding candidate intervals are respectively formed. The candidate interval set reflects the physical feasible interval of the three-phase current under the current dead-time effect and voltage utilization conditions, and is used to limit the boundary range of the inference path in the symbolic logic inference module:

[0038]

[0039] S35. Initialize the rule weight parameter set Ω for the candidate rule subset k :

[0040]

[0041] where, Ω k represents the rule weight parameter set initialized for the candidate rule subset selected by the symbolic logic inference module under the current pulse width modulation period. Each weight parameter in the rule weight parameter set corresponds to the contribution degree of the rule R to the inference credibility of the phase current during this period. ‖Θ i ‖ represents the Hamming weight of the sampling saturation flag set, that is, the number of saturated states in the three-phase sampling channels during the current period. α k represents the sensitivity factor of the rule R i to the bus voltage utilization rate, and β i represents the suppression coefficient of the rule R i to the sampling saturation degree. i

[0042] Optionally, the construction method of the hypergraph static boundary rule library includes:

[0043] Based on the pulse width modulation geometric constraint, define the legal combination between the PWM sector number σ k and the polarity discrimination code stream Π k to construct a sector-polarity correspondence matrix;​

[0044] Based on the voltage-current physical constraints and combined with the DC bus voltage utilization rate μ k , dead time Δt dz and the motor stator phase inductance L s Establish a current slope model to describe the possible current change amplitude within a unit dead time;

[0045] Based on the sampling saturation flag set Θ k Define the inferability states of each phase channel and generate a sampling availability mask;

[0046] Map the boundary information to a six-dimensional discrete state node graph, where the edges represent the constraint coupling relationships between rules and the nodes represent the feasible phase current change patterns. Finally, fuse them to form a hypergraph-style static boundary rule base R with topological connectivity and electrical interpretability. Each rule R in the hypergraph-style static boundary rule base i Is mapped to a specific candidate current interval structure.

[0047] Optionally, S4 includes the following steps:

[0048] S41. Within the current pulse width modulation period T k , call the three-phase current reconstruction vector generated in the previous period T k-1 And perform differential calculation with the three-phase instantaneous current observation values in the preliminary sampling frame of the current period To obtain a three-phase residual error set, which reflects the difference between the reconstruction result of the previous period and the current actual sampling;

[0049] S42. Based on the three-phase residual error set, perform residual response analysis on each rule in the candidate rule subset Involved in symbolic logic reasoning during the current pulse width modulation period, and generate an error response function for the current period. The error response function is used to characterize the contribution of each rule to the current residual in the previous period reconstruction. The larger the residual amplitude of the error response function, the lower the fitness of the rule in the previous period;

[0050] S43. Use an exponential regularization decreasing mechanism to update the rule weight parameter set of the previous period, and introduce a regularization adjustment coefficient during the update process to obtain the rule weight parameter set Ω k of the current period. The rule weight parameter set is used to characterize the contribution degree of each rule to the credibility of symbolic logic reasoning in the current period;

[0051] S44. After completing the update of the rule weight parameter set, use the rule weight parameter set Ω k of the current period to Perform the re - reasoning operation, perform credibility weighting on each phase current under multiple candidate intervals through a weighted fusion strategy, and output a set of weighted candidate current intervals

[0052] Optionally, S5 includes the following steps:

[0053] S51. Read the set of weighted candidate current intervals within the current pulse - width modulation period T k and synchronously obtain the polarity discrimination code stream provided by the comparator array in the boundary input data set;

[0054] S52. For each phase current x ∈ {a, b, c}, construct an interval elimination rule based on symbol constraints in the set of weighted candidate current intervals Define the effective polarity direction according to the polarity discrimination value to generate a set of candidate current intervals after boundary consistency verification

[0055] S53. For each phase current x, construct a sliding weighted fitting window within the set of candidate current intervals The fitting window introduces the reconstructed current value of the previous period with the current period interval as the core as the regression reference, and perform a weighted least - deviation fitting strategy: where,

[0056]

[0057] represents the reconstructed current value of phase x in the current period, w1 and w2 are fitting weight coefficients that balance historical trends and inference stability, and δ i represents the deviation factor between the candidate point and the current central interval position, controlling the error growth caused by the central offset of the interval;

[0058] S54. Combine the fitting results of the three - phase currents and output the reconstructed three - phase current vector of the current period

[0059] When the set of candidate current intervals is non - empty and the fitting error is less than the set threshold, the reconstructed three - phase current vector is the minimum - deviation point within the interval;

[0060] When the set of candidate current intervals is empty, or the fitting error exceeds the limit, the reconstructed three - phase current vector value of the previous week is used as the compensation output.

[0061] Optionally, the effective polarity direction definition rule is:

[0062] When the polarity discrimination code stream Retain the part where the upper bound of the reserved interval is greater than 0;

[0063] When the polarity discrimination code stream Retain the part where the lower bound of the reserved interval is less than 0;

[0064] When the polarity discrimination code stream Retain the entire interval;

[0065] The rejection operation is performed separately for the three phases to generate a set of candidate current intervals after boundary consistency verification

[0066] The beneficial effects of the present invention are as follows:

[0067] (1) By constructing a hypergraph - type static boundary rule library, the present invention fuses the pulse - width modulation sector number, voltage utilization rate, sampling saturation state, and polarity information to construct a multi - dimensional rule node graph, realizes the logical modeling of complex factors such as dead - zone, inductance characteristics, and sampling polarity changes, and discretizes these complex physical boundary conditions into logical node constraints. On this basis, a set of rule weight parameters is introduced and dynamically updated through a periodic residual feedback mechanism, effectively avoiding the problem of mis - reasoning caused by insufficient generalization of the initial rule library, and enabling the candidate current interval to have a higher inference confidence within the physical boundary conditions.

[0068] (2) The present invention introduces a periodic residual error feedback mechanism. Based on the difference between the three - phase current reconstruction result of the previous cycle and the current - cycle preliminary sampling frame, a residual response function is constructed to perform fitness analysis on the symbolic inference rules, and the set of rule weight parameters is dynamically adjusted, making the inference process more inclined to the rule path with smaller residuals and stronger fitting ability, effectively overcoming the logical drift problems brought by polarity inversion and signal masking.

[0069] (3) Before the inference result is output, based on the polarity discrimination code stream, symbolic consistency rejection is performed on the candidate current interval to further exclude invalid intervals with logical conflicts. Subsequently, a sliding fitting window is constructed in combination with the reconstructed current vector of the previous cycle, and the current - cycle current estimation result is generated through a weighted least - deviation fitting strategy, comprehensively considering the historical trend, current interval confidence, and sampling boundary stability, realizing the steady - state fitting of the current estimation process and avoiding abnormal outputs at mutation points. Description of the Drawings

[0070] The drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the present invention, but do not constitute a limitation to the present invention. In the drawings:

[0071] Figure 1Flow chart of a phase current reconstruction method for high modulation ratio three - resistor sampling of a permanent magnet synchronous motor proposed by the present invention. Detailed implementation manners

[0072] Now, the present invention will be further described in detail with reference to the accompanying drawings. These drawings are all simplified schematic diagrams, only illustrating the basic structure of the present invention in a schematic way, so they only show the components related to the present invention.

[0073] Refer to Figure 1 , a phase current reconstruction method for high modulation ratio three - resistor sampling of a permanent magnet synchronous motor, comprising the following steps:

[0074] S1. Synchronously sample the three - phase current through three sampling resistors to obtain a preliminary sampling frame;

[0075] S2. Within the same pulse - width modulation period, read the current pulse - width modulation sector number, DC bus voltage utilization rate and sampling saturation flag, as well as the determination results of the comparator array on the polarities and amplitude thresholds of the three - phase sampling signals, and generate boundary input data;

[0076] S3. Input the preliminary sampling frame and the boundary input data into the symbolic logic reasoning module, perform rule matching using the static boundary rule library, output the first candidate current interval set, and simultaneously initialize the rule weight parameter set;

[0077] S4. Call the dynamic weight adaptive layer in the symbolic logic reasoning module, update the rule weight parameter set according to the residual error between the reconstruction result of the previous cycle and the current preliminary sampling frame, and perform re - reasoning on the first candidate current interval set to obtain a weighted candidate current interval set;

[0078] S5. Cross - check the weighted candidate current interval set with the output of the comparator array, eliminate the inference branches that do not meet the hardware polarity and amplitude limitations, obtain the candidate current interval set after boundary reduction, and fuse it with the reconstructed three - phase current vector of the previous pulse - width modulation cycle to generate the reconstructed three - phase current vector of the current cycle;

[0079] S6. Periodically and cyclically execute steps S1 to S5. After the reconstructed three - phase current vector continuously maintains an error less than a preset value under high modulation ratio driving conditions, the real - time closed - loop control of the phase current reconstruction method for high modulation ratio three - resistor sampling of the permanent magnet synchronous motor is completed.

[0080] In this embodiment, S1 includes the following steps:

[0081] S11. Initialize the pulse - width modulation control of the three - phase inverter of the permanent magnet synchronous motor, and set the modulation ratio threshold γ th , when the modulation ratio γ ≥ γ th , enable the high modulation ratio observation mode. The modulation ratio γ is defined as:

[0082]

[0083] wherein, is the maximum value of the output voltage of the three-phase inverter, V dc is the DC bus voltage;

[0084] S12. In the high modulation ratio observation mode, three low-resistance sampling resistors R a , R b , R c are respectively connected in series to the branches of the A-phase, B-phase and C-phase of the motor, and the instantaneous voltages at both ends of each sampling resistor are synchronously sampled through a high-precision analog-to-digital conversion module ;

[0085] S13. According to Ohm's law, the system divides the voltages at both ends of the three low-resistance sampling resistors by their corresponding resistance values respectively to calculate the observed values of the three-phase instantaneous currents. The observed values of the instantaneous currents are respectively for the A-phase, for the B-phase,

[0086] S14. The observed values of the three-phase instantaneous currents collected constitute a preliminary sampling frame within the current pulse width modulation period wherein, t k represents the time stamp of the current sampling moment.

[0087] In this embodiment, S2 includes the following steps:

[0088] S21. Read the sector number σ k of the current PWM modulation within the current pulse width modulation period T k ∈{1, 2, 3, 4, 5, 6} for indicating the region to which the inverter output voltage vector belongs in the six-sector division of space. The sector number has a direct correspondence with the position of the current vector;

[0089] S22. Synchronously obtain the DC bus voltage utilization rate μ k of the current cycle. The DC bus voltage utilization rate reflects the voltage modulation depth and dead zone masking degree:

[0090]

[0091] wherein, represents the amplitude of the average output voltage of the three phases of the inverter within the kth modulation cycle, V dc represents the bus voltage;

[0092] S23. Identify the sampling states of the three-phase channels during the sampling process of the previous cycle, and record the sampling saturation flag set Among them, represents whether the analog-to-digital converter input is oversaturated in the x-phase during the k-th period. 1 represents oversaturation, and 0 represents normal, where x ∈ {a, b, c};

[0093] S24. Synchronously extract the polarity state and amplitude threshold judgment result of the three-phase instantaneous current signal in the current period to obtain the polarity discrimination code stream Among them:

[0094]

[0095] Among them, represents the instantaneous current observation value of the x-phase in the current period, V th+ and V th- are the positive and negative threshold levels, which are used to delimit the boundary conditions of the current signal;

[0096] S25. Integrate the PWM sector number σ k , the bus voltage utilization rate μ k , the sampling saturation flag set Θ k and the polarity discrimination code stream Π k in the current period to form the boundary input data set B k .

[0097] In this embodiment, S3 includes the following steps:

[0098] S31. Input the preliminary sampling frame F init and the boundary input data set B k into the extended symbol logic reasoning module. The extended symbol logic reasoning module includes a hypergraph static boundary rule library R = {R1, R2,..., R m};

[0099] S32. According to the PWM sector number σ k , the DC bus voltage utilization rate μ k and the stator phase inductance L s of the permanent magnet synchronous motor in the current pulse width modulation period, construct a sector-voltage gradient parameter set in the symbol logic reasoning module. The sector-voltage gradient parameter set is used to characterize the maximum voltage change trend of each phase current under the given sector and bus conditions. The voltage gradient magnitude of each phase current is proportional to the unit voltage direction symbol of the phase in the current PWM sector number and is affected by the inductance characteristics;

[0100]

[0101] Among them, is the unit voltage direction coefficient of the x-phase in the PWM sector;

[0102] S33. Combine the sector-voltage gradient parameter set with the preset dead time Δt dz , calculate the maximum disturbance amplitude of the three-phase current in the current modulation period, and form a current boundary change vector. The current boundary change vector describes the maximum increase and decrease amplitude of the three-phase current caused by the change of the driving voltage during the dead time action, and is respectively defined as that of phase A that of phase B that of phase C

[0103] S34. Based on the observed values of the three-phase instantaneous current and the corresponding current boundary change vector construct the first candidate current interval set of the current period The candidate interval of the candidate current of phase A is formed by extending the current instantaneous value of phase A up and down by a maximum disturbance amplitude Similarly for phase B and phase C, and the corresponding candidate intervals are respectively formed. The candidate interval set reflects the physical feasible interval of the three-phase current under the current dead time action and voltage utilization conditions, and is used to limit the boundary range of the inference path in the symbolic logic inference module:

[0104]

[0105] S35. Initialize the rule weight parameter set Ω for the candidate rule subset k :

[0106]

[0107] where, Ω k represents the rule weight parameter set initialized for the candidate rule subset selected by the symbolic logic inference module under the current pulse width modulation period. Each weight parameter in the rule weight parameter set corresponds to the contribution degree of the rule R to the inference credibility of the phase current in this period. ‖Θ i ‖ represents the Hamming weight of the sampling saturation flag set, that is, the number of saturated states in the three-phase sampling channels in the current period. α k represents the sensitivity factor of the rule R i to the bus voltage utilization rate, and β i represents the suppression coefficient of the rule R i to the sampling saturation degree. i In this embodiment, the construction method of the hypergraph static boundary rule library includes:

[0108] Based on the pulse width modulation geometric constraint, define the PWM sector number σ

[0109]

[0109] k The legal combinations with the polarity discrimination code stream Π k are used to construct the sector-polarity correspondence matrix;

[0110] Based on the physical constraints of voltage-current, combined with the DC bus voltage utilization rate μ k , the dead time Δt dz and the motor stator phase inductance L s a current slope model is established to describe the possible current change amplitude within a unit dead time;

[0111] Based on the sampling saturation flag set Θ k the inferable states of each phase channel are defined, and a sampling availability mask is generated;

[0112] The boundary information is mapped to a six-dimensional discrete state node graph, where the edges represent the constraint coupling relationships between rules, and the nodes represent the feasible phase current change patterns. Finally, they are fused to form a hypergraph-style static boundary rule library R with topological connectivity and electrical interpretability. Each rule R in the hypergraph-style static boundary rule library i is mapped to a specific candidate current interval structure.

[0113] In this embodiment, S4 includes the following steps:

[0114] S41. Within the current pulse width modulation period T k , the three-phase current reconstruction vector generated in the previous period T k-1 is called and its difference is calculated with the three-phase instantaneous current observation values in the current period's preliminary sampling frame to obtain a three-phase residual error set, which reflects the difference between the reconstruction result of the previous period and the current actual sampling;

[0115] S42. Based on the three-phase residual error set, a residual response analysis is performed on each rule in the candidate rule subset involved in the symbolic logic reasoning within the current pulse width modulation period, and an error response function for the current period is generated. The error response function is used to characterize the contribution of each rule to the current residual in the previous period's reconstruction. The larger the residual amplitude of the error response function, the lower the fitness of the rule in the previous period;

[0116] S43. An exponential regularization decreasing mechanism is used to update the rule weight parameter set of the previous period, and a regularization adjustment coefficient is introduced during the update process to obtain the rule weight parameter set Ω k of the current period. The rule weight parameter set is used to characterize the contribution degree of each rule to the credibility of the symbolic logic reasoning in the current period;

[0117] ​S44. After completing the update of the rule weight parameter set, utilize the rule weight parameter set Ω of the current period k to perform re-inference operations on the first candidate current interval set by performing credibility weighting on each phase current under multiple candidate intervals through a weighted fusion strategy, and output a weighted candidate current interval set

[0118] In this embodiment, S5 includes the following steps:

[0119] S51. Read the weighted candidate current interval set within the current pulse width modulation period T k and synchronously obtain the polarity discrimination code stream provided by the comparator array in the boundary input data set;

[0120] S52. For each phase current x ∈ {a, b, c}, construct an interval elimination rule based on symbol constraints in the weighted candidate current interval set define the effective polarity direction according to the polarity discrimination value and generate a candidate current interval set after boundary consistency verification

[0121] S53. For each phase current x, construct a sliding weighted fitting window within the candidate current interval set The fitting window takes the current period interval as the core and introduces the reconstructed current value of the previous period as the regression reference and execute the weighted least deviation fitting strategy: where

[0122]

[0123] represents the reconstructed current value of phase x in the current period, w1 and w2 are fitting weight coefficients for balancing historical trends and inference stability, and δ i represents the deviation factor between the candidate point and the current center interval position, controlling the error growth caused by the interval center offset;

[0124] S54. Combine the fitting results of the three-phase currents and output the reconstructed three-phase current vector of the current period

[0125] When the candidate current interval set is non-empty and the fitting error is less than the set threshold, the reconstructed three-phase current vector is the minimum deviation point within the interval;

[0126] When the candidate current interval set is empty, or the fitting error exceeds the limit, then use the reconstructed three-phase current vector value of the previous week As a compensation output.

[0127] In this embodiment, the rule for defining the effective polarity direction is as follows:

[0128] When the polarity discrimination code stream The part where the upper bound of the reserved interval is greater than 0;

[0129] When the polarity discrimination code stream The part where the lower bound of the reserved interval is less than 0;

[0130] When the polarity discrimination code stream Reserve the entire interval;

[0131] The rejection operation is performed separately for the three phases to generate a set of candidate current intervals after boundary consistency verification

[0132] Example 1:

[0133] When Company A was conducting performance debugging on a new type of electric injection molding machine drive system, it encountered the problem that the permanent magnet synchronous motor frequently triggered overcurrent protection under high-speed and high-load conditions. The modulation ratio of this type of motor has been running at an extremely high level all year round, and has been stably maintained between 0.93 and 0.97 for a long time, belonging to a typical high-modulation ratio application scenario. The company's engineer, Liang Weiqing, found that under this modulation ratio, the traditional current reconstruction method using two resistors frequently led to incorrect judgments by the control system due to insufficient sampling windows or incorrect polarity inferences, resulting in abnormal current limiting and interference feedback, seriously restricting the continuity and stability of equipment operation.

[0134] To solve this problem, the project team decided to introduce the present invention on its development platform and arranged a three-day system comparison test. The test location was the automation equipment debugging room on the third floor of Building A of Hengtu Zhikong Experimental Center in Bao'an District, Shenzhen. TMS320F28379D was used as the main control chip, and a self-developed three-channel current sampling module was installed.

[0135] The project team completed the system initialization configuration, including motor parameter entry, current sampling channel calibration, PWM modulation setting. The three-phase sampling resistors were all 0.005 ohms, the sampling rate was 10 kHz, the test object was a permanent magnet synchronous motor with a rated current of 10 amperes, the DC bus voltage was 540 volts, and the load was 0.7 times the rated torque.

[0136] The formal comparison experiment began. The modulation ratios were set at three levels: 0.89, 0.94, and 0.97. Under the same working conditions, the traditional two-resistor inverse solution method and the algorithm of the present invention were respectively run, and key index data were recorded. For the convenience of unified analysis, all data were synchronously collected to the background server through the CAN port and recorded in CSV format.

[0137] Taking the group B experiment under a modulation ratio of 0.94 as an example, during the experiment that lasted for about 1 hour, a total of 840 groups of current data samples were collected. Due to the failure of the traditional algorithm to handle some polarity jumps and dead-time boundaries, the maximum error in current reconstruction reached 1.78 amperes, and the mean square error was 0.523 ampere squared. While under the same conditions, the maximum error of the method of the present invention was controlled within 0.88 amperes, the average error was 0.176, and the error fluctuation was only one-third of that of the traditional scheme.

[0138] At a specific moment, the system detected that the current of phase C fluctuated greatly. Due to misjudging the polarity direction, the traditional method identified the negative current as the positive direction, and the reconstructed value deviated from the actual value by more than 9 amperes, directly triggering the wrong current limit logic and causing the system speed limit action. While the method of the present invention correctly identified the polarity of the current in the current interval and output a current value close to the actual value through the boundary rule logic combined with the sampling upper and lower limit fluctuation intervals, and the system operated without error.

[0139] The project team further compared the frequencies of controller action interference caused by misjudgment of current reconstruction during the continuous operation of the two methods. Within 20 minutes of continuous operation, when the modulation ratio of the traditional method exceeded 0.95 under high load operation, a total of 11 misoperations occurred, and some occurred in the high-speed operation section, resulting in equipment pauses. After using the method of the present invention, the number of such misoperations was reduced to 2 times, and both were marginal samples with incomplete recognition of short-term voltage jitters, without affecting the normal operation of the equipment.

[0140] In addition, in order to further verify the universality of the algorithm under different working modes, the project team constructed three groups of AI training data sets, which were respectively from three typical working condition scenarios of low-speed constant torque, high-speed climbing section, and frequent load disturbances, with a total of 1850 samples, and applied them to the fitting parameter adjustment of the rule weight adaptive layer. The training results showed that after three rounds of training iterations, the errors of the sample sets converged stably to 0.084, 0.143, and 0.097 respectively, demonstrating extremely strong adaptability and learning efficiency.

[0141] In terms of operation stability, the method of the present invention showed obvious advantages. Under the extreme working condition where the bus voltage fluctuation amplitude was as high as 20%, the curve fluctuation of the three-phase current output decreased by more than 41% compared with the traditional scheme, the drive system response was smoother, and the system oscillation problem caused by error accumulation was effectively reduced.

[0142] After the experiment, the project team summarized the following performance indicators:

[0143] Table 1 Data comparison between the method of the present invention and the traditional algorithm

[0144]

[0145]

[0146] The experimental results fully verify that the method of the present invention has good current reconstruction accuracy, polarity recognition robustness and real-time inference ability under high modulation ratio and non-ideal sampling conditions. Subsequently, the company decided to solidify this algorithm in its new generation of T900 series industrial motor controllers and include it in the default product configuration for factory shipment in the first quarter of 2025.

[0147] The present invention constructs a hypergraph static boundary rule base, fuses the pulse width modulation sector number, voltage utilization rate, sampling saturation state and polarity information to construct a multi-dimensional rule node graph, realizes the logical modeling of complex factors such as dead zone, inductance characteristics and sampling polarity change, and discretizes these complex physical boundary conditions into logical node constraints. On this basis, a set of rule weight parameters is introduced and dynamically updated through a periodic residual feedback mechanism, effectively avoiding the problem of false inference caused by insufficient generalization of the initial rule base, and enabling the candidate current interval to have a higher inference confidence within the physical boundary conditions.

[0148] The present invention introduces a periodic residual error feedback mechanism. Based on the difference between the three-phase current reconstruction result of the previous cycle and the preliminary sampling frame of the current cycle, a residual response function is constructed to analyze the fitness of the symbolic inference rules, and the set of rule weight parameters is dynamically adjusted, making the inference process more inclined to the rule path with smaller residuals and stronger fitting ability, effectively overcoming the logical drift problem caused by polarity inversion and signal masking.

[0149] Before the inference result is output, the present invention performs symbolic consistency elimination on the candidate current interval based on the polarity discrimination code stream, further excluding invalid intervals with logical conflicts. Subsequently, a sliding fitting window is constructed by combining the reconstructed current vector of the previous cycle, and the current estimation result of the current cycle is generated through a weighted least deviation fitting strategy. Considering the historical trend, the current interval confidence and the sampling boundary stability, the steady-state fitting of the current estimation process is realized, avoiding abnormal output at mutation points.

[0150] As described above, only the preferred specific embodiments of the present invention are provided, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and all should be covered by the protection scope of the present invention.

Claims

1. A method for reconstructing phase current with high modulation ratio three-resistance sampling of a permanent magnet synchronous motor, characterized in that, Including the following steps: S1. Synchronously sample the three-phase current through three sampling resistors to obtain a preliminary sampling frame; S2. Within the same pulse width modulation period, read the current pulse width modulation sector number, DC bus voltage utilization rate, sampling saturation flag, and the determination results of the comparator array on the polarities and amplitude thresholds of the three-phase sampling signals, and generate boundary input data; S3. Input the preliminary sampling frame and the boundary input data into the symbolic logic inference module, perform rule matching using the static boundary rule library, output the first candidate current interval set, and simultaneously initialize the rule weight parameter set; S4. Invoke the dynamic weight adaptive layer in the symbolic logic inference module, update the rule weight parameter set according to the residual error between the reconstruction result of the previous cycle and the current preliminary sampling frame, and perform re-inference on the first candidate current interval set to obtain a weighted candidate current interval set; S5. Perform cross-checking on the weighted candidate current interval set and the output of the comparator array, eliminate the inference branches that do not meet the hardware polarity and amplitude limitations, obtain the candidate current interval set after boundary reduction, and fuse it with the reconstructed three-phase current vector of the previous pulse width modulation cycle to generate the reconstructed three-phase current vector of the current cycle; S6. Periodically and circularly execute steps S1 to S5. After the reconstructed three-phase current vector continuously maintains an error less than the preset value under the high modulation ratio driving condition, complete the real-time closed-loop control of the phase current reconstruction method for the permanent magnet synchronous motor with high modulation ratio three-resistor sampling.

2. The phase current reconstruction method of high modulation ratio three-resistance sampling for a permanent magnet synchronous motor according to claim 1, wherein The S1 includes the following steps: S11. Initialize the pulse width modulation control of the three-phase inverter of the permanent magnet synchronous motor and set the modulation ratio threshold γ th , when the modulation ratio γ ≥ γ th , enable the high modulation ratio observation mode; S12. In the high modulation ratio observation mode, three low-value sampling resistors R a , R b , R c are respectively connected in series to the branch circuits of the A-phase, B-phase, and C-phase of the motor, and the instantaneous voltages across each sampling resistor are synchronously sampled through a high-precision analog-to-digital conversion module ; S13. The system divides the voltages across the three low-resistance sampling resistors by their corresponding resistance values according to Ohm's law to calculate the observed values of the three-phase instantaneous currents. The observed values of the instantaneous currents are respectively for phase A for phase B for phase C S14. The collected three-phase instantaneous current observation values are used to form a preliminary sampling frame within the current pulse width modulation period where t k represents the time stamp at the current sampling moment.

3. A phase current reconstruction method for a permanent magnet synchronous motor with high modulation ratio three-resistor sampling according to claim 2, characterized in that, The S2 includes the following steps: S21. Read the sector number σ where the current PWM modulation is located within the current pulse width modulation period T k to indicate the area where the inverter output voltage vector belongs in the six-sector spatial division, where the sector number has a direct correspondence with the position of the current vector; k σ ∈ {1, 2, 3, 4, 5, 6} S22. Synchronously obtain the utilization rate μ of the DC bus voltage in the current cycle k , where the utilization rate of the DC bus voltage reflects the voltage modulation depth and the dead-time masking degree; S23. Identify the sampling status of the three-phase channels during the sampling process of the previous cycle, and record the sampling saturation flag set Among them, indicates whether the input of the analog-to-digital converter is oversaturated in the x-phase during the k-th cycle. 1 indicates oversaturation, and 0 indicates normal, where x ∈ {a, b, c}; S24. Synchronously extract the polarity state of the three-phase instantaneous current signal and the judgment result of the amplitude threshold in the current cycle to obtain the polarity discrimination code stream Wherein: Among them, represents the observed value of the instantaneous current of the x-phase in the current cycle, V th+ and V th- are the positive and negative threshold levels, which are used to define the boundary conditions of the current signal; S25. Integrate the PWM sector number σ k of the current cycle, the bus voltage utilization rate μ k , the sampling saturation flag set Θ k and the polarity discrimination code stream Π k to form the boundary input data set B k .

4. A phase current reconstruction method for high modulation ratio three-resistance sampling of a permanent magnet synchronous motor according to claim 3, characterized in that The S3 includes the following steps: S31. Input the preliminary sampling frame F init and the boundary input data set B k into the extended symbol logic inference module. The extended symbol logic inference module includes a hypergraph-based static boundary rule base R = {R1, R2, …, R m}; S32. According to the PWM sector number σ of the current pulse width modulation period k , the utilization rate μ of the DC bus voltage k , and the stator phase inductance L of the permanent magnet synchronous motor s , a sector-voltage gradient parameter set is constructed within the symbolic logic reasoning module. The sector-voltage gradient parameter set is used to characterize the maximum voltage change trend of each phase current under given sector and bus conditions. The magnitude of the voltage gradient of each phase current is proportional to the sign of the unit voltage direction of the phase in the current PWM sector number and the utilization rate of the DC bus voltage, and is affected by the inductance characteristics; S33. Combine the sector-voltage gradient parameter set with the preset dead time Δt dz , calculate the maximum disturbance amplitude of the three-phase current occurring within the current modulation period, and form a current boundary change vector. The current boundary change vector describes the maximum increase and decrease amplitude of the three-phase current caused by the change in the drive voltage during the dead time action, and is respectively defined as that of phase A that of phase B that of phase C S34. Based on the observed values of three-phase instantaneous current and the corresponding current boundary change vectors construct the set of first candidate current intervals for the current period The candidate interval of the candidate current of phase A extends up and down from the current instantaneous current value of phase A by a maximum disturbance amplitude to form. The same applies to phases B and C, respectively forming corresponding candidate intervals. The candidate interval set reflects the physical feasible intervals of the three-phase current under the current dead-time effect and voltage utilization conditions, and is used to limit the boundary range of the inference path in the symbolic logic inference module; S35. For the candidate rule subset Initialize the rule weight parameter set Ω k : Among them, Ω k represents the subset of candidate rules selected for the symbolic logic inference module in the current pulse width modulation period The set of rule weight parameters initialized and generated, and each weight parameter in the set of rule weight parameters corresponds to the rule R i The contribution degree of credibility to the phase current inference within this period, ‖Θ k ‖ represents the Hamming weight of the sampling saturation flag set, that is, the number of saturated states in the three-phase sampling channels in the current period, α i represents the rule R i The sensitivity factor for the utilization rate of the bus voltage, β i represents the rule R i The suppression coefficient for the sampling saturation degree.

5. A method for reconstructing phase current with three-resistor sampling of high modulation ratio for a permanent magnet synchronous motor according to claim 4, characterized in that The construction method of the hypergraph static boundary rule library includes: Define the PWM sector number σ based on the geometric constraints of pulse width modulation k and the polarity discrimination code stream Π k to construct a sector-polarity correspondence matrix for their legal combinations; Based on the voltage-current physical constraints, combined with the DC bus voltage utilization rate μ k , dead time Δt dz and the motor stator phase inductance L s A current slope model is established to describe the possible current change amplitude within a unit dead time; Based on the sampling saturation flag set Θ k Define the inferable states of each phase channel and generate a sampling availability mask; Map the boundary information into a six-dimensional discrete state node graph, where the edges represent the constraint coupling relationships between rules, and the nodes represent the feasible phase current change patterns. Finally, a hypergraph-style static boundary rule base R with topological connectivity and electrical interpretability is formed. Each rule R in the hypergraph-style static boundary rule base i is mapped to a specific candidate current interval structure.

6. A phase current reconstruction method for a permanent magnet synchronous motor with a high modulation ratio and three - resistance sampling, characterized in that, The S4 includes the following steps: S41. During the current pulse width modulation period T k call the three-phase current reconstruction vector generated in the previous period T k-1 and perform differential calculation on it and the three-phase instantaneous current observation values in the preliminary sampling frame of the current period to obtain a three-phase residual error set, which reflects the difference between the reconstruction result of the previous period and the current actual sampling; ​ S42. Based on the three-phase residual error set, perform residual response analysis on each rule in the candidate rule subset participating in symbolic logic inference within the current pulse width modulation period, and generate an error response function for the current period. The error response function is used to characterize the contribution of each rule to the current residual in the reconstruction of the previous period. The larger the residual amplitude of the error response function, the lower the fitness of the rule in the previous period; Among them, for each rule, perform residual response analysis to generate an error response function for the current period. The error response function is used to describe the contribution of each rule to the current residual in the reconstruction of the previous period. The larger the residual amplitude of the error response function, the lower the fitness of the rule in the previous period; S43. Update the set of rule weight parameters in the previous cycle by using an exponential regularization decay mechanism, and introduce a regularization adjustment coefficient during the update process to obtain the set of rule weight parameters Ω in the current cycle. k The set of rule weight parameters is used to represent the contribution degree of each rule to the credibility of symbolic logical reasoning in the current cycle. S44. After completing the update of the rule weight parameter set, use the rule weight parameter set Ω of the current period k to perform a re-inference operation on the first set of candidate current intervals by performing credibility weighting on each phase current under multiple candidate intervals through a weighted fusion strategy, and output a weighted set of candidate current intervals 7. A method for reconstructing phase current of a permanent magnet synchronous motor with high modulation ratio by three - resistor sampling, as claimed in claim 6, wherein The S5 includes the following steps: S51. Read the weighted candidate current interval set within the current pulse width modulation period T k and synchronously obtain the polarity discrimination code stream provided by the comparator array in the boundary input data set; For each phase current \(x\in\{a, b, c\}\), construct an interval elimination rule based on symbol constraints in the weighted candidate current interval set and define the effective polarity direction according to the polarity discrimination value to generate a candidate current interval set S53. For each phase current x, construct a sliding weighted fitting window in the candidate current interval set The fitting window takes the current cycle interval as the core and introduces the reconstructed current value of the previous cycle As the regression reference, execute the weighted least deviation fitting strategy: As the regression reference, execute the weighted least deviation fitting strategy: Among them, represents the reconstructed current value of the x-phase in the current cycle, w1 and w2 are fitting weight coefficients for balancing historical trends and inferred stability, and δ i represents the deviation factor between the candidate point and the current center interval position, controlling the error growth caused by the center offset of the control interval; Combine the three-phase current fitting results and output the reconstructed three-phase current vector for the current period When the set of candidate current intervals is non-empty and the fitting error is less than the set threshold, reconstruct the three-phase current vector as the minimum deviation point within the interval; When the set of candidate current intervals is empty or the fitting error exceeds the limit, the reconstructed three-phase current vector values of the previous week are used as the compensation output.

8. A method for reconstructing phase current with high modulation ratio three-resistance sampling of a permanent magnet synchronous motor according to claim 7, characterized in that, The effective polarity direction definition rule is: When the polarity discrimination bit stream Retain the part where the upper bound of the reserved interval is greater than 0; When the polarity discrimination bit stream Retain the part where the lower bound of the reserved interval is less than 0; When the polarity discrimination bit stream Reserve the entire interval; The rejection operation is performed separately for the three phases to generate a set of candidate current intervals after boundary consistency verification

Images