A Precise Control Method and System for Unmanned Aerial Vehicle Motors Based on Adaptive Algorithms

Abstract

This invention discloses a precise control method and system for UAV motors based on an adaptive algorithm, comprising: collecting motor operation data, dividing phase regions according to electrical angles, and mapping to generate segmented state sequences; identifying the effects of in-phase recurrence, generating phase debt units, and constructing a debt lattice structure by interlocking; extracting multi-channel state sequences, inputting them into an improved MCformer model, and obtaining debt acceptance results; identifying regions that meet the redemption closure conditions, constructing borrow redemption sidechains, and forming a locking graph; calculating the recovery index of each redemption path, selecting the optimal target path, and generating a corrected control sequence; dividing the witness sealing segments, configuring release priority conditions, and sequentially outputting PWM or current commands. This invention achieves high-precision stable operation and rapid dynamic response of UAV motors under complex operating conditions through an adaptive control mechanism that constructs an electrical angle phase debt lattice and a cross-phase redemption locking graph.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) control technology, and in particular to a method and system for precise control of UAV motors based on an adaptive algorithm. Background Technology

[0002] In scenarios such as inspection, surveying, logistics, and emergency operations, unmanned aerial vehicles (UAVs) typically rely on multi-motor coordinated drive to achieve lift output and attitude adjustment. The accuracy of motor control directly affects the UAV's flight stability, disturbance rejection capability, and energy utilization efficiency. Existing UAV motor control schemes generally achieve control through speed closed-loop, current closed-loop, or a combination of both, and combine electronic speed controllers (ESCs) to adjust the PWM duty cycle, current reference value, or voltage vector. Under normal flight conditions, this type of control can meet basic drive requirements. However, under complex operating conditions such as sudden gusts, rapid turns, rapid acceleration and deceleration, and long-term continuous operation, the actual motor response is simultaneously affected by multiple factors such as abrupt changes in aerodynamic load, electromagnetic rebound changes, temperature rise accumulation, and vibration propagation, resulting in significant dynamic non-uniformity in the control process.

[0003] Existing technologies typically address overall time-domain errors by directly performing in-phase, in-cycle compensation within the current control cycle based on the deviation between the target and actual speeds, phase current fluctuations, or temperature changes. While these approaches can correct instantaneous errors, they generally analyze the motor control response as a holistic sequence, lacking a segmented identification mechanism based on rotor electrical angle position. This makes it difficult to distinguish historical control effects that have been applied but not yet closed within a specific phase's effect area, and also difficult to determine whether such control effects continue to manifest in adjacent electrical angle areas and form in-phase recurrences over multiple electrical angle cycles. Furthermore, most existing control strategies employ in-situ compensation, applying corrections only in the phase where a deviation occurs. This lacks cross-phase borrowing and redemption mechanisms to address differences in the carrying capacity of different phase effect areas, and also lacks structured processing for locking and sequential constraints on redemption paths.

[0004] Existing technologies are prone to problems such as local overcompensation, insufficient phase current contraction, back EMF rebound mismatch, vibration diffusion amplification, and temperature rise burden accumulation under complex working conditions, which further lead to motor stabilization delay, discontinuous control output, and fluctuation in the overall machine attitude response.

[0005] Therefore, how to provide a precise control method and system for UAV motors based on adaptive algorithms is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose a precise control method and system for UAV motors based on an adaptive algorithm. This invention comprehensively utilizes electrical angle segmentation identification, phase debt lattice construction, cross-phase borrowing and transfer determination of the improved MCformer model, and witness sealing control mechanism to identify, migrate, redeem, and release the unclosed control effects of UAV motors under complex operating conditions. The invention details the precise control implementation process under conditions of sudden wind disturbances, rapid attitude changes, and temperature rise accumulation, and has the advantages of high control accuracy, good vibration suppression effect, low thermal burden, and fast recovery speed.

[0007] A precise motor control method for unmanned aerial vehicles based on an adaptive algorithm according to an embodiment of the present invention includes:

[0008] Collect the operating data of the UAV motor, divide the phase action area according to the rotor electrical angle position, map the operating data to the corresponding phase action area, and generate an electrical angle segmentation state sequence;

[0009] In each phase action region, the same phase recurrence control effect of the cross-electric angle region is identified, phase debt units are generated, and each phase debt unit is locked according to the same phase recurrence relationship in the electric angle period to construct an electric angle phase debt lattice.

[0010] For each phase debt unit in the electrical angle phase debt lattice, the multi-channel state sequence corresponding to the debt source phase slice and the candidate redemption phase slice is extracted, input into the improved MCformer model, and phase debt borrowing and transfer processing is performed to obtain the debt acceptance result corresponding to each candidate redemption phase slice.

[0011] Based on the debt assumption results, candidate redemption phase areas that meet the redemption closure conditions are identified, and phase debt units and candidate redemption phase areas are driven to form borrowed redemption edge chains to construct cross-phase redemption lock-in graphs.

[0012] In the cross-phase borrowing and redemption map, for the current phase debt unit, the comprehensive operation recovery index corresponding to each redemption path is calculated, the redemption path with the optimal comprehensive operation recovery index is selected as the target redemption path, and the modified control quantity sequence is generated.

[0013] The corrected control sequence is divided into multiple witness sealing segments. Each witness sealing segment is configured with an action range, release order, cancellation condition, and witnessing condition. Witness sealing segments that meet the witnessing conditions are released in sequence according to the release order, and PWM duty cycle control commands or current reference values ​​are output.

[0014] Optionally, the operating data includes target speed, actual speed, three-phase current, bus voltage, PWM duty cycle, motor winding temperature, back EMF estimate, rotor electrical angle position, body attitude angular velocity, and propeller vibration response.

[0015] Optionally, the step of dividing the phase action area according to the rotor electrical angle position and mapping the operating data to the corresponding phase action area includes:

[0016] The rotor mechanical angle is obtained and converted into the rotor electrical angle sequence by combining the number of motor pole pairs. Each electrical angle is defined as an electrical angle cycle of 360 degrees. Each electrical angle cycle is divided into N consecutive phase action regions according to a preset angle step size. Each phase action region is assigned a region identifier, where N is an integer greater than or equal to two. The corresponding rotor electrical angle position is extracted according to the sampling time of each operating data to determine the phase action region to which each operating data belongs. The operating data belonging to the same phase action region are collected in chronological order to generate the segmented state data of the corresponding phase action region. The segmented state data of each phase action region are arranged in the order of the electrical angle cycle to obtain the electrical angle segmented state sequence.

[0017] Optionally, the construction of the electrical angle phase debt lattice includes:

[0018] For each phase action zone in each electrical angle cycle, the change data of target speed, actual speed, phase current, back EMF, temperature and vibration are extracted, and combined with the continuous changes at the zone boundary, a zone drive response record of the phase action zone is formed.

[0019] Based on the area drive response record, identify the control effects that have been applied in the current phase action area but have not converged in the current phase action area and continue to be maintained in the electrical angle continuous area, and mark the control effects as cross-area continuous display control effects;

[0020] For each cross-regional continuous display control impact, a corresponding phase debt unit is generated, and the source phase region identifier, continuous display region identifier, continuous display direction, continuous display status and occurrence order are written in. The phase debt units in the same phase action region are arranged according to the occurrence order to form a debt unit sequence.

[0021] For debt unit sequences in adjacent electrical angle cycles that are at the same phase position, compare the display direction identifier, display area identifier, display holding status and occurrence order of each phase debt unit. When the content is consistent, determine that there is a same phase recurrence relationship between the corresponding phase debt units, and generate a locking mark for the phase debt units that have the same phase recurrence relationship.

[0022] Using phase debt units as lattice nodes, the continuity correspondence between source phase regions and continuity regions as region continuity edges, and the in-phase recurrence relationship with locking marks as periodic locking edges, the lattice nodes, region continuity edges, and periodic locking edges are associated and combined to construct an electrical angle phase debt lattice.

[0023] Optionally, obtaining the debt assumption results corresponding to each candidate redemption phase area includes:

[0024] For each phase debt unit, extract the speed change data, phase current change data, bus voltage change data, motor winding temperature change data, back EMF estimation change data, body attitude angular velocity change data, and propeller vibration response change data corresponding to the source phase area and candidate redemption phase areas to form a multi-channel state sequence corresponding to the source phase area and each candidate redemption phase area.

[0025] An improved MCformer model is constructed, which consists of a phase debt injection layer, a hybrid channel borrowing encoding layer, a transfer competition convergence layer, and a result output layer connected in sequence.

[0026] The phase debt injection layer writes the source phase area identifier, continuation area identifier, continuation direction identifier, continuation hold status, continuation termination status and occurrence order of the phase debt unit into the multi-channel state sequence corresponding to the source phase area, generating a source input sequence with debt marking;

[0027] The source input sequence with debt label and the multi-channel state sequence corresponding to the candidate redemption phase area are input into the hybrid channel borrowing coding layer. The channels are grouped according to driving response, electromagnetic rebound, thermal holding and attitude vibration, and hybrid channel borrowing coding is performed to obtain the borrowing coding result corresponding to each candidate redemption phase area.

[0028] Each borrowing code result is input into the transfer competition convergence layer. Based on whether the control influence in the source phase area can complete the acceptance closure in the candidate redemption phase area, whether it causes the temperature holding state to continue to increase, whether it causes the vibration response to continue to spread, and whether it maintains the consistent direction of the continuous manifestation, phase debt borrowing transfer processing is performed on each candidate redemption phase area to generate the corresponding transfer competition result.

[0029] The results of each transfer competition are input into the acceptance result output layer. The acceptance capacity of each candidate redemption phase area is sorted, and the acceptance priority, acceptance maintenance status and acceptance termination status are output to form the debt acceptance result corresponding to each candidate redemption phase area.

[0030] Optionally, constructing the cross-phase redemption locking graph includes:

[0031] For each phase debt unit, read the corresponding debt acceptance result, extract the acceptance order, acceptance hold status and acceptance termination status of each candidate redemption phase area, and generate the redemption candidate sequence corresponding to the phase debt unit according to the arrangement order of the candidate redemption phase areas in the electrical angle period.

[0032] For each candidate redemption phase segment in the redemption candidate sequence, check in turn whether the acceptance and maintenance status in the corresponding debt acceptance result is "continuously maintained" or "not terminated". Check whether the candidate redemption phase segment still maintains the same continuing display direction as the source phase debt unit after acceptance, and identify the candidate redemption phase segments that meet the redemption closure conditions.

[0033] For each candidate redemption phase segment that meets the redemption closure conditions, the corresponding phase debt unit is attached to the candidate redemption phase segment by borrowing, generating a borrowing redemption edge. Multiple borrowing redemption edges corresponding to the same phase debt unit are connected in sequence according to the order of acceptance to form a borrowing redemption edge chain corresponding to the phase debt unit.

[0034] For each borrowing redemption edge chain, compare the source phase area, redemption phase area and acceptance and maintenance status of adjacent borrowing redemption edges within the edge chain. When the redemption phase area of ​​adjacent borrowing redemption edges can form continuous acceptance of the same phase debt unit, establish a locking relationship between the corresponding borrowing redemption edges and mark the borrowing redemption edge chain with the locking relationship as a redemption locking chain.

[0035] Using phase debt units as the starting node of the graph, candidate redemption phase areas as the receiving nodes of the graph, borrowed redemption edges as the connecting edges of the graph, and redemption lock chains as the locking edge groups of the graph, the starting nodes, receiving nodes, connecting edges, and locking edge groups of the graph are associated and combined to construct a cross-phase redemption lock graph.

[0036] Optionally, generating the modified control quantity sequence includes:

[0037] Read each redemption lock chain corresponding to the phase debt unit in the cross-phase redemption lock map, extract the acceptance order, acceptance maintenance status, acceptance termination status, source phase area identifier and redemption phase area identifier corresponding to each borrowing redemption edge in each redemption lock chain, and form a path recovery candidate sequence.

[0038] For each path recovery candidate sequence, the changes in rotational speed, current, back EMF, temperature and vibration state of each redemption phase area before and after acceptance are extracted sequentially along the borrow redemption edge. Based on the consistency of the state change direction, the continuity of acceptance and the consistency of the display direction, the corresponding recovery continuous record is generated.

[0039] For each recovery continuous record, the recovery amount of speed stabilization, current contraction, springback closure, temperature rise and pressure drop, and vibration convergence is determined respectively, and the recovery continuity length, lockout holding length, and termination interruption position are further determined.

[0040] For each redemption lock chain, a basic recovery result is generated based on the degree of common maintenance of the speed stabilization recovery, current contraction recovery, springback closure recovery, temperature rise and pressure drop recovery, and vibration convergence recovery in the redemption lock chain. Then, a continuation gain processing is performed on the basic recovery result based on the recovery continuity length, a stabilization and reinforcement processing is performed on the basic recovery result based on the lock holding length, and an interruption deduction processing is performed on the basic recovery result based on the termination and interruption position to obtain the comprehensive operation recovery index corresponding to the redemption lock chain.

[0041] Compare the comprehensive operation recovery indicators corresponding to each redemption lock chain, determine the redemption lock chain with the optimal comprehensive operation recovery indicators as the target redemption path, extract the control correction amount of the corresponding redemption phase slice according to the arrangement order of each borrow redemption edge in the target redemption path, and generate the correction control amount sequence.

[0042] Optionally, the output PWM duty cycle control command or current reference value includes:

[0043] For the sequence of corrected control quantities corresponding to the target redemption path, each corrected control quantity is divided into multiple witness sealing fragments according to the phase order in the electrical angle cycle. Each witness sealing fragment is configured with a corresponding electrical angle action range, release order, cancellation condition and witnessing condition.

[0044] Read each witness sealing disc in sequence according to the release order, apply the corresponding control correction amount of the witness sealing disc within the corresponding electrical angle action range, and simultaneously collect the current phase current change state, back EMF change state and vibration response change state of the motor to generate the corresponding execution observation sequence.

[0045] For the currently executed witness sealing crack, the phase current, back electromotive force and vibration state are determined according to the execution observation sequence to see if the preset witness conditions are met. When the determination result is met, the next witness sealing crack in the release order is executed. When the determination result is not met, the cancellation condition of the current witness sealing crack is triggered and the unexecuted witness sealing crack is terminated.

[0046] After all witness sealing and splitting operations that meet the witnessing conditions are completed, the control correction values ​​corresponding to each witness sealing and splitting operation are superimposed according to the electrical angle action range to generate control output, and output PWM duty cycle control command or current reference value.

[0047] A precise control system for unmanned aerial vehicle (UAV) motors based on an adaptive algorithm, according to an embodiment of the present invention, includes:

[0048] The data slicing module is used to collect the operating data of the UAV motors and divide the phase action area to generate an electrical angle slicing state sequence;

[0049] A phase debt lattice construction module is used to identify the control effect of in-phase recurrence across regions based on the electrical angle sliced ​​state sequence, and to construct an electrical angle phase debt lattice.

[0050] The borrowing and transfer determination module is used to extract multi-channel state sequences based on the phase debt lattice of the electrical angle, input them into the improved MCformer model to perform phase debt borrowing and transfer processing, and obtain the debt acceptance result.

[0051] The redemption lockout graph module is used to borrow and link phase debt units with candidate redemption phase areas based on debt assumption results, forming borrowed redemption edge chains and constructing cross-phase redemption lockout graphs.

[0052] The path decision module is used to calculate the comprehensive operation recovery index of each redemption path based on the cross-phase redemption lockout map, determine the target redemption path, and generate a sequence of corrective control quantities.

[0053] The witness sealing module is used to divide the corrected control quantity sequence into segments and release segments that meet the witnessing conditions in sequence, and output PWM duty cycle control commands or current reference values.

[0054] The beneficial effects of this invention are:

[0055] Compared with existing technologies, this invention no longer performs direct compensation for the UAV motor based on the overall error in the time domain, but instead segments the operating state according to the rotor's electrical angle position. Within each phase action area, it identifies the recurring control influences that continue to appear across electrical angle areas, generates phase debt units, and constructs an electrical angle phase debt lattice. This allows it to separate historical control influences that have been applied but not yet closed under complex operating conditions from the overall response, achieving structured identification of the source, recurring position, and recurrence relationship of control influences. This improves the accuracy of control state identification for UAV motors under conditions of sudden wind disturbances, rapid attitude changes, and temperature accumulation, and reduces the problems of local misjudgment and repeated compensation caused by mixed processing of overall errors in existing technologies.

[0056] This invention, based on an improved MCformer model, performs phase debt borrowing and transfer processing on the multi-channel state sequences corresponding to the debt source phase region and the candidate redemption phase region, obtaining the debt acceptance result corresponding to the candidate redemption phase region. Based on this, a cross-phase redemption lockout graph is constructed, and the comprehensive operational recovery index corresponding to each redemption path is calculated. A target redemption path is selected to generate a modified control quantity sequence. Through this method, the compensation process, originally limited to the original phase, can be extended to a cross-phase redemption process oriented towards the acceptance capabilities of different phases. This allows control effects to be resolved within a more suitable phase region, effectively reducing problems such as local overcompensation, persistent phase current expansion, back EMF rebound mismatch, accumulated temperature rise burden, and vibration diffusion amplification, thereby improving the recovery efficiency and operational stability of UAV motors under complex operating conditions.

[0057] This invention further divides the corrected control sequence into multiple witness sealing segments and performs segment-by-segment release control based on the action range, release order, cancellation condition, and witness condition. When the witness condition is met, subsequent segments are released; when the witness condition is not met, cancellation and termination processing are performed. This transforms continuous control output into a verifiable, interruptible, and sequentially released sealing control process, ensuring that the application of the control correction is synchronized with the real-time operating state of the motor. This avoids the control overshoot and output oscillation problems caused by traditional one-time compensation methods, thereby improving the output stability of PWM duty cycle control commands or current reference values. This enables high-precision, low-vibration, low-heat-burden, and stabilizable control of UAV motors under complex operating conditions. Attached Figure Description

[0058] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0059] Figure 1 This is a flowchart of a precise motor control method for unmanned aerial vehicles based on an adaptive algorithm proposed in this invention;

[0060] Figure 2 This is a schematic diagram of the structure of a UAV motor precision control system based on an adaptive algorithm proposed in this invention. Detailed Implementation

[0061] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0062] refer to Figure 1 A precise control method for UAV motors based on an adaptive algorithm, comprising:

[0063] Collect the operating data of the UAV motor, divide the phase action area according to the rotor electrical angle position, map the operating data to the corresponding phase action area, and generate an electrical angle segmentation state sequence;

[0064] In each phase action region, the same phase recurrence control effect of the cross-electric angle region is identified, phase debt units are generated, and each phase debt unit is locked according to the same phase recurrence relationship in the electric angle period to construct an electric angle phase debt lattice.

[0065] For each phase debt unit in the electrical angle phase debt lattice, the multi-channel state sequence corresponding to the debt source phase slice and the candidate redemption phase slice is extracted, input into the improved MCformer model, and phase debt borrowing and transfer processing is performed to obtain the debt acceptance result corresponding to each candidate redemption phase slice.

[0066] Based on the debt assumption results, candidate redemption phase areas that meet the redemption closure conditions are identified, and phase debt units and candidate redemption phase areas are driven to form borrowed redemption edge chains to construct cross-phase redemption lock-in graphs.

[0067] In the cross-phase borrowing and redemption map, for the current phase debt unit, the comprehensive operation recovery index corresponding to each redemption path is calculated, the redemption path with the optimal comprehensive operation recovery index is selected as the target redemption path, and the modified control quantity sequence is generated.

[0068] The corrected control sequence is divided into multiple witness sealing segments. Each witness sealing segment is configured with an action range, release order, cancellation condition, and witnessing condition. Witness sealing segments that meet the witnessing conditions are released in sequence according to the release order, and PWM duty cycle control commands or current reference values ​​are output.

[0069] In this embodiment, the operating data includes target speed, actual speed, three-phase current, bus voltage, PWM duty cycle, motor winding temperature, back EMF estimate, rotor electrical angle position, body attitude angular velocity, and propeller vibration response.

[0070] In this embodiment, the step of dividing the phase action area according to the rotor electrical angle position and mapping the operating data to the corresponding phase action area includes:

[0071] The rotor mechanical angle is obtained and converted into the rotor electrical angle sequence by combining the number of motor pole pairs. Each electrical angle is defined as an electrical angle cycle of 360 degrees. Each electrical angle cycle is divided into N consecutive phase action regions according to a preset angle step size. Each phase action region is assigned a region identifier, where N is an integer greater than or equal to two. The corresponding rotor electrical angle position is extracted according to the sampling time of each operating data to determine the phase action region to which each operating data belongs. The operating data belonging to the same phase action region are collected in chronological order to generate the segmented state data of the corresponding phase action region. The segmented state data of each phase action region are arranged in the order of the electrical angle cycle to obtain the electrical angle segmented state sequence.

[0072] In this embodiment, constructing the electrical angle phase debt lattice includes:

[0073] For each phase action zone in each electrical angle cycle, the change data of target speed, actual speed, phase current, back EMF, temperature and vibration are extracted, and combined with the continuous changes at the zone boundary, a zone drive response record of the phase action zone is formed.

[0074] Based on the area drive response record, identify the control effects that have been applied in the current phase action area but have not converged in the current phase action area and continue to be maintained in the electrical angle continuous area, and mark the control effects as cross-area continuous display control effects;

[0075] For each cross-regional continuous display control effect, a corresponding phase debt unit is generated, and the source phase region identifier, continuous display region identifier, continuous display direction, continuous display status, and occurrence order are written into it. The phase debt units within the same phase effect region are arranged according to their occurrence order to form a debt unit sequence. Specifically, the generation of the corresponding phase debt unit for each cross-regional continuous display control effect is as follows:

[0076] Obtain the source phase area identifier and the continuing display area identifier corresponding to the cross-region continuing display control effect; obtain the start sampling position and end sampling position of the cross-region continuing display control effect within the source phase area and determine the occurrence order; determine the continuing display direction based on the state change direction of the cross-region continuing display control effect at the boundary of the source phase area; determine the continuing display state based on the maintenance status of the cross-region continuing display control effect within the continuing display area, and generate a phase debt unit containing the source phase area identifier, the continuing display area identifier, the occurrence order, the continuing display direction, and the continuing display state;

[0077] For debt unit sequences in adjacent electrical angle cycles that are at the same phase position, compare the display direction identifier, display area identifier, display holding status and occurrence order of each phase debt unit. When the content is consistent, determine that there is a same phase recurrence relationship between the corresponding phase debt units, and generate a locking mark for the phase debt units that have the same phase recurrence relationship.

[0078] Using phase debt units as lattice nodes, the continuity correspondence between the source phase region and the continuing phase region as the region continuity edge, and the in-phase recurrence relationship with locking marks as the periodic locking edge, the lattice nodes, region continuity edges, and periodic locking edges are associated and combined to construct an electrical angle phase debt lattice. The electrical angle phase debt lattice refers to a lattice where each phase debt unit is a lattice node, region continuity edges represent the continuity correspondence between the source phase region and the continuing phase region, and periodic locking edges represent adjacent electrical angle cycles. The graph structure set of in-phase recurrence relationships is described in the middle. Each lattice node carries the source phase region identifier, the recurrence region identifier, the recurrence direction, the recurrence state, and the occurrence order information. Each recurrence edge of the region carries at least the source phase region identifier and the recurrence region identifier information. Each periodic locking edge carries the in-phase position identifier and locking mark information. The lattice nodes, the recurrence edges of the regions, and the periodic locking edges are associated in the order of the electrical angle period to form a lattice structure used to characterize the control effect of recurrence and in-phase recurrence across electrical angle regions.

[0079] In this embodiment, obtaining the debt assumption results corresponding to each candidate redemption phase area includes:

[0080] For each phase debt unit, extract the speed change data, phase current change data, bus voltage change data, motor winding temperature change data, back EMF estimation change data, body attitude angular velocity change data, and propeller vibration response change data corresponding to the source phase area and candidate redemption phase areas to form a multi-channel state sequence corresponding to the source phase area and each candidate redemption phase area.

[0081] An improved MCformer model is constructed, which consists of a phase debt injection layer, a hybrid channel borrowing encoding layer, a transfer competition convergence layer, and a result output layer connected in sequence.

[0082] The improved MCformer model is derived from the hybrid channel modeling backbone of the traditional MCformer model. It retains the multi-channel joint encoding skeleton of the traditional MCformer model, adds a phase debt injection layer at the input end, and writes the source phase slice identifier, continuation slice identifier, continuation direction identifier, continuation hold state, continuation termination state, and occurrence order into the source sequence. The traditional channel hybrid encoding part is transformed into a hybrid channel borrowing encoding layer, which groups and encodes according to driving response, electromagnetic rebound, thermal hold, and attitude vibration. A newly added relinquishment competition convergence layer is connected after the encoding result to perform phase debt borrowing relinquishment processing on each candidate redemption phase slice. The original prediction output end is transformed into a receiving result output layer to output the receiving order, receiving hold state, and receiving termination state, thus forming the improved MCformer model.

[0083] The phase debt injection layer writes the source phase area identifier, continuation area identifier, continuation direction identifier, continuation hold status, continuation termination status and occurrence order of the phase debt unit into the multi-channel state sequence corresponding to the source phase area, generating a source input sequence with debt marking;

[0084] The source input sequence with debt tags and the multi-channel state sequence corresponding to the candidate redemption phase area are input into the hybrid channel borrowing encoding layer. Channels are grouped according to driving response, electromagnetic rebound, thermal hold, and attitude vibration, and hybrid channel borrowing encoding is performed to obtain the borrowing encoding results corresponding to each candidate redemption phase area. Specifically, the hybrid channel borrowing encoding process is as follows:

[0085] The source input sequence with debt markers and the multi-channel state sequences corresponding to the candidate redemption phase areas are aligned by time and mapped to the drive response channel group, electromagnetic rebound channel group, thermal holding channel group, and attitude vibration channel group, respectively, to obtain the in-group input sequence corresponding to each channel group. In-group borrowing correlation coding is performed on the in-group input sequences corresponding to each channel group to extract the state difference features, state continuity features, and state closure features of the source phase area and the candidate redemption phase area in the same channel group, to obtain the in-group borrowing coding results corresponding to each channel group. The in-group borrowing coding results corresponding to each channel group are concatenated across groups in the order of source phase area first and candidate redemption phase area last, and cross-group correlation coding is performed to generate the borrowing coding results corresponding to the candidate redemption phase area.

[0086] The borrowing codes are input into the transfer competition convergence layer. Based on whether the control influence in the source phase area can complete the acceptance closure within the candidate redemption phase area, whether it causes the temperature holding state to continue to strengthen, whether it causes the vibration response to continue to spread, and whether it maintains consistent display direction, phase debt borrowing transfer processing is performed on each candidate redemption phase area, generating the corresponding transfer competition results. Specifically, the phase debt borrowing transfer processing for each candidate redemption phase area is as follows:

[0087] Read the borrowing code results corresponding to each candidate redemption phase area, and extract the control influence representation corresponding to the source phase area, the acceptance response representation, temperature holding representation, vibration diffusion representation, and continuation direction representation corresponding to the candidate redemption phase area; align and compare the control influence representation corresponding to the source phase area with the acceptance response representation corresponding to the candidate redemption phase area to determine the acceptance closure state; compare the temperature holding representation, vibration diffusion representation, and continuation direction representation with the corresponding representations of the source phase area to determine the temperature holding change state, vibration diffusion change state, and continuation direction consistency state; based on the acceptance closure state, temperature holding change state, vibration diffusion change state, and continuation direction consistency state, generate a transfer permission mark, transfer holding mark, or transfer termination mark for each candidate redemption phase area, and form the transfer competition result of the corresponding candidate redemption phase area according to the marking results;

[0088] The results of each transfer competition are input into the debt assumption result output layer. The assumption capacity of each candidate redemption phase area is ranked, and the assumption priority, assumption maintenance status, and assumption termination status are output to form the debt assumption result corresponding to each candidate redemption phase area. The ranking of the assumption capacity of each candidate redemption phase area is as follows:

[0089] Read the transfer competition results corresponding to each candidate redemption phase region, and extract the transfer permission flag, transfer hold flag, and transfer termination flag. Select candidate redemption phase regions with the transfer permission flag as priority receiving regions, those with the transfer hold flag as secondary receiving regions, and those with the transfer termination flag as termination receiving regions. Within the priority receiving regions, arrange them according to the degree of completion of the receiving closure state, the degree of weakening of the temperature hold change state, the degree of convergence of the vibration diffusion change state, and the degree of maintaining the consistency of the continuing display direction. Within the secondary receiving regions, arrange them according to the continuity of the receiving hold state. Finally, output the receiving order of each candidate redemption phase region in the order of priority receiving region, secondary receiving region, and termination receiving region.

[0090] In this embodiment, constructing the cross-phase redemption locking graph includes:

[0091] For each phase debt unit, read the corresponding debt acceptance result, extract the acceptance order, acceptance hold status and acceptance termination status of each candidate redemption phase area, and generate the redemption candidate sequence corresponding to the phase debt unit according to the arrangement order of the candidate redemption phase areas in the electrical angle period.

[0092] For each candidate redemption phase segment in the redemption candidate sequence, check in turn whether the acceptance and maintenance status in the corresponding debt acceptance result is "continuously maintained" or "not terminated". Check whether the candidate redemption phase segment still maintains the same continuing display direction as the source phase debt unit after acceptance, and identify the candidate redemption phase segments that meet the redemption closure conditions.

[0093] For each candidate redemption phase segment that meets the redemption closure conditions, the corresponding phase debt unit is linked to the candidate redemption phase segment by borrowing, generating a borrowed redemption edge. Multiple borrowed redemption edges corresponding to the same phase debt unit are then chained together according to their order of acceptance, forming a borrowed redemption edge chain corresponding to the phase debt unit. Specifically, the borrowing linking of the corresponding phase debt unit to the candidate redemption phase segment is performed as follows:

[0094] Read the source phase area identifier, re-display area identifier, re-display direction, and re-display status of the phase debt unit, and read the candidate redemption phase area identifier and corresponding succession order that meet the redemption closure conditions; establish a borrowing association from the source phase area to the candidate redemption phase area, with the phase debt unit as the attachment starting point and the candidate redemption phase area as the attachment ending point; write the re-display direction and re-display status of the phase debt unit into the borrowing association to generate the corresponding borrowing redemption edge;

[0095] For each borrowing redemption edge chain, compare the source phase area, redemption phase area and acceptance and maintenance status of adjacent borrowing redemption edges within the edge chain. When the redemption phase area of ​​adjacent borrowing redemption edges can form continuous acceptance of the same phase debt unit, establish a locking relationship between the corresponding borrowing redemption edges and mark the borrowing redemption edge chain with the locking relationship as a redemption locking chain.

[0096] Using phase debt units as the starting node of the graph, candidate redemption phase areas as the receiving nodes, borrowed redemption edges as connecting edges, and redemption closed chains as closed edge groups, the graph's starting nodes, receiving nodes, connecting edges, and closed edge groups are associated and combined to construct a cross-phase redemption closed graph. This cross-phase redemption closed graph refers to a graph where phase debt units are the starting nodes, candidate redemption phase areas are the receiving nodes, and borrowed redemption edges represent the borrowed redemption phase areas. The borrowing and redemption edges that form a borrowing and redemption relationship in the selected redemption phase area are the graph connection edges, and the redemption lock chain that represents the continuous acceptance and locking relationship between multiple borrowing and redemption edges is the graph lock edge group. The graph start node, graph acceptance node, graph connection edge, and graph lock edge group are associated according to the acceptance order, acceptance maintenance state, and acceptance termination state. This is used to represent the borrowing and redemption path and locking acceptance relationship between phase debt units in different redemption phase areas.

[0097] In this embodiment, generating the corrected control quantity sequence includes:

[0098] Read each redemption lock chain corresponding to the phase debt unit in the cross-phase redemption lock map, extract the acceptance order, acceptance maintenance status, acceptance termination status, source phase area identifier and redemption phase area identifier corresponding to each borrowing redemption edge in each redemption lock chain, and form a path recovery candidate sequence.

[0099] For each path recovery candidate sequence, the changes in rotational speed, current, back EMF, temperature and vibration state of each redemption phase area before and after acceptance are extracted sequentially along the borrow redemption edge. Based on the consistency of the state change direction, the continuity of acceptance and the consistency of the display direction, the corresponding recovery continuous record is generated.

[0100] For each recovery continuous record, the recovery amounts for speed stabilization, current contraction, springback closure, temperature rise and pressure drop, and vibration convergence are determined respectively. Furthermore, the recovery continuity length, lockout holding length, and termination / interruption position are determined, where:

[0101] Read the deviation between the target speed and the actual speed of each redemption phase area before acceptance, and the deviation between the target speed and the actual speed after acceptance; compare the speed deviation before and after acceptance, and extract the change in the actual speed approaching the target speed; accumulate the change in the actual speed in each redemption phase area that continuously approaches the target speed without any reverse deviation expansion according to the borrowing redemption edge order, and determine the speed stabilization recovery amount;

[0102] Read the phase current expansion state of each redemption phase area before acceptance and the phase current contraction state after acceptance; compare the current change direction and change magnitude before and after acceptance, and extract the effective change amount from expansion to contraction; accumulate the effective change amount in each redemption phase area that has maintained the contraction direction and has not returned to the expansion state according to the borrowing redemption edge order to determine the current contraction recovery amount.

[0103] Read the back potential mismatch state of each redemption phase area before acceptance and the back potential rebound closure state after acceptance; compare the rebound correspondence before and after acceptance, and extract the effective improvement amount from mismatch to closure; accumulate the effective improvement amount of each redemption phase area that maintains the closure trend and has not reappeared the rebound mismatch according to the borrow redemption edge order, and determine the rebound closure recovery amount.

[0104] Read the temperature enhancement state of each redemption phase area before acceptance and the temperature reduction state after acceptance; compare the temperature maintenance trend before and after acceptance, and extract the effective change amount of temperature from continuous rise to controlled decline or slow decline; accumulate the effective change amount of temperature burden in each redemption phase area that has continuously weakened and has not returned to the enhancement state in the order of borrowing redemption edge to determine the temperature rise and pressure drop recovery amount.

[0105] Read the vibration diffusion state of each redemption phase area before acceptance and the vibration convergence state after acceptance; compare the vibration propagation trend and vibration amplitude change before and after acceptance, and extract the effective improvement amount from diffusion to convergence; accumulate the effective improvement amount of vibration in each redemption phase area that continues to converge and has not returned to diffusion state according to the borrowing redemption edge order to determine the vibration convergence recovery amount.

[0106] For each redemption lockout chain, a basic recovery result is generated based on the degree to which the recovery amounts of speed stabilization, current contraction, springback closure, temperature rise and pressure drop, and vibration convergence are maintained within the redemption lockout chain. Then, a continuation gain processing is applied to the basic recovery result based on the recovery continuity length, a stabilization and reinforcement processing is applied based on the lockout holding length, and an interruption deduction processing is applied based on the termination / interruption position. This yields the comprehensive operational recovery index corresponding to the redemption lockout chain. Specifically, the generation of the basic recovery result is as follows:

[0107] Read the speed stabilization recovery amount, current contraction recovery amount, springback closure recovery amount, temperature rise and voltage drop recovery amount, and vibration convergence recovery amount corresponding to the same redemption lock chain; sequentially check the holding state of each recovery amount in each borrow redemption edge of the redemption lock chain, and identify the combination of recovery amounts that are simultaneously in an effective holding state; take the speed stabilization recovery amount, current contraction recovery amount, springback closure recovery amount, temperature rise and voltage drop recovery amount, and vibration convergence recovery amount that are simultaneously in an effective holding state as a common recovery amount set; determine the basic recovery result of the redemption lock chain according to the continuous distribution range, common holding length, and consistent holding state of the common recovery amount set in the redemption lock chain;

[0108] Compare the comprehensive operation recovery indicators corresponding to each redemption lock chain, determine the redemption lock chain with the optimal comprehensive operation recovery indicators as the target redemption path, extract the control correction amount of the corresponding redemption phase slice according to the arrangement order of each borrow redemption edge in the target redemption path, and generate the correction control amount sequence.

[0109] In this embodiment, the output PWM duty cycle control command or current reference value includes:

[0110] For the sequence of corrected control quantities corresponding to the target redemption path, each corrected control quantity is divided into multiple witness sealing fragments according to the phase order in the electrical angle cycle. Each witness sealing fragment is configured with a corresponding electrical angle action range, release order, cancellation condition and witnessing condition.

[0111] Read each witness sealing disc in sequence according to the release order, apply the corresponding control correction amount of the witness sealing disc within the corresponding electrical angle action range, and simultaneously collect the current phase current change state, back EMF change state and vibration response change state of the motor to generate the corresponding execution observation sequence.

[0112] For the currently executed witness sealing crack, the phase current, back EMF, and vibration state are determined based on the execution observation sequence to see if the preset witness conditions are met. If the determination result is met, the next witness sealing crack in the release sequence is executed. If the determination result is not met, the cancellation condition of the current witness sealing crack is triggered, and the unexecuted witness sealing cracks are terminated. The preset witness conditions are that the phase current contraction condition, the back EMF rebound closure condition, and the vibration convergence condition are met simultaneously. Specifically, the phase current contraction condition is that the peak phase current at the end of the corresponding action interval of the current witness sealing crack is less than the peak phase current at the beginning of the action interval, and no current re-expansion occurs in the two consecutive sampling windows after the action interval. The back EMF rebound closure condition is that after the execution of the current witness sealing crack, the back EMF state changes from a mismatch state to a rebound closure state, and the rebound closure is maintained in the two consecutive sampling windows. The vibration convergence condition is that the peak vibration response at the end of the corresponding action interval of the current witness sealing crack is less than the peak vibration response at the beginning of the action interval, and no vibration re-diffusion occurs in the two consecutive sampling windows after the action interval.

[0113] After all witness sealing and splitting operations that meet the witnessing conditions are completed, the control correction values ​​corresponding to each witness sealing and splitting operation are superimposed according to the electrical angle action range to generate control output, and output PWM duty cycle control command or current reference value.

[0114] refer to Figure 2 A precision control system for drone motors based on an adaptive algorithm, comprising:

[0115] The data slicing module is used to collect the operating data of the UAV motors and divide the phase action area to generate an electrical angle slicing state sequence;

[0116] A phase debt lattice construction module is used to identify the control effect of in-phase recurrence across regions based on the electrical angle sliced ​​state sequence, and to construct an electrical angle phase debt lattice.

[0117] The borrowing and transfer determination module is used to extract multi-channel state sequences based on the phase debt lattice of the electrical angle, input them into the improved MCformer model to perform phase debt borrowing and transfer processing, and obtain the debt acceptance result.

[0118] The redemption lockout graph module is used to borrow and link phase debt units with candidate redemption phase areas based on debt assumption results, forming borrowed redemption edge chains and constructing cross-phase redemption lockout graphs.

[0119] The path decision module is used to calculate the comprehensive operation recovery index of each redemption path based on the cross-phase redemption lockout map, determine the target redemption path, and generate a sequence of corrective control quantities.

[0120] The witness sealing module is used to divide the corrected control quantity sequence into segments and release segments that meet the witnessing conditions in sequence, and output PWM duty cycle control commands or current reference values.

[0121] Example 1: During a continuous low-altitude patrol cycle of a multi-rotor UAV, the controller receives online operational data from one drive motor during a complex disturbance flight segment. The continuous sampling duration is 18 seconds, the sampling frequency is 2kHz, and a total of 36,000 sampling points are obtained. This flight segment includes hovering, lateral movement, wind resistance correction, and deceleration and re-acceleration processes. The target speed switches between 4520 r / min, 4280 r / min, and 4650 r / min. The bus voltage fluctuates between 21.8V and 23.4V. The peak three-phase current occurs between 7.6A and 10.1A. The winding temperature rises from 46.2℃ to 54.8℃. The peak body attitude angular velocity reaches 128° / s. The propeller vibration response amplitude varies between 0.18g and 0.64g. There is significant sudden wind disturbance in this segment. The maximum instantaneous speed downsurge caused by wind load reaches 312 r / min. Furthermore, in several electrical angle regions, there are historical control effects where the drive was applied but not closed. To verify the feasibility of the method of the present invention, the system pre-constructed 9600 sets of training samples, including 4200 sets of normal closure samples, 2800 sets of cross-regional continuation samples, 1600 sets of in-phase reproduction samples, and 1000 sets of failure samples; and 1200 sets of test samples, of which complex perturbation segment samples accounted for 41.7%.

[0122] After operation begins, the system first reads the rotor mechanical angle sequence and calculates the rotor electrical angle sequence by combining it with the number of motor pole pairs. Each 360° electrical angle is considered one electrical angle cycle, and each cycle is further divided into 12 phase action zones with a 30° step size. After mapping 36,000 sampling points within 18 seconds, a total of 1,492 complete electrical angle cycles are formed, resulting in 17,904 phase action zone segments. The system aggregates the target speed, actual speed, phase current, back EMF estimate, temperature, and vibration response by zone, forming an electrical angle segmented state sequence. Statistical analysis shows that the average speed deviation in zone 3 is 186 r / min, the average phase current expansion in zone 4 is 1.42 A, and the average vibration diffusion in zone 7 is 0.11 g. The continuous state difference at the zone boundaries is higher than the average value of the steady-state segments.

[0123] The system then identifies the in-phase recurrence control effect of continuous display across electrical angle zones within each phase action zone. Taking zone 3 as an example, the system extracts drive response records from 24 consecutive electrical angle cycles and finds that 17 cycles did not converge after the end of zone 3, but continued to maintain speed lag and current expansion in zone 4. In 11 of these cycles, the effect recurred in zone 3 of the next cycle. Based on this, the system generates 17 phase debt units and writes the source phase zone identifier, continuous display zone identifier, continuous display direction, continuous display status, and occurrence order. After processing other zones, a total of 41 phase debt units are generated, with 17 corresponding to zone 3, 9 to zone 4, 8 to zone 7, and 7 to the remaining zones. After locking according to the in-phase recurrence relationship, 9 debt locking chains are formed, with the longest chain spanning 5 consecutive electrical angle cycles and the shortest chain spanning 2 cycles. The final constructed electrical angle phase debt lattice contains 41 lattice nodes, 56 area continuation edges, and 23 periodic locking edges. Compared with the traditional time-domain in-phase direct compensation method, this invention additionally identifies 14 sets of cross-area continuation control effects that were not identified by the traditional method in this segment.

[0124] After the phase debt lattice is established, the system extracts multi-channel state sequences of the source phase region and candidate redemption phase regions for each phase debt unit, and inputs them into the improved MCformer model to perform phase debt borrowing and transfer processing. The improved MCformer model consists of four layers: a phase debt injection layer, a hybrid channel borrowing encoding layer, a transfer competition convergence layer, and a result output layer. Taking a group of phase debt units originating from the third region as an example, the system selects the fourth, fifth, sixth, and seventh regions as candidate redemption phase regions along the electrical angle borrowing search direction, and combines the rotational speed, current, voltage, temperature, back EMF, attitude angular velocity, and vibration data of the source region and each candidate region into four groups of multi-channel state sequences. After entering the hybrid channel borrowing encoding layer, the average response value of the drive response channel group reaches 0.82 in the sixth region, the electromagnetic rebound channel group reaches 0.74 in the fifth region, the thermal holding channel group is lowest in the seventh region at 0.31, and the attitude vibration channel group has the smallest disturbance coupling value in the sixth region at 0.27. After the competitive allocation process, the system assigns debts in the following order: Zone 6, Zone 5, Zone 7, and Zone 4. Zone 6 maintains its debt assignment status continuously, Zone 5 maintains it partially, Zone 7 is suitable for assignment but recovery is slow, and Zone 4 is deemed to have a high risk of termination. After processing all 41 phase debt units, a total of 126 debt assignment results are output, including 63 continuous holding results, 38 partially holding results, and 25 termination risk results.

[0125] The system constructs a cross-phase redemption lockout graph. For each phase debt unit, candidate redemption phase regions that meet the redemption closure conditions are first identified. Then, the phase debt unit is linked to the corresponding redemption phase region by borrowing, forming a borrowed redemption edge chain. Taking the debt unit in the third region as an example, the system identifies the sixth and fifth regions as meeting the redemption closure conditions, thus forming two borrowed redemption edges, which are then linked together in order of acceptance to form one edge chain. After processing all 41 phase debt units, a total of 58 borrowed redemption edges, 19 borrowed redemption edge chains, and 7 redemption lockout chains are formed, ultimately constructing a cross-phase redemption lockout graph. The longest lockout chain in this graph contains 5 borrowed redemption edges, and the shortest contains 2, indicating that the debt resolution process has changed from in-situ compensation to cross-phase ordered acceptance.

[0126] After completing the map construction, the system calculates the comprehensive operational recovery index for each redemption path. Taking three candidate locking chains as examples, the first chain has a rotational speed stabilization recovery of 241 r / min, a current contraction recovery of 1.16 A, a springback closure recovery of 26 μs, a temperature rise and pressure drop recovery of 0.38℃, and a vibration convergence recovery of 0.14 g. The corresponding data for the second chain are 228 r / min, 1.32 A, 31 μs, 0.44℃, and 0.19 g, respectively. The corresponding data for the third chain are 205 r / min, 0.97 A, 22 μs, 0.29℃, and 0.11 g, respectively. After combining the recovery continuity length, locking holding length, and interruption position, the system determines that the comprehensive operational recovery index of the second chain is optimal and generates a correction control sequence accordingly. This sequence contains four control segments, corresponding to four electrical angle action intervals, with correction amplitudes of 2.8%, 1.9%, 1.3%, and 0.8%, respectively.

[0127] Finally, the system divided the corrected control sequence into four witness-sealing segments and released them sequentially. After the first segment was released, the actual rotational speed increased from 4216 r / min to 4348 r / min, the peak phase current decreased from 9.4 A to 8.8 A, the back EMF rebound delay shortened from 118 μs to 95 μs, and the vibration response decreased from 0.52 g to 0.41 g, meeting the witnessing conditions. After the second segment was released, the actual rotational speed further increased to 4426 r / min, the phase current continued to shrink to 8.3 A, and the vibration decreased to 0.34 g. After the third segment was released, the winding temperature rise rate decreased from 0.46℃ / s to 0.23℃ / s. After the fourth segment was released, the actual rotational speed stabilized within the target speed range of ±58 r / min. The system finally output a PWM duty cycle control command, and the entire correction process lasted 0.17 s.

[0128] In the comparative experiment, using the same 9600 training samples and 1200 test samples, the method of this invention was compared with the traditional time-domain in-phase direct compensation method. For the aforementioned complex 18s segment, the stabilization time of the traditional method was 0.31s, while that of this invention was 0.17s; the average rotational speed stabilization error of the traditional method was 132 r / min, while that of this invention was 49 r / min; the peak phase current of the traditional method was 10.1A, while that of this invention was 8.8A; the average back EMF rebound delay of the traditional method was 127μs, while that of this invention was 93μs; the peak vibration of the traditional method was 0.57g, while that of this invention was 0.34g; and the cumulative increase in winding temperature rise of the traditional method was 3.1℃, while that of this invention was 1.9℃. On 1200 test samples, the traditional method had a recognition rate of 68.5% for the impact of cross-regional display control, while the present invention had a recognition rate of 91.8%; the traditional method had a redemption path effectiveness rate of 61.2%, while the present invention had an effectiveness rate of 88.4%; the traditional method had a secondary overcompensation occurrence rate of 27.6%, while the present invention reduced it to 8.9%; and the traditional method had a witness execution success rate of 70.3%, while the present invention had a success rate of 92.1%.

[0129] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A precise control method for UAV motors based on an adaptive algorithm, characterized in that, include: Collect the operating data of the UAV motor, divide the phase action area according to the rotor electrical angle position, map the operating data to the corresponding phase action area, and generate an electrical angle segmentation state sequence; In each phase action region, the same phase recurrence control effect of the cross-electric angle region is identified, phase debt units are generated, and each phase debt unit is locked according to the same phase recurrence relationship in the electric angle period to construct an electric angle phase debt lattice. For each phase debt unit in the electrical angle phase debt lattice, the multi-channel state sequence corresponding to the debt source phase slice and the candidate redemption phase slice is extracted, input into the improved MCformer model, and phase debt borrowing and transfer processing is performed to obtain the debt acceptance result corresponding to each candidate redemption phase slice. Based on the debt assumption results, candidate redemption phase areas that meet the redemption closure conditions are identified, and phase debt units and candidate redemption phase areas are driven to form borrowed redemption edge chains to construct cross-phase redemption lock-in graphs. In the cross-phase borrowing and redemption map, for the current phase debt unit, the comprehensive operation recovery index corresponding to each redemption path is calculated, the redemption path with the optimal comprehensive operation recovery index is selected as the target redemption path, and the modified control quantity sequence is generated. The corrected control sequence is divided into multiple witness sealing segments. Each witness sealing segment is configured with an action range, release order, cancellation condition, and witnessing condition. Witness sealing segments that meet the witnessing conditions are released in sequence according to the release order, and PWM duty cycle control commands or current reference values ​​are output.

2. The method for precise control of UAV motors based on an adaptive algorithm according to claim 1, characterized in that, The operating data includes target speed, actual speed, three-phase current, bus voltage, PWM duty cycle, motor winding temperature, back EMF estimate, rotor electrical angle position, body attitude angular velocity, and propeller vibration response.

3. The method for precise control of UAV motors based on an adaptive algorithm according to claim 1, characterized in that, The process of dividing the phase action area according to the rotor's electrical angle position and mapping the operating data to the corresponding phase action area includes: The rotor mechanical angle is obtained and converted into the rotor electrical angle sequence by combining the number of motor pole pairs. Each electrical angle is defined as an electrical angle cycle of 360 degrees. Each electrical angle cycle is divided into N consecutive phase action regions according to a preset angle step size. Each phase action region is assigned a region identifier, where N is an integer greater than or equal to two. The corresponding rotor electrical angle position is extracted according to the sampling time of each operating data to determine the phase action region to which each operating data belongs. The operating data belonging to the same phase action region are collected in chronological order to generate the segmented state data of the corresponding phase action region. The segmented state data of each phase action region are arranged in the order of the electrical angle cycle to obtain the electrical angle segmented state sequence.

4. The method for precise control of UAV motors based on an adaptive algorithm according to claim 1, characterized in that, The construction of the electrical angle phase debt lattice includes: For each phase action zone in each electrical angle cycle, the change data of target speed, actual speed, phase current, back EMF, temperature and vibration are extracted, and combined with the continuous changes at the zone boundary, a zone drive response record of the phase action zone is formed. Based on the area drive response record, identify the control effects that have been applied in the current phase action area but have not converged in the current phase action area and continue to be maintained in the electrical angle continuous area, and mark the control effects as cross-area continuous display control effects; For each cross-regional continuous display control impact, a corresponding phase debt unit is generated, and the source phase region identifier, continuous display region identifier, continuous display direction, continuous display status and occurrence order are written in. The phase debt units in the same phase action region are arranged according to the occurrence order to form a debt unit sequence. For debt unit sequences in adjacent electrical angle cycles that are at the same phase position, compare the display direction identifier, display area identifier, display holding status and occurrence order of each phase debt unit. When the content is consistent, determine that there is a same phase recurrence relationship between the corresponding phase debt units, and generate a locking mark for the phase debt units that have the same phase recurrence relationship. Using phase debt units as lattice nodes, the continuity correspondence between source phase regions and continuity regions as region continuity edges, and the in-phase recurrence relationship with locking marks as periodic locking edges, the lattice nodes, region continuity edges, and periodic locking edges are associated and combined to construct an electrical angle phase debt lattice.

5. The method for precise control of UAV motors based on an adaptive algorithm according to claim 1, characterized in that, The process of obtaining the debt assumption results corresponding to each candidate redemption phase area includes: For each phase debt unit, extract the speed change data, phase current change data, bus voltage change data, motor winding temperature change data, back EMF estimation change data, body attitude angular velocity change data, and propeller vibration response change data corresponding to the source phase area and candidate redemption phase areas to form a multi-channel state sequence corresponding to the source phase area and each candidate redemption phase area. An improved MCformer model is constructed, which consists of a phase debt injection layer, a hybrid channel borrowing encoding layer, a transfer competition convergence layer, and a result output layer connected in sequence. The phase debt injection layer writes the source phase area identifier, continuation area identifier, continuation direction identifier, continuation hold status, continuation termination status and occurrence order of the phase debt unit into the multi-channel state sequence corresponding to the source phase area, generating a source input sequence with debt marking; The source input sequence with debt label and the multi-channel state sequence corresponding to the candidate redemption phase area are input into the hybrid channel borrowing coding layer. The channels are grouped according to driving response, electromagnetic rebound, thermal holding and attitude vibration, and hybrid channel borrowing coding is performed to obtain the borrowing coding result corresponding to each candidate redemption phase area. Each borrowing code result is input into the transfer competition convergence layer. Based on whether the control influence in the source phase area can complete the acceptance closure in the candidate redemption phase area, whether it causes the temperature holding state to continue to increase, whether it causes the vibration response to continue to spread, and whether it maintains the consistent direction of the continuous manifestation, phase debt borrowing transfer processing is performed on each candidate redemption phase area to generate the corresponding transfer competition result. The results of each transfer competition are input into the acceptance result output layer. The acceptance capacity of each candidate redemption phase area is sorted, and the acceptance priority, acceptance maintenance status and acceptance termination status are output to form the debt acceptance result corresponding to each candidate redemption phase area.

6. The method for precise control of UAV motors based on an adaptive algorithm according to claim 1, characterized in that, The construction of the cross-phase redemption locking graph includes: For each phase debt unit, read the corresponding debt acceptance result, extract the acceptance order, acceptance hold status and acceptance termination status of each candidate redemption phase area, and generate the redemption candidate sequence corresponding to the phase debt unit according to the arrangement order of the candidate redemption phase areas in the electrical angle period. For each candidate redemption phase segment in the redemption candidate sequence, check in turn whether the acceptance and maintenance status in the corresponding debt acceptance result is "continuously maintained" or "not terminated". Check whether the candidate redemption phase segment still maintains the same continuing display direction as the source phase debt unit after acceptance, and identify the candidate redemption phase segments that meet the redemption closure conditions. For each candidate redemption phase segment that meets the redemption closure conditions, the corresponding phase debt unit is attached to the candidate redemption phase segment by borrowing, generating a borrowing redemption edge. Multiple borrowing redemption edges corresponding to the same phase debt unit are connected in sequence according to the order of acceptance to form a borrowing redemption edge chain corresponding to the phase debt unit. For each borrowing redemption edge chain, compare the source phase area, redemption phase area and acceptance and maintenance status of adjacent borrowing redemption edges within the edge chain. When the redemption phase area of ​​adjacent borrowing redemption edges can form continuous acceptance of the same phase debt unit, establish a locking relationship between the corresponding borrowing redemption edges and mark the borrowing redemption edge chain with the locking relationship as a redemption locking chain. Using phase debt units as the starting node of the graph, candidate redemption phase areas as the receiving nodes of the graph, borrowed redemption edges as the connecting edges of the graph, and redemption lock chains as the locking edge groups of the graph, the starting nodes, receiving nodes, connecting edges, and locking edge groups of the graph are associated and combined to construct a cross-phase redemption lock graph.

7. The method for precise control of UAV motors based on an adaptive algorithm according to claim 1, characterized in that, The generation of the corrected control quantity sequence includes: Read each redemption lock chain corresponding to the phase debt unit in the cross-phase redemption lock map, extract the acceptance order, acceptance maintenance status, acceptance termination status, source phase area identifier and redemption phase area identifier corresponding to each borrowing redemption edge in each redemption lock chain, and form a path recovery candidate sequence. For each path recovery candidate sequence, the changes in rotational speed, current, back EMF, temperature and vibration state of each redemption phase area before and after acceptance are extracted sequentially along the borrow redemption edge. Based on the consistency of the state change direction, the continuity of acceptance and the consistency of the display direction, the corresponding recovery continuous record is generated. For each recovery continuous record, the recovery amount of speed stabilization, current contraction, springback closure, temperature rise and pressure drop, and vibration convergence is determined respectively, and the recovery continuity length, lockout holding length, and termination interruption position are further determined. For each redemption lock chain, a basic recovery result is generated based on the degree of common maintenance of the speed stabilization recovery, current contraction recovery, springback closure recovery, temperature rise and pressure drop recovery, and vibration convergence recovery in the redemption lock chain. Then, a continuation gain processing is performed on the basic recovery result based on the recovery continuity length, a stabilization and reinforcement processing is performed on the basic recovery result based on the lock holding length, and an interruption deduction processing is performed on the basic recovery result based on the termination and interruption position to obtain the comprehensive operation recovery index corresponding to the redemption lock chain. Compare the comprehensive operation recovery indicators corresponding to each redemption lock chain, determine the redemption lock chain with the optimal comprehensive operation recovery indicators as the target redemption path, extract the control correction amount of the corresponding redemption phase slice according to the arrangement order of each borrow redemption edge in the target redemption path, and generate the correction control amount sequence.

8. The method for precise control of UAV motors based on an adaptive algorithm according to claim 1, characterized in that, The output PWM duty cycle control command or current reference value includes: For the sequence of corrected control quantities corresponding to the target redemption path, each corrected control quantity is divided into multiple witness sealing fragments according to the phase order in the electrical angle cycle. Each witness sealing fragment is configured with a corresponding electrical angle action range, release order, cancellation condition and witnessing condition. Read each witness sealing disc in sequence according to the release order, apply the corresponding control correction amount of the witness sealing disc within the corresponding electrical angle action range, and simultaneously collect the current phase current change state, back EMF change state and vibration response change state of the motor to generate the corresponding execution observation sequence. For the currently executed witness sealing crack, the phase current, back electromotive force and vibration state are determined according to the execution observation sequence to see if the preset witness conditions are met. When the determination result is met, the next witness sealing crack in the release order is executed. When the determination result is not met, the cancellation condition of the current witness sealing crack is triggered and the unexecuted witness sealing crack is terminated. After all witness sealing and splitting operations that meet the witnessing conditions are completed, the control correction values ​​corresponding to each witness sealing and splitting operation are superimposed according to the electrical angle action range to generate control output, and output PWM duty cycle control command or current reference value.

9. A precise control system for UAV motors based on an adaptive algorithm, executing the precise control method for UAV motors based on an adaptive algorithm as described in any one of claims 1 to 8, characterized in that, include: The data slicing module is used to collect the operating data of the UAV motors and divide the phase action area to generate an electrical angle slicing state sequence; A phase debt lattice construction module is used to identify the control effect of in-phase recurrence across regions based on the electrical angle sliced ​​state sequence, and to construct an electrical angle phase debt lattice. The borrowing and transfer determination module is used to extract multi-channel state sequences based on the phase debt lattice of the electrical angle, input them into the improved MCformer model to perform phase debt borrowing and transfer processing, and obtain the debt acceptance result. The redemption lockout graph module is used to borrow and link phase debt units with candidate redemption phase areas based on debt assumption results, forming borrowed redemption edge chains and constructing cross-phase redemption lockout graphs. The path decision module is used to calculate the comprehensive operation recovery index of each redemption path based on the cross-phase redemption lockout map, determine the target redemption path, and generate a sequence of corrective control quantities. The witness sealing module is used to divide the corrected control quantity sequence into segments and release segments that meet the witnessing conditions in sequence, and output PWM duty cycle control commands or current reference values.

Images