Brushless motor control system and control method

Abstract

The invention discloses a brushless motor control system and control method, and belongs to the technical field of motor control and driving. Comprising a real-time detection module, a control, regulation and control analysis module and a self-adaptive compensation module, multi-source operation data of the brushless motor are collected in real time, the control, regulation and control analysis module carries out depth correlation analysis, potential characteristics such as parameter sensitivity, observer coupling degree and stability attenuation index are mined, potential influences are quantified through analog simulation, and a comprehensive influence value is obtained. And the self-adaptive compensation module intelligently selects a compensation type according to the numerical value, calculates an accurate compensation data value, and performs dynamic closed-loop correction on controller parameters. According to the method, the problems of insufficient adaptive capacity and compensation lag caused by difficulty in coping with parameter time variation and potential coupling in the prior art are solved, the conversion from passive response to active prevention is realized, and the robustness, the control precision and the long-term operation stability under complex working conditions are improved.

Description

Technical Field

[0001] This invention relates to the field of motor control and drive technology, specifically to a brushless motor control system and control method. Background Technology

[0002] Brushless motors are widely used in industrial automation, electric vehicles, and home appliances due to their high efficiency, high reliability, and excellent speed regulation performance. Traditional brushless motor control systems, typically based on fixed control parameters or simple feedback adjustment, struggle to cope with the drift of internal parameters and changes in dynamic characteristics caused by factors such as temperature variations, component aging, and load disturbances under complex operating conditions. While some adaptive control methods exist, most focus on adjusting known, explicit variables, lacking the ability to deeply explore and proactively intervene in potential coupling relationships and implicit performance degradation trends within multi-source data. This results in lag in response and insufficient compensation accuracy when facing sudden disturbances or gradual performance degradation, affecting the stability, accuracy, and efficiency of the control. Summary of the Invention

[0003] The purpose of this invention is to address the problems of insufficient adaptive capability, delayed compensation, and lack of precision in existing brushless motor control systems when faced with time-varying internal parameters, multivariable coupling, and potential performance degradation. Therefore, this invention proposes a brushless motor control system and control method.

[0004] The objective of this invention can be achieved through the following technical solution: This invention provides a brushless motor control system, including a real-time detection module, a control and regulation analysis module, and an adaptive compensation module; The real-time detection module collects various real-time data of the brushless motor during operation, including electrical parameters, status parameters, control parameters, and physical parameters. The control and regulation analysis module receives real-time data and performs deep correlation analysis to obtain potential data correlation data values; it determines potential correlation based on the potential data correlation data values ​​to obtain potential data; it simulates brushless motor operation based on the potential data and the real-time data to obtain the impact of potential data and quantifies the simulated impact value of each potential data impact; and it corrects the simulated impact value to obtain the comprehensive impact value. The adaptive compensation module acquires the comprehensive impact value and performs compensation analysis to obtain the adaptive compensation data type and compensation data value, and then dynamically corrects the controller parameters based on the compensation data type and compensation data value.

[0005] In a preferred embodiment of the present invention, the process of receiving real-time data and performing deep correlation analysis to obtain potential data correlation values ​​is as follows: The real-time data is used to collect monitoring variables, and each variable is processed by timestamp alignment and normalization to obtain a normalized time series. The normalized time series is then sliced ​​according to a preset window length and step size to obtain each state slice. Based on each state slice, the nonlinear correlation between any two variables is calculated to form an instantaneous correlation tensor. A deep model is constructed based on the instantaneous correlation tensor and the state slice, and the potential correlation feature vector is output.

[0006] In a preferred embodiment of the present invention, the process of calculating the nonlinear correlation between any two sets of variables is as follows: For variables i and j within a slice, the mutual information is calculated using the mutual information formula. The weighted Euclidean distance between variables i and j is calculated using the weighted Euclidean distance formula. The mutual information and weighted Euclidean distance are fused using a hybrid correlation index formula. Calculate the mixed correlation of variables i and j in slice k. Where β is the mutual information contribution weight; This represents the maximum weighted Euclidean distance between all variable pairs in the k-th slice. To convert distance into similarity.

[0007] In a preferred embodiment of the present invention, the process of determining the potential correlation of potential data correlation values ​​is as follows: Based on a preset set of potential relevance determination thresholds, each potential feature in the potential relevance feature vector is determined to obtain a potential relevance identifier, which is then summarized into a potential relevance identifier set. All judgment results in the potential relevance identifier set are weighted and fused to obtain the potential relevance level for each time slice; The potential data is generated based on the potential relevance feature vector, the potential relevance identifier set, and the potential relevance level.

[0008] As a preferred embodiment of the present invention, the influence values ​​are substituted into the actual operation process of the brushless motor, specifically as follows: Based on real-time collected multi-source operating data, the model instance that best matches the current operating state is selected from the pre-set fully parameterized motor model library to obtain the optimal set of model parameters. Based on the potential data set and its potential correlation data values, the relevant potential parameters are substituted into the high-fidelity reduced-order equivalent model for calculation, generating multiple sets of perturbation parameter sets; For each disturbance scenario's disturbance parameter set, numerical simulation and analysis are performed to calculate the changes in key performance indicators; the changes in key performance indicators are then input into a pre-trained impact analysis neural network model to obtain comprehensive simulated impact values.

[0009] As a preferred embodiment of the present invention, the process of obtaining comprehensive impact values ​​and performing compensation analysis is as follows: The current compensation mode is determined based on the comprehensive impact value. The compensation modes include no compensation, slight compensation, moderate compensation, and strong compensation. Based on the compensation mode, select the corresponding compensation data type from the predetermined set of compensation types; Based on the comprehensive impact value I (p) Sensitivity factor s of the parameter i and the corresponding action coefficient η for the compensation type i Through formula C i =γ×I (p) ×∣s i ∣×η i The calculated compensation data value C i , where γ is the global compensation gain coefficient; The compensation data type and the compensation data value are output to the controller parameter correction unit to generate a set of controller parameters after real-time correction.

[0010] Another aspect of the present invention provides a brushless motor control method, comprising the following steps: Step S1: Obtain real-time data such as electrical parameters, status parameters, and temperature of the brushless motor; Step S2: Perform deep correlation analysis on the real-time data to obtain potential data and potential correlation data values; Step S3: Build an operational model based on potential data and real-time data, perform disturbance simulation, obtain the impact of potential data, and quantify the simulated impact value; Step S4: Correct the simulated impact values ​​and the actual operational deviations to obtain the comprehensive impact values; Step S5: Generate compensation data types and compensation data values ​​based on the comprehensive impact values; Step S6: Perform real-time adaptive dynamic correction of the controller parameters based on the compensation data.

[0011] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention collects multi-source data through a real-time detection module and performs in-depth correlation analysis and operation simulation using a control and regulation analysis module. It can extract potential characteristics such as parameter sensitivity, observer coupling degree, and stability decay index from massive operational data, quantify the impact of potential risks, and achieve accurate insight into the latent state and future trend of the system.

[0012] 2. Based on the comprehensive impact values ​​obtained from in-depth analysis, the adaptive compensation module of this invention can automatically determine the risk level, intelligently match the compensation type (such as gain, observer parameters, filter coefficients, etc.), and calculate accurate compensation data values ​​to perform real-time, dynamic closed-loop correction of controller parameters. This enables the system to proactively adjust before potential problems become explicit, improving the system's robustness, control accuracy, and long-term operational stability, achieving a leap from passive response to proactive prevention. Attached Figure Description

[0013] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0014] Figure 1 This is a schematic diagram of the principle of the present invention; Figure 2 This is a flowchart illustrating the steps and methods of the present invention. Detailed Implementation

[0015] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0016] It should be understood that the terms “comprising” and “including” used in this disclosure and claims indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0017] It should also be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of this disclosure. As used in this disclosure and claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this disclosure and claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes such combinations.

[0018] Please see Figure 1 As shown, one aspect of the present invention provides a brushless motor control system, including: a real-time detection module, a control and regulation analysis module, and an adaptive compensation module; The real-time detection module collects various real-time data of the brushless motor during operation, including electrical parameters such as three-phase current and DC bus voltage, status parameters such as rotor position and speed, control parameters such as current loop output, and physical parameters such as motor and driver temperature.

[0019] The control and regulation analysis module receives all data and performs in-depth correlation analysis to obtain potential data correlation values, and determines potential correlation based on these values; thus obtaining potential data. Based on the potential data and all data, a brushless motor operation simulation was performed to obtain the impact of the potential data; and the simulated impact value of each potential data impact during the brushless motor operation was quantified. Furthermore, the influence values ​​are substituted into the actual operation of the brushless motor to correct the simulated influence values ​​and obtain the comprehensive influence values. The adaptive compensation module acquires the comprehensive impact values ​​for compensation analysis, obtaining the adaptive compensation data type and compensation data value; and then dynamically corrects the controller parameters in real time.

[0020] The process by which the control and regulation analysis module receives all data and performs deep correlation analysis to obtain potential data correlation values ​​is as follows: Let M be the number of several monitored variables in the real-time data, and let X = {x1(t), x2(t), ..., x...} be the original data. M (t)}; Perform timestamp alignment and normalization on each variable to obtain the normalized time series. ,Right now: We obtain, where x i (t) represents the original time series of the i-th variable, μi is the mean, and σ is the mean. i Standard deviation; Next, the sequence is sliced ​​according to the window length W and the step size S to obtain the k-th state slice S. k ,Right now: Where W is the window length (e.g., one electrical cycle), S is the sliding step size between adjacent slices, and t k This is the start time of the k-th window; It should be noted that: each slice S k It represents the local intact state within a certain time period, and all subsequent correlation calculations will be based on a single slice.

[0021] Based on each slice S k The nonlinear correlation between any two sets of variables is calculated to form an instantaneous correlation tensor. The specific process is as follows: For slice S k The variables i and j within the formula are used to... Output the mutual information between variables i and j in slice k. , Let be the joint probability that variable i falls within interval a and variable j falls within interval b. These are the marginal probabilities of variables i and j, respectively; The process of calculating the distance between variables i and j in a slice: using the formula Output the weighted Euclidean distance between variables i and j , where ω(t) is the weight coefficient (the closer to the current time, the greater the weight); (where α is the weight decay coefficient). The mutual information and Euclidean distance are fused to obtain the mixed correlation of variables i and j in slice k; that is, by constructing a mixed correlation index: through the formula, Mixed correlation of output variables i and j in slice k Where β is the mutual information contribution weight, 0≤β≤1; This represents the maximum weighted Euclidean distance between all variable pairs in the k-th slice. To convert distance into similarity; Then combine all the mixed correlations Arrange the variables in order to form a three-dimensional tensor; C k Let be the instantaneous correlation tensor of the k-th slice, with dimensions M×M.

[0022] Based on three-dimensional tensor C k and slice S k Building a deep model, specifically: Depth model with slice S k The variables are graph nodes, represented by the three-dimensional tensor C of their correlations. k Let the edge weights be dynamic; let the graph structure be G. k =(V, E) k ), where V = {v1, v2, ..., v} M}E is a node for each variable. k Let E be the dynamic edge set of the k-th time slice, and E k Including A k =C k .

[0023] The update formula for each layer is as follows: In the process of updating the formula, No. Layer node feature matrix, A k−τ Let be the adjacency matrix for the first τ time steps. T is the weight matrix of the spatiotemporal convolution kernel, σ(·) is the activation function (such as ReLU), and T is the weight matrix of the spatiotemporal convolution kernel. s The temporal convolution depth; Then, by performing a heaping operation on each layer, the high-order latent feature matrix of the final layer's output is obtained. ,Right now: H (0) =S k .

[0024] Finally, through the model formula Z k =f( Output the latent correlation feature vector Z k Where f(·) is the feature aggregation function, used to compress high-dimensional graph features into low-dimensional indices; Output the latent correlation feature vector Z from the deep model k And Z k ={z1, z2, ..., z P}, where each z P Characterize a certain type of potential correlation features of the system in the k-th time slice, including parameter sensitivity factor, observer coupling degree, stability decay index, etc. Furthermore, a preset threshold set φ for determining potential relevance is defined. φ = {θ1, θ2, ..., θ...} P} is determined in the following way: for each latent feature z p Calculate its distribution in a historical dataset covering multiple system states, and analyze z p The correspondence between the numerical value and the actual operating state is determined, and an appropriate statistic (such as a specific percentile value in the "normal" state sample set) is selected as the threshold θp.

[0025] The threshold set φ is determined based on a preset potential relevance, where φ = {θ1, θ2, ..., θ}. P}; for each threshold θ P Used to define the corresponding potential feature z P Whether the triggering conditions for potential relevance have been met; the threshold is derived from historical data statistics.

[0026] For each latent feature z p Execution function determination: Obtain the set of potential relevance identifiers R k ={R1, R2, ..., R P}; where R P When R = 1, it indicates that the feature has potential correlation. P When the value is 0, it means that the feature has not met the potential relevance condition.

[0027] All judgment results are weighted and fused to obtain the potential relevance level of the k-th time slice. Through the formula: We obtain, where w pPreset indicator weights to emphasize the importance of key features; Based on the calculation The potential states of the system are classified into levels, when When L1 < L1, it is considered a normal state; when L1 ≤ L1, it is considered a normal state. When L2 < L2, it is considered a slight potential coupling; when L2 ≤ When L < L3, it is judged as a significant potential correlation; when L k When the value is ≥L3, it is considered a strong potential risk.

[0028] Finally, based on the latent correlation feature vector Z k Potential correlation marker R k and potential relevance level L k Generate potential data: D k =g(Z k R k L k ).

[0029] Furthermore, based on the latent correlation feature vector Z k Row decomposition yields the sensitivity factor, observer coupling degree, and stability decay index of the corresponding parameters; The sensitivity factor is the critical energy measurement y versus a certain parameter p. i The sensitivity is calculated as follows: Output sensitivity factor ;where p i Let y be a parameter related to the motor or drive in the system, and y be a critical energy measurement. The model outputs its sensitivity to input variables. The partial derivative of the performance with respect to the parameters represents the degree of influence of small changes; The observer coupling degree is the position observer error ε. o With current harmonic components h n The correlation, calculation process Output observer coupling ,in These are the vectors representing the observer error and the nth harmonic in the characteristic space, respectively. The stability decay index is a comprehensive indicator for predicting future oscillation trends, expressed by the formula... Output stability decay index ;in Let A be the k-th correlation matrix. k The largest eigenvalue, is the time decay factor, γ is the time decay coefficient, and γ > 0.

[0030] Based on the potential data and all data, a brushless motor operation simulation was performed to obtain the impact of the potential data; and the simulated impact value of each potential data influence during the brushless motor operation was quantified; specifically: Based on real-time collected multi-source operating data (including but not limited to current, voltage, speed, and temperature data), the model m instance that best matches the current operating state is selected from the pre-set fully parameterized motor model library (the m-th candidate model in the pre-set model library has parameters Φ). (m) ); The matching error is calculated based on a set of key physical quantities, using the formula The matching error of output model m Where Q is the set of observations involved in the matching. To normalize the weights, The physical quantities calculated for model m These are real-time measured values; then based on... Obtain the optimal model parameter set ;in, Index the selected optimal model; Select the model with the highest matching degree. The corresponding parameter set The instance is used as a high-fidelity reduced-order equivalent model for the current working condition; Based on the previously obtained latent dataset and its latent correlation data values, the relevant latent parameters are substituted into the high-fidelity reduced-order equivalent model for calculation. The specific process is as follows: Potential correlation feature vector Z k And the set of parameter sensitivity factors PSF={s1, s2, ..., s ∣Φ∣}; where s i Indicates the parameter ϕ i Local sensitivity to key performance y; generating multiple perturbation scenarios based on parameter sensitivity factors: for any model parameter ϕ i Let the reference disturbance amplitude Δ i base (e.g., ±5%), and scale it up or down according to sensitivity: Δ i =κ×∣s i ∣×Δ i base For parameter ϕ i The actual perturbation amplitude; where κ is the global scale factor; Δ i base The basic disturbance percentage, such as 1%, 5%, etc.; To generate a composite scene, define a scene vector p: ,in Choose the following construction strategy: Typical scenario: =sign(s i )×0.5Δi Worst-case scenario: Pressing |s i | Take the first K parameters in descending order, and make them =sign(s i )×Δ i Other parameters are 0; Progressive scenario: from 0 to Δ i Subdividing the interval into several steps yields a series of δ i The values ​​form an asymptotic trajectory.

[0031] Based on the disturbance amplitude, different disturbance combinations such as typical scenario, worst-case scenario, and asymptotic scenario are obtained, and multiple sets of disturbance parameter sets Φ are obtained. (p) ; For each disturbance scenario Φ (p) Perform the following detailed steps to quantify the impact of the potential data: Based on the reduced-order equivalent model, the disturbance scenario Φ (p) Perform detailed numerical simulations and analyses, including: The fourth-order Runge-Kutta method with fixed step size is used to calculate the dynamic response curve y of the key state variables under parameter perturbation. sim (p) (t); where key state variables include, but are not limited to, phase current and rotational speed.

[0032] Linearization is performed at the operating point to construct its open-loop and closed-loop transfer functions G(s), and frequency domain stability indices under this disturbance scenario are extracted, mainly including phase margin PM. (p) With cutoff frequency ω c (p) ; Record the parameter perturbation δ i (p) The propagation sequence and amplification relationship of the signal changes caused in various links of the control system (such as the current loop and the speed loop) form a parameter disturbance propagation diagram. This diagram is a directed graph structure, in which the nodes represent signals or performance indicators inside the system, the directed edges represent the influence path and direction, and the weight of the edge represents the gain and sign of the influence, clearly describing the causal chain structure from the disturbance source to the final performance indicator.

[0033] The simulation and analysis results under the disturbance scenario are y sim (p) The result y under the baseline scenario (without perturbation) sim (base) By comparison, the change Δy of a set of key performance indicators is calculated. i (p) : Time-domain indicators, including: adjustment time variation ΔT s (p)The overshoot change ΔOS is defined as the time required for the system response to reach and remain within the error band of ±2% of the steady-state value. (p) ΔOS (p) =OS (p) -OS (base) Where OS is the percentage by which the maximum output value exceeds the steady-state value during the response process; and the absolute change in steady-state error is |ΔE. ss (p) |:|ΔE ss (p) |=|E ss (p) -E ss (base) |, where E ss This represents the deviation between the expected value and the actual value after reaching steady state. Frequency domain metrics, including: phase margin attenuation ΔPM (p) ΔPM (p) =PM (base) -PM (p) (Unit: degrees); Cutoff frequency offset Δωc (p) Δωc (p) =ωc(p)−ωc(base) (unit: rad / s).

[0034] Stability index: Change in the real part of the dominant eigenvalue Δσ max (p) Calculate the dominant eigenvalue (the one with the largest real part) of the Jacobian matrix in the baseline and perturbed scenarios, and the change is Δσ. max (p) =σ max (p) −σσ max (base) A positive increase in this value indicates decreased stability.

[0035] The extracted changes in all key performance indicators {Δy1 (p) Δy2 (p) , …, Δy n (p) The features are combined into a feature vector and input into a pre-trained neural network model for influence analysis. NN ; The combined feature vectors are input into a pre-trained influence analysis neural network model; influence analysis neural network model f NN Preferably, a multilayer perceptron with two hidden layers is used, where the number of input layer nodes corresponds to the change in the key performance indicator Δy. iThe number of hidden layer nodes can be 64 or 32, respectively, using the ReLU activation function. The output layer has one node and uses the Sigmoid activation function. Training data is generated through extensive simulations based on a high-fidelity reduced-order equivalent model with various parameter perturbations. The true value I of the comprehensive influence degree is labeled for the changes in each group of key performance indicators according to preset rules. true The training process aims to minimize the difference between the predicted value I and the true value I. true The objective is to calculate the binary cross-entropy loss between the two variables, using the Adam optimizer. Its output is a dimensionless combined influence value I between 0 and 1. (p) :I (p) =f NN (Δy1 (p) Δy2 (p) , …, Δy n (p) );f NN To influence the analysis of neural networks; Δy i (p) Let I be the change in the i-th performance index; the comprehensive simulation influence value I (p) The closer it is to 1, the greater the impact of the corresponding potential parameter on the system; The adaptive compensation module obtains comprehensive impact values ​​and performs compensation analysis. The specific process is as follows: The current system risk level and response requirements are determined based on the comprehensive simulation impact value I, using the formula... The determination is made, where M0 is the final compensation mode, and θ1, θ2, and θ3 are preset thresholds; M none In the no-compensation mode, M soft M is in mild compensation mode. nmcdium For the moderate compensation mode, M hard This is a strong compensation model.

[0036] Based on the compensation pattern M0, select the corresponding compensation data type from the compensation type set: T={T R T K T G T F T S}; where T R For resistance-related parameter compensation types (such as resistance temperature drift compensation), T K For observer gain compensation type, T G For controller proportional / integral / feedforward gain compensation type, T F For filter coefficient compensation type, T S For stability boundary compensation types; based on the comprehensive impact values ​​and the main paths appearing in the disturbance propagation diagram, determine the compensation type corresponding to the main affected components; for example, when the disturbance mainly affects the current loop phase margin, select T. G or TS If the effect comes from the temperature drift parameter, then select T. R .

[0037] Furthermore, a mapping relationship between the compensation data type T and specific controller parameters is pre-stored. For example, T G The proportional gain K mapped to the controller p and / or integral gain K i ;T K The feedback gain matrix G mapped to the observer obs ;T F Mapped to the time constant or cutoff frequency of the filtering stage. Compensation data value C i Based on this mapping relationship applied to the corresponding parameters, the update rule can be expressed as follows: for gain-type parameters, a multiplicative correction Θ is used. new =Θ old +ΔΘ, for bias-type parameters, use additive correction Θ new =Θ old +ΔΘ. The corrected parameters need to be clipped.

[0038] After determining the compensation type, the actual compensation data value is further calculated; the compensation data value is based on the comprehensive impact value I. (p) Sensitivity factor s i And the target performance offset is used to construct an adjustable compensation amount, through formula C. i =γ×I (p) ×∣s i ∣×η i Among them, C i γ is the compensation data value; I is the global compensation gain coefficient; |s i | is the parameter ϕ i The absolute value of the sensitivity factor; η i These are the action coefficients corresponding to different compensation types (such as different scaling factors for proportional gain, integral gain, and observer gain).

[0039] The compensation data value is either positive or negative compensation, determined by the sign of the disturbance propagation path. For example, when the disturbance causes a degraded system performance and follows a positive amplification chain, C... i A positive value indicates a positive result; conversely, a negative value indicates a negative result.

[0040] Finally, the compensation data type T and the compensation data value C will be... i The output is sent to the controller parameter correction unit to form a set of controller parameters corrected in real time: using formula Θ new =Θ old +ΔΘ outputs the controller parameters updated in real time; where Θ old The original controller parameter set; ΔΘ is the dynamic correction amount determined by the compensation data value.

[0041] Furthermore, taking proportional gain compensation as an example: K p new =K p old ×(1+C i Output compensation proportional gain K p new ; Observer gain compensation as an example: G obs new =G obs old +C i Output compensated observer gain G obs new ; For example, filter compensation: α new =α old -C i Output compensated filter coefficient α new Among them, K p old G obs old and α old These are the original proportional gain, the original observer gain, and the original filter coefficients, respectively.

[0042] It should be noted that some of the key parameters mentioned above (such as mutual information contribution weight β, overall scale factor κ, global compensation gain coefficient γ, weight attenuation coefficient α, etc.) need to be tuned according to the specific requirements. As an initial reference for a typical embodiment: β is recommended to have an initial value of 0.5, κ is recommended to have an initial value of 1.0, γ is recommended to have an initial value of 0.1, and α is recommended to have a typical value of 0.1. In practical applications, fine-tuning can be performed through offline simulation or testing on a real platform, with performance indicators as the optimization target.

[0043] Please see Figure 2 As shown, another aspect of the present invention provides a brushless motor control method, the steps of which are as follows: Step S1: Collect real-time data from multiple sources during the operation of the brushless motor, including three-phase current, DC bus voltage, rotor position, speed, motor temperature, driver temperature, and current loop output, and send the data to the control and regulation analysis module.

[0044] Step S2: Perform time alignment, normalization, and window slicing operations on all real-time data. Calculate the instantaneous correlation tensor based on the mutual information-Euclidean distance hybrid correlation and construct a deep spatiotemporal graph model to extract potential correlation feature vectors. Then, determine the potential correlation of each potential feature according to a preset threshold set to obtain the potential correlation level and generate potential data.

[0045] Step S3: Based on the potential data and all real-time data, select a high-fidelity reduced-order equivalent model that matches the current operating state from the model library, inject the parameter perturbations corresponding to the potential data, and generate a composite perturbation scenario; run a hybrid domain-frequency domain simulation under each perturbation scenario to form a parameter perturbation propagation diagram, and extract key performance indicators such as settling time, overshoot, steady-state error, phase margin, cutoff frequency offset, and eigenvalue changes.

[0046] Step S4: Extract the changes in multidimensional performance indicators, use the trained impact assessment network to calculate the simulated impact value of each potential data point, and align and correct it with the observation deviations collected during actual operation to obtain the final comprehensive impact value.

[0047] Step S5: Determine the compensation level based on the comprehensive impact value, and calculate the compensation increment of the controller by combining the sensitivity factor of the potential data, the observer coupling degree and the stability decay index, and generate the corresponding compensation data type and compensation data value.

[0048] Step S6: The adaptive compensation module performs real-time online adjustments to the current loop, speed loop, or observer parameters based on the compensation increment, so that the motor control system can maintain stable and high-performance operation even in the presence of potential disturbances.

[0049] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A brushless motor control system, comprising a real-time detection module, a control and regulation analysis module, and an adaptive compensation module; characterized in that: The real-time detection module collects various real-time data of the brushless motor during operation, including electrical parameters, status parameters, control parameters, and physical parameters. The control and regulation analysis module receives real-time data and performs deep correlation analysis to obtain potential data correlation data values; it determines potential correlation based on the potential data correlation data values ​​to obtain potential data; it performs brushless motor operation simulation based on the potential data and the real-time data to obtain the impact of potential data and quantifies the simulated impact value of each potential data impact; and it corrects the simulated impact value to obtain the comprehensive impact value. The adaptive compensation module acquires the comprehensive impact value and performs compensation analysis to obtain the adaptive compensation data type and compensation data value, and then dynamically corrects the controller parameters based on the compensation data type and compensation data value.

2. The brushless motor control system according to claim 1, characterized in that, The process of receiving real-time data and performing deep correlation analysis to obtain potential data correlation values ​​is as follows: The real-time data is used to collect monitoring variables, and each variable is processed by timestamp alignment and normalization to obtain a normalized time series. The normalized time series is then sliced ​​according to a preset window length and step size to obtain each state slice. Based on each state slice, the nonlinear correlation between any two variables is calculated to form an instantaneous correlation tensor. A deep model is constructed based on the instantaneous correlation tensor and the state slice, and the potential correlation feature vector is output.

3. The brushless motor control system according to claim 2, characterized in that, The process of calculating the nonlinear correlation between any two sets of variables is as follows: For variables i and j within a slice, the mutual information is calculated using the mutual information formula. The weighted Euclidean distance between variables i and j is calculated using the weighted Euclidean distance formula. The mutual information and weighted Euclidean distance are fused using a hybrid correlation index formula. Calculate the mixed correlation of variables i and j in slice k. Where β is the mutual information contribution weight; This represents the maximum weighted Euclidean distance between all variable pairs in the k-th slice. To convert distance into similarity.

4. The brushless motor control system according to claim 2, characterized in that, The process of determining the potential correlation of data values ​​is as follows: Based on a preset set of potential relevance determination thresholds, each potential feature in the potential relevance feature vector is determined to obtain a potential relevance identifier, which is then summarized into a potential relevance identifier set. All judgment results in the potential relevance identifier set are weighted and fused to obtain the potential relevance level for each time slice; The potential data is generated based on the potential relevance feature vector, the potential relevance identifier set, and the potential relevance level.

5. A brushless motor control system according to claim 2, characterized in that, The factors affecting the numerical values ​​applied to the actual operation of a brushless motor are as follows: Based on real-time collected multi-source operating data, the model instance that best matches the current operating state is selected from the pre-set fully parameterized motor model library to obtain the optimal set of model parameters. Based on the potential data set and its potential correlation data values, the relevant potential parameters are substituted into the high-fidelity reduced-order equivalent model for calculation, generating multiple sets of perturbation parameter sets; For each disturbance scenario's disturbance parameter set, numerical simulation and analysis are performed to calculate the changes in key performance indicators; the changes in key performance indicators are then input into a pre-trained impact analysis neural network model to obtain comprehensive simulated impact values.

6. The brushless motor control system according to claim 1, characterized in that, The process of obtaining comprehensive impact values ​​and conducting compensation analysis is as follows: The current compensation mode is determined based on the comprehensive impact value. The compensation modes include no compensation, slight compensation, moderate compensation, and strong compensation. Based on the compensation mode, select the corresponding compensation data type from the predetermined set of compensation types; Based on the comprehensive impact value I (p) Sensitivity factor s of the parameter i and the corresponding action coefficient η for the compensation type i Through formula C i =γ×I (p) ×∣s i ∣×η i The calculated compensation data value C i , where γ is the global compensation gain coefficient; The compensation data type and the compensation data value are output to the controller parameter correction unit to generate a set of controller parameters after real-time correction.

7. A brushless motor control method, characterized in that... Implementing a brushless motor control system according to any one of claims 1-6, comprising: Step S1: Obtain real-time data such as electrical parameters, status parameters, and temperature of the brushless motor; Step S2: Perform deep correlation analysis on the real-time data to obtain potential data and potential correlation data values; Step S3: Build an operational model based on potential data and real-time data, perform disturbance simulation, obtain the impact of potential data, and quantify the simulated impact value; Step S4: Correct the simulated impact values ​​and the actual operational deviations to obtain the comprehensive impact values; Step S5: Generate compensation data types and compensation data values ​​based on the comprehensive impact values; Step S6: Perform real-time adaptive dynamic correction of the controller parameters based on the compensation data.

Images