Motor efficiency optimal search control method and system for variable working conditions

Abstract

The application relates to the technical field of motor control, and particularly discloses a motor energy efficiency optimal search control method and system for variable working conditions, which comprises the following steps: synchronously collecting motor multi-physical field original signals, performing multi-component frequency domain feature stripping through adaptive Kalman filtering, and outputting purified actual current, actual temperature and actual vibration effective amplitude; constructing a dynamic digital twin model of electromagnetic-thermal-mechanical multi-physical field coupling, inputting the purified actual current, the actual temperature and the actual vibration effective amplitude into the dynamic digital twin model, and outputting real-time dynamic electrical parameters and a subdivided health index. According to the scheme, the purified actual current and other characteristic signals are extracted through adaptive Kalman filtering, and the real-time dynamic electrical parameters and the subdivided health index are outputted in combination with the dynamic digital twin model of multi-physical field coupling, so that accurate data basis is provided for the prediction of the variable working condition trend.

Description

Technical Field

[0001] This invention relates to the field of motor control technology, specifically to a motor energy efficiency optimal search control method and system for variable operating conditions. Background Technology

[0002] Existing motor drive control systems typically employ offline parameter calibration combined with online optimization algorithms to improve energy efficiency when dealing with variable operating conditions characterized by frequent changes in load and speed. Traditional online optimization control often relies on fixed decision thresholds and a single time scale, continuously injecting current disturbance signals into the control system and observing changes in system energy consumption to search for the most efficient point under the current operating conditions.

[0003] This conventional continuous disturbance injection optimization scheme has specific application drawbacks in complex variable operating conditions. The continuously injected disturbance signal induces additional stator current harmonics and electromagnetic torque pulsations, increasing system transient losses and accelerating the aging of motor stator insulation components, demagnetization of permanent magnets, and wear of mechanical bearing components. Furthermore, existing optimization mechanisms typically only consider energy consumption reduction as an optimization metric, lacking quantitative assessment and constraints on the equipment lifespan costs caused by disturbances. This leads to the motor being prone to steady-state oscillations and over-adjustment during dynamic optimization, reducing the overall operational stability of the system and the equipment's service life. Summary of the Invention

[0004] This invention aims to at least partially solve one of the technical problems in related technologies. Therefore, the objective of this invention is to propose a motor energy efficiency optimal search control method and system for variable operating conditions, so as to improve the control convergence and overall life-cycle benefits of motors under variable operating conditions.

[0005] To achieve the above objectives, a first aspect of the present invention proposes a motor energy efficiency optimal search control method for variable operating conditions, comprising:

[0006] The original signals of the multi-physics field of the motor are acquired synchronously, and multi-component frequency domain features are stripped by adaptive Kalman filtering to output the purified actual current, actual temperature and actual effective vibration amplitude.

[0007] A dynamic digital twin model coupling electromagnetic-thermal-mechanical multiphysics fields is constructed. The purified actual current, actual temperature and actual vibration effective amplitude are input into the dynamic digital twin model, and the real-time dynamic electrical parameters and detailed health index are output.

[0008] Based on the real-time dynamic electrical parameters, the operating condition trend is predicted within a preset time window, and the estimated optimal current vector angle and the predicted steady-state duration are output.

[0009] The marginal lifetime cost and expected energy efficiency benefits of a single optimization are quantified in real time, the marginal cost-effectiveness ratio of optimization is calculated, the optimization granularity mode is autonomously decided based on the optimization marginal cost-effectiveness ratio, and the maximum allowable step size and convergence judgment threshold corresponding to the optimization granularity mode are output.

[0010] Based on the maximum allowable step size, a dual-time-scale hierarchical architecture is adopted to perform dual-objective optimization control for energy efficiency and lifespan: the upper search loop of the first control cycle performs variable step size iterative optimization based on gradient and curvature dual-drive mechanism, and the lower suppression loop of the second control cycle performs torque ripple and harmonic dynamic suppression locally, wherein the first control cycle is longer than the second control cycle.

[0011] A multi-dimensional adaptive threshold is used to determine the extreme value neighborhood. When the convergence determination threshold is met, steady-state zero disturbance locking is triggered, the injection of optimization disturbance is stopped, and a search control closed loop is formed.

[0012] To achieve the above objectives, a second aspect of the present invention provides a motor energy efficiency optimal search and control system for variable operating conditions, comprising:

[0013] The multi-physics sensing and adaptive Kalman filter purification unit is used to synchronously acquire the original multi-physics signals of the motor, perform multi-component frequency domain feature stripping through adaptive Kalman filtering, and output the purified actual current, actual temperature and actual vibration effective amplitude.

[0014] A multi-physics dynamic digital twin unit is used to construct a dynamic digital twin model of electromagnetic-thermal-mechanical multi-physics coupling. The purified actual current, actual temperature and actual vibration effective amplitude are input into the dynamic digital twin model, and the real-time dynamic electrical parameters and detailed health index are output.

[0015] The operating condition trend prediction unit is used to predict the operating condition trend within a preset time window based on the real-time dynamic electrical parameters, and output the estimated optimal current vector angle and the predicted steady-state duration.

[0016] The marginal cost-effectiveness ratio decision unit is used to quantify the marginal lifetime cost and expected energy efficiency benefits of a single optimization in real time, calculate the optimization marginal cost-effectiveness ratio, autonomously decide the optimization granularity mode based on the optimization marginal cost-effectiveness ratio, and output the maximum allowable step size and convergence judgment threshold corresponding to the optimization granularity mode.

[0017] A dual-timescale optimization control unit is used to perform dual-objective optimization control of energy efficiency and lifespan based on the maximum allowable step size and using a dual-timescale hierarchical architecture: the upper search loop of the first control cycle performs variable step size iterative optimization based on a gradient and curvature dual-drive mechanism, and the lower suppression loop of the second control cycle locally performs torque ripple and harmonic dynamic suppression, wherein the first control cycle is longer than the second control cycle.

[0018] The steady-state locking unit is used to determine the extreme value neighborhood using a multi-dimensional adaptive threshold. When the convergence determination threshold is met, steady-state zero-disturbance locking is triggered, the injection of optimization disturbances is stopped, and a search control closed loop is formed.

[0019] To achieve the above objectives, a third aspect of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory. When the computer program is executed by the processor, it implements the above-described motor energy efficiency optimal search control method for variable operating conditions.

[0020] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0021] The present invention extracts and purifies characteristic signals such as actual current through adaptive Kalman filtering, and outputs real-time dynamic electrical parameters and detailed health indices in conjunction with a multi-physics coupled dynamic digital twin model, providing an accurate data foundation for predicting the trend of changing operating conditions. In actual optimization operation, the system quantifies the expected energy efficiency benefits and marginal lifetime costs of a single optimization in real time, autonomously decides the optimization step size based on the marginal cost-effectiveness ratio, and adopts a dual-time-scale hierarchical architecture. It performs variable step-size iterative optimization in a longer control cycle and dynamically suppresses torque ripple and harmonics in a shorter control cycle. Finally, it triggers steady-state locking to stop disturbances when the convergence condition is met through multi-dimensional adaptive thresholds.

[0022] The above scheme quantitatively balances energy efficiency optimization with motor life loss, avoids continuous physical damage to motor insulation, permanent magnets and mechanical components caused by ineffective disturbance injection, effectively suppresses torque pulsation during the optimization process, and improves the control convergence and overall life-cycle benefits of the motor under varying operating conditions. Attached Figure Description

[0023] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals are used to refer to the same parts. Wherein:

[0024] Figure 1 This is a flowchart illustrating the optimal energy efficiency search control method for motors under varying operating conditions provided by the present invention.

[0025] Figure 2 This is a schematic diagram illustrating the implementation of the optimal energy efficiency search and control system for motors under varying operating conditions provided by this invention.

[0026] Figure 3 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0027] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0028] The following description, with reference to the accompanying drawings, outlines an embodiment of the optimal energy efficiency search control method, system, and electronic device for motors operating under varying conditions.

[0029] Example 1:

[0030] Figure 1 This is a flowchart illustrating an embodiment of the optimal energy efficiency search control method for motors under varying operating conditions according to the present invention.

[0031] Specifically, this embodiment provides a motor energy efficiency optimal search control method for variable operating conditions. This method, through a hardware-software co-working architecture, achieves comprehensive control of high energy efficiency and long lifespan during motor drive based on multi-physics data fusion. The control method constructs a complete search control closed loop through fine-grained management of the control cycle, disturbance step size, and multiple objective functions. The method includes the following steps:

[0032] The original multi-physics field signals of the motor are acquired synchronously, and multi-component frequency domain features are stripped using adaptive Kalman filtering to output the purified actual current. The actual temperature and actual effective vibration amplitude; constructing a dynamic digital twin model of electromagnetic-thermal-mechanical multiphysics coupling, and refining the actual current. The actual temperature and the actual effective vibration amplitude are input into the dynamic digital twin model, which outputs real-time dynamic electrical parameters and detailed health indices.

[0033] Based on the real-time dynamic electrical parameters, the operating condition trend is predicted within a preset time window, and the estimated optimal current vector angle and the predicted steady-state duration are output. The marginal lifetime cost and expected energy efficiency benefits of a single optimization are quantified in real time, the optimization marginal cost-effectiveness ratio is calculated, and the optimization granularity mode is autonomously decided according to the optimization marginal cost-effectiveness ratio. The maximum allowable step size and convergence judgment threshold corresponding to the optimization granularity mode are output.

[0034] Based on the maximum allowable step size, a dual-time-scale hierarchical architecture is adopted to perform dual-objective optimization control for energy efficiency and lifespan: the upper search loop of the first control cycle performs variable step-size iterative optimization based on a gradient and curvature dual-drive mechanism, and the lower suppression loop of the second control cycle performs torque ripple and harmonic dynamic suppression locally, wherein the first control cycle is longer than the second control cycle; a multi-dimensional adaptive threshold is used to determine the extreme value neighborhood, and when the convergence determination threshold is met, steady-state zero disturbance locking is triggered, the injection of optimization disturbances is stopped, and a search control closed loop is formed.

[0035] For example, the process of removing multi-component frequency domain features by adaptive Kalman filtering for the original multi-physics field signals of the synchronously acquired motor described above specifically includes:

[0036] Simultaneously collect multiphase stator current, DC bus voltage, rotor position, mechanical speed, stator winding temperature, bearing vibration acceleration, and permanent magnet temperature.

[0037] In practical industrial applications, multiphase stator current is typically acquired using high-precision Hall current sensors located at the inverter output. The sampling frequency is limited by the underlying hardware's analog-to-digital converter configuration and is usually set to the megahertz level to capture high-frequency harmonics. Stator winding temperature and permanent magnet temperature can be obtained using thermistors, such as the PT100 type, embedded inside the stator slots and on the rotor surface. Bearing vibration acceleration is acquired using piezoelectric accelerometers mounted on the outside of the motor end cover. A coordinate transformation is then performed on the multiphase stator current to obtain the real-time current in an orthogonal stationary coordinate system. vector.

[0038] This coordinate transformation process primarily utilizes the Clark transformation matrix to project the time-varying multiphase asymmetric AC current onto a two-phase orthogonal stationary coordinate plane. This decouples the complex coupled system into mutually perpendicular components, providing a standardized input reference for subsequent filtering and observation. It is important to clarify here that the real-time current... It represents the transient raw sample value, which includes stray components caused by low-level high-frequency switching noise, electromagnetic interference, and mechanical vibration. Its value fluctuates dramatically.

[0039] Optionally, in order to extract useful state information from the aforementioned mixed noise, the system is constructed to include the actual current. A discrete state-space model is constructed for the components of the motor's internal physical quantities, including the optimization disturbance current component, the actual winding temperature component, and the effective vibration amplitude component. This discrete state-space model defines the dynamic evolution of each physical quantity within the motor in the time domain.

[0040] Furthermore, the adaptive Kalman filter performs filtering iterations, adjusting the amplitude of the process noise covariance matrix in real time based on the transient and steady-state operating conditions determined by the current change rate, and outputting the actual current after removing the optimization disturbance current component. The actual temperature and the actual effective amplitude of vibration.

[0041] In the process of performing filtering iterations in the above adaptive Kalman filter, its core discrete state-space model and observation update equation are expressed as follows:

[0042] ;

[0043] ;

[0044] In the formula, Indicates the first The system state vector at each discrete time step contains the actual current. Equal portions; Indicates the first A system state vector with a discrete time step; This represents the system state transition matrix, used to describe the inherent decay and evolution characteristics of physical quantities over time; Represents the system control input matrix; Indicates the first A control input vector with a discrete time step; Represents the process noise vector; Indicates the first Each discrete-time step measurement observation vector contains the real-time current. Original sensor sample values; This represents the observation matrix, used to establish a linear mapping between state variables and measurement variables; This represents the measurement noise vector.

[0045] It should also be noted that when applying the above formula for iteration, the system will monitor the real-time current. The time-varying rate of change is calculated. When the absolute value of the current change rate exceeds a set threshold, the system is determined to be in a transient state, at which point the algorithm amplifies the process noise vector. The corresponding covariance matrix magnitude. The engineering significance of this action is to reduce the model's confidence in historical prior estimates, improve the filter's tracking bandwidth for abrupt signals, and prevent severe phase lag during rapid acceleration and deceleration.

[0046] Conversely, when the absolute value of the current change rate is less than or equal to the set change rate threshold, the system is determined to be in steady-state operation. In this case, the algorithm will reduce the amplitude of the process noise covariance matrix. This enhances the filtering depth and further suppresses measurement noise during steady-state operation. After multiple rounds of prediction and update corrections, the system finally outputs a smooth and accurate actual current. This variable filters out high-frequency glitches and artificially injected optimization disturbances, and can truly reflect the core electromagnetic work capacity of the motor body under the current load.

[0047] Specifically, the steps for constructing a dynamic digital twin model of electromagnetic-thermal-mechanical multi-physics coupling and outputting real-time dynamic electrical parameters and detailed health indices include, for example: acquiring pre-stored offline motor design parameters and constructing a multi-physics coupling digital twin base, and using a dynamic time warping algorithm to refine the actual current. The actual temperature and actual effective vibration amplitude are aligned with the timestamp of the multiphysics coupled digital twin base. This is because electrical signals in the motor control system, such as actual current... The changes are in the millisecond or even microsecond range, while the changes in thermodynamic signals (such as actual temperature) and mechanical aging characteristics (such as actual vibration effective amplitude) are often in the second or even hour range.

[0048] Direct fusion of data from multiple time scales can lead to ill-conditioned matrix computation. Therefore, the dynamic time warping algorithm constructs a time warping cumulative distance matrix to nonlinearly stretch and map the low-frequency acquisition sequence while maintaining the chronological order of temporal features. This ensures that electrical, thermal, and mechanical three-dimensional data share a unified timestamp reference within the same computation cycle.

[0049] Optionally, the system combines extended Kalman filtering for multiphysics state estimation, outputting the real-time dynamic electrical parameters, which include dynamic stator resistance, dynamic direct-axis inductance, dynamic quadrature-axis inductance, and dynamic permanent magnet flux linkage. Extended Kalman filtering is performed by expanding the partial derivatives of the nonlinear motor flux linkage and voltage equations using the actual real-time current. Based on the bus voltage input, the motor parameters affected by temperature rise and magnetic circuit saturation effect are estimated online in a rolling manner.

[0050] Based on this, the system calculates the estimation error of the multiphysics state estimation, and outputs the subdivided health index based on the estimation error. The subdivided health index includes the stator insulation health index, the permanent magnet health index, and the mechanical transmission system health index. The comprehensive health index of the motor is obtained by weighted summation of the subdivided health indices.

[0051] For example, when the estimated dynamic stator resistance exceeds its nominal value at room temperature by an abnormally large proportion, it indicates that the stator insulation layer may be aging and degrading due to long-term thermal stress, thus lowering the insulation health index. When the estimated dynamic permanent magnet flux exhibits an irreversible amplitude drop, it indicates local demagnetization of the permanent magnet, thus lowering the permanent magnet health index. When the energy integral of the actual vibration effective amplitude within a specific characteristic frequency range exceeds the baseline safety threshold, it indicates that the bearing raceway is peeling or the clearance is increasing, thus lowering the mechanical transmission system health index. These three sub-indices are linearly summed using their respective preset lifespan importance weighting coefficients to form the global motor comprehensive health index required for closed-loop control.

[0052] For example, for predicting the operating condition trend within a preset time window based on the real-time dynamic electrical parameters, and outputting the estimated optimal current vector angle and the predicted steady-state duration, specifically, it includes: using the time series data of speed and torque within a historical preset time period as input, and using a time series prediction neural network model and a feature regression model to generate the speed and torque prediction curves within the future preset time window.

[0053] In practical embedded controller implementations, the historical preset time period can be configured to be the past 30 seconds. The controller caches discrete points of rotational speed and torque at a fixed downsampling frequency, forming a feature tensor input to a time-series prediction neural network model, such as a long short-term memory network with forget gates, input gates, and output gates. This model can uncover periodic load fluctuation patterns or trending acceleration command intentions in historical operating conditions. The feature regression model is responsible for calibrating and smoothing the nonlinear boundaries of the neural network output, ultimately piecing together a future preset time window, such as a continuous prediction curve for the next 10 seconds.

[0054] It is also important to note that the system identifies the steady-state operating condition intervals in the speed and torque prediction curves and outputs the predicted steady-state duration. The system calculates the variance or gradient of the prediction curve within a sliding time window. When this gradient is continuously lower than a set operating condition switching threshold, it is marked as an upcoming steady-state operating condition interval, and the time span of this interval is accumulated as the predicted steady-state duration.

[0055] Furthermore, by combining the real-time dynamic electrical parameters with the predicted operating conditions within the steady-state operating range, the estimated optimal current vector angle is obtained by matching from the dynamic maximum torque-current ratio database. This database is typically a two-dimensional or three-dimensional interpolation table generated through large-scale offline calibration during bench testing. It provides an initial, efficient operating point as an anchor point for subsequent online fine-tuning, effectively shortening the convergence distance of closed-loop control.

[0056] Specifically, for real-time quantification of the marginal lifetime cost of a single optimization, examples include: obtaining a preset system reference maximum step size and motor calibration parameters, and calculating the effective value of current ripple and torque pulsation amplitude caused by the optimization based on the system reference maximum step size and the motor calibration parameters. The system reference maximum step size is the theoretical maximum disturbance injection amount without considering any constraints, and the motor calibration parameters include the number of pole pairs, moment of inertia, and electromagnetic torque coefficient of the motor. Using these parameters, it is possible to analytically calculate how much current ripple will be generated in the stator winding and how much torque pulsation will be generated at the rotor shaft end if the optimization signal is injected according to the reference step size.

[0057] Optionally, the estimated optimization convergence time is obtained; three types of optimization marginal lifetime costs are calculated respectively, including: calculating the insulation thermal aging marginal cost based on the current ripple RMS value, the estimated optimization convergence time, and the stator insulation health index; calculating the permanent magnet demagnetization marginal cost based on the current ripple RMS value, the estimated optimization convergence time, and the permanent magnet health index; calculating the mechanical wear marginal cost based on the torque pulsation amplitude, the estimated optimization convergence time, and the mechanical transmission system health index; and weighted summing the insulation thermal aging marginal cost, the permanent magnet demagnetization marginal cost, and the mechanical wear marginal cost to obtain the total marginal lifetime cost.

[0058] The quantitative weighting process for the above three types of marginal costs is calculated using the following formula 2:

[0059] ;

[0060] In the formula, Represents the total marginal life cost; This represents the weighting factor for the importance of stator insulation; This represents the marginal cost of thermal aging of insulation. This represents the weighting coefficient for the importance of permanent magnets; This represents the marginal cost of permanent magnet demagnetization; Indicates the weighting coefficient for the importance of mechanical transmission; This represents the marginal cost of mechanical wear and tear.

[0061] The core technical logic of this formula lies in the fact that, since finding the optimal energy efficiency point requires injecting disturbances into the motor, these additional current ripples increase copper and iron losses, leading to a greater temperature rise in the windings and permanent magnets; additional torque pulsations are applied to the bearing surface, accelerating fatigue wear. Furthermore, when the motor's current sub-health index is low, i.e., in the middle to late stage of the equipment's lifespan, its ability to withstand these disturbances decreases exponentially, and the calculated marginal costs will increase significantly at this time.

[0062] For example, the process of calculating the optimal marginal cost-effectiveness ratio for expected energy efficiency gains specifically includes: based on the actual current... Based on the real-time dynamic electrical parameters, calculate the total system energy efficiency loss at the current operating point of the motor. This loss typically includes the copper loss of the stator windings (proportional to the square of the actual current and the dynamic stator resistance), the iron loss of the stator core (eddy current and hysteresis losses), and the conduction and switching losses of the inverter switching transistors. Based on the total system energy efficiency loss, calculate the reduction in energy efficiency loss per unit time between the current initial operating point and the estimated optimal point. Obtain the preset average transient loss increment during the optimization process; calculate the total expected energy efficiency gain for a single optimization by dividing the reduction in energy efficiency loss per unit time by the difference in stable operating time, where the stable operating time difference... The predicted steady-state duration Subtract the estimated convergence time for optimization .

[0063] Furthermore, dividing the total expected energy efficiency benefit by the total marginal lifetime cost yields the optimal marginal cost-effectiveness ratio, the relationship of which can be expressed by the following formula:

[0064] ;

[0065] In the formula, This indicates the marginal utility-cost ratio for finding the optimal solution. This indicates the reduction in energy loss per unit time. Indicates the time difference of stable operation; This represents the average transient loss increment during the optimization process; This indicates the estimated convergence time for optimization. This represents the total marginal lifetime cost. This formula not only considers the energy-saving benefits brought about by long-term operation after successful optimization, but also rigorously deducts the additional energy consumption cost incurred due to the injection of disturbances during the optimization process. Finally, it compares this cost with the lifetime cost to form a core decision-making indicator to measure whether the optimization action is "worthwhile".

[0066] Specifically, based on the optimization marginal cost-effectiveness ratio, the system autonomously decides the optimization granularity mode and outputs the maximum allowable step size and convergence judgment threshold corresponding to the optimization granularity mode. This includes setting a high-efficiency cost-effectiveness ratio threshold, a medium-efficiency cost-effectiveness ratio threshold, and a low-efficiency cost-effectiveness ratio threshold, with the values ​​of the three decreasing sequentially. When the optimization marginal cost-effectiveness ratio is greater than the high-efficiency cost-effectiveness ratio threshold, it is determined to be in full-scale high-precision mode, and the maximum allowable step size equal to the system's baseline maximum step size, as well as the corresponding convergence judgment threshold and pulsation threshold, are output. This means that the energy-saving benefits brought by optimization far outweigh the resulting equipment losses, and the system allows the maximum search step size to quickly approach the optimal point.

[0067] When the marginal cost-effectiveness ratio of the optimization is between the medium cost-effectiveness ratio threshold and the high cost-effectiveness ratio threshold, it is determined to be a standard convergence mode. The maximum allowable step size is reduced by a preset first ratio, as well as the corresponding convergence determination threshold and pulsation threshold, wherein the first ratio is less than 1.

[0068] When the marginal cost-effectiveness ratio of the optimization is between the low cost-effectiveness ratio threshold and the medium cost-effectiveness ratio threshold, it is determined to be a fast approximation mode, and the maximum allowable step size of the system benchmark maximum step size is amplified by a preset second ratio, as well as the corresponding convergence determination threshold and pulsation threshold, wherein the second ratio is greater than 1.

[0069] When the marginal cost-effectiveness ratio of optimization is less than or equal to the inefficiency-cost ratio threshold, it is determined that the optimization mode should be abandoned, the current working point should be maintained and no optimization disturbance should be injected.

[0070] This mode effectively avoids the risk of irreversible physical damage to aging motors under short-term steady-state conditions in order to save a small amount of energy, demonstrating the reliability constraints of the control strategy.

[0071] Optionally, for the upper-level search loop of the first control cycle, a variable-step-size iterative optimization process is performed based on a gradient and curvature dual-drive mechanism. Specifically, this includes: constructing a dual-objective overall optimization function for energy efficiency and lifespan, the value of which is a weighted sum of the equivalent energy efficiency loss calculated based on the total system energy efficiency loss and the equivalent lifespan loss calculated based on the motor's comprehensive health index, weighted according to preset weights; and calculating the first-order gradient and second-order curvature of the dual-objective overall optimization function with respect to the current vector angle command. The system typically uses the central difference method or backward difference equations to obtain the gradient and curvature in the discrete-time domain. The gradient indicates the steepest descent direction of the objective function, while the curvature reflects the gentleness or steepness of the spatial region where the objective function resides.

[0072] Based on the maximum allowable step size, the first-order gradient, the second-order curvature, and the health constraint factor, the current optimization step size amplitude is dynamically updated through absolute value scalar operations, wherein the value of the health constraint factor decreases as the comprehensive health index of the motor decreases; the inverse sign component of the first-order gradient is extracted, and the inverse sign component is multiplied by the current optimization step size amplitude to update the current vector angle command and send it to the lower-level suppression loop of the second control cycle.

[0073] The update rule for the current optimization step size is defined by the following formula:

[0074] ;

[0075] In the formula, Indicates the current optimization step size; Indicates the maximum allowed step size; Represents the hyperbolic tangent function; This represents the first-order gradient adjustment coefficient; Indicates the first-order gradient; Represents an exponential function with the natural constant as its base; This represents the second-order curvature adjustment coefficient; Indicates second-order curvature; This indicates a small constant bias to prevent the denominator from being zero; This represents a health constraint factor.

[0076] According to this formula, when the first-order gradient is large, the hyperbolic tangent function approaches 1, ensuring sufficient step size when far from the optimum. When the second-order curvature increases significantly compared to the first-order gradient, i.e., near the valley region of the extreme point, the exponential function term decays rapidly, forcing the current optimization step size to adaptively shrink, effectively suppressing steady-state high-frequency oscillations near the optimum. At the same time, the health constraint factor serves as a global scaling factor, ensuring that the more severe the motor aging, the more conservative the exploration action.

[0077] It should also be noted that the lower-level suppression loop of the second control cycle locally performs torque ripple and harmonic dynamic suppression, specifically including: real-time detection of the total harmonic distortion rate of the stator current and the amplitude of electromagnetic torque ripple in the lower-level current loop controller; when the total harmonic distortion rate of the stator current or the amplitude of electromagnetic torque ripple exceeds the ripple threshold corresponding to the current optimization granularity mode, a feedforward compensation current command is generated to offset the transient ripple caused by optimization, and the steady-state operating point output by the upper-level search loop remains unchanged.

[0078] Because the second control cycle (e.g., 100 microseconds) is significantly shorter than the first control cycle (e.g., 10 milliseconds), the lower-level suppression loop has extremely high time response resolution. It extracts the disturbance harmonics (e.g., the 5th or 7th harmonics) caused by optimization by connecting a harmonic resonator of a specific frequency in parallel within a synchronously rotating coordinate system, and injects compensation voltages with opposite phases and equal amplitudes. This reduces electromagnetic noise and torque tremors at the microscopic level without affecting the macroscopic energy efficiency convergence path of the upper-level optimization.

[0079] For example, a multi-dimensional adaptive threshold is used to determine the extreme value neighborhood. When the convergence determination threshold is met, steady-state zero-disturbance locking is triggered. This includes: using a first-order gradient threshold, a second-order curvature threshold, and an objective function volatility threshold to form the multi-dimensional adaptive threshold as the convergence determination threshold. The multi-dimensional adaptive threshold is dynamically and adaptively scaled according to the current optimization granularity mode. When, within a consecutive preset number of optimization cycles, the first-order gradient, the second-order curvature, and the volatility of the bi-objective total optimization function all simultaneously meet their corresponding multi-dimensional adaptive thresholds, the working point is determined to enter the preset extreme value optimal neighborhood. Steady-state zero-disturbance locking is triggered, and the optimization step size is controlled to smoothly decay to zero according to an exponential law. In the locked state, the motor parameter drift is monitored in real time through the dynamic digital twin model to evaluate the conditions for triggering optimization restart.

[0080] The volatility of the objective function is typically in a length of The statistical variance calculated within the sliding data window is used to characterize the degree of abrupt changes in the system's energy dissipation. Only when the gradient is close to zero, the curvature meets the convexity requirement, and the system's own fluctuations are minimal, and this state is maintained for several cycles, is the system confirmed that the current operating point has entered the optimal neighborhood of the extreme value. Once locked, no further disturbances are injected into the system, and the motor enters a stable and efficient operation phase without external disturbances.

[0081] Based on existing technology analysis, traditional online efficiency optimization methods mostly rely on single-dimensional disturbance observation. This means that to obtain the control gradient, sinusoidal or step disturbances must be continuously injected into the motor system, causing the motor to oscillate continuously throughout its steady-state lifespan. In specific applications such as electric vehicles or precision industrial control, this traditional unconstrained optimization not only offsets some energy-saving benefits due to additional harmonic copper losses, but the accompanying torque pulsations also significantly accelerate the fatigue wear of transmission bearings, shortening the overall lifespan of the machine and leading to a decline in the overall economic efficiency of the system.

[0082] The method disclosed in this embodiment pre-estimates optimization potential by constructing a digital twin base and innovatively introduces the "marginal cost-effectiveness ratio" into the control decision layer, enabling the control system to have a mechanism for comprehensively evaluating energy efficiency gains and equipment lifespan losses. Under the synergy of a dual-timescale architecture, this solution can dynamically manage the convergence step size based on the motor's health status at a macroscopic level to ensure lifespan boundaries, and dynamically filter parasitic vibrations caused by optimization at a microscopic level, implementing zero-disturbance locking after confirming the arrival of the optimal point. Overall, this technical solution effectively overcomes the irreconcilable contradiction between optimization accuracy and physical losses in existing technologies, significantly improving the operational stability and overall mechanical reliability of equipment under varying operating conditions throughout its entire lifecycle while ensuring high-efficiency electromagnetic energy conversion.

[0083] Example 2:

[0084] In multi-motor cooperative drive systems, such as four-wheel independent drive electric vehicles, multi-joint industrial robotic arms, or distributed rail transit traction systems, the individual motors do not operate in isolation, but rather interact strongly through physical structures or electrical networks.

[0085] To manage this interaction, the multi-motor cooperative scheduling step specifically includes: constructing a directed graph topology model for the multi-motor cooperative drive system, where graph nodes represent individual motor operating units, graph edges represent electrical and mechanical coupling relationships between individual motor operating units, and the weight values ​​of the graph edges characterize the coupling strength of the electrical and mechanical coupling relationships.

[0086] In practical engineering applications, constructing the above-mentioned directed graph topology model is a prerequisite for achieving system-level decoupled scheduling. The graph nodes correspond to independent motor drive modules in physical space, typically including an inverter, the motor itself, and a matching local microcontroller. The graph edges represent the implicit physical interconnections between these nodes.

[0087] The electrical coupling relationship usually originates from multiple single motor operating units connected in parallel on the same DC bus. When a motor performs optimization operation and generates high-frequency current ripple, the ripple will cause the charging and discharging fluctuations of the DC bus capacitor, thereby affecting the input voltage stability of other motors connected in parallel on the bus.

[0088] The mechanical coupling relationship arises from multiple motors acting on the same rigid or semi-rigid load object. For example, the left and right wheels of an electric vehicle are connected to the chassis frame via the chassis suspension, or multiple joint motors of a robotic arm are connected by mechanical linkages. The torque pulsation of one motor will be transmitted to the shaft ends of other motors in the form of elastic waves through the mechanical structure, forming mechanical crosstalk.

[0089] To accurately quantify the potential harm of this crosstalk, the system constructs a data matrix reflecting the aforementioned interconnections within the embedded central controller. It then uses online system identification algorithms, such as recursive least squares with a forgetting factor, to estimate the equivalent impedance of the bus and the equivalent stiffness of the mechanical structure in real time, thereby generating weight values ​​to characterize the coupling strength. The larger the weight value, the more severe the mutual influence between the two individual motor operating units, and the more likely the spillover of optimization disturbances will occur.

[0090] Optionally, in order to find the optimal optimization execution order and strategy in a complex coupled network, the system extracts the current subdivided health index and the optimization marginal cost ratio of each single motor operating unit as node features input to the graph neural network model, and calculates the topological centrality and disturbance crosstalk sensitivity of each graph node through the graph neural network model.

[0091] In the above steps of calculation using a graph neural network model, the system leverages the advantages of graph neural networks in processing non-Euclidean space topological data. The system concatenates and compresses the stator insulation health index, permanent magnet health index, mechanical transmission system health index, and optimization marginal cost-effectiveness ratio output in Example 1 to form the initial feature vector for each graph node. The graph neural network model aggregates the state information of adjacent nodes through a hierarchical information transmission mechanism. Its core graph convolutional state update algorithm is expressed by the following formula:

[0092] ;

[0093] In the formula, This represents the node feature update matrix of a single layer output of a graph neural network model. This represents a non-linear activation function, configured in this specific application as a linear rectified function with leakage correction to avoid gradient vanishing; This represents the graph edge weight matrix containing self-loops after Laplace normalization, whose internal elements are composed of the coupling strength of the aforementioned electrical and mechanical coupling relationships. This represents the input node feature matrix, which includes the detailed health index and optimization marginal cost ratio of each individual motor operating unit; This represents the learnable parameter matrix of the current hidden layer in the graph neural network model.

[0094] It is also important to note that through multiple forward propagation calculations in the above formula, the graph neural network model can output a dimensionality-reduced high-order feature representation, which is ultimately mapped to two key system-level metrics: topological centrality and disturbance crosstalk sensitivity. Topological centrality reflects the "hub" status of a particular motor in the entire physical network.

[0095] For example, in a multi-axis drive system, the motor located at the input end of the main drive shaft has extremely high topological centrality. Its minute torque ripples are amplified by the subsequent multi-stage gear network and transmitted to all driven motors. Disturbance crosstalk sensitivity reflects the probability that the optimization disturbance signal injected by this motor, under the current specific health condition, will cause global system resonance or bus voltage drop. By training a graph neural network model, the system adaptively increases the output value of the disturbance crosstalk sensitivity for motor nodes that frequently cause errors or current overshoots in other nodes in the historical operation logs. Through this deep learning-based graph topology analysis, the controller can grasp the disturbance propagation characteristics of the entire system at a macroscopic level, laying a data foundation for subsequent computational resource allocation.

[0096] For example, after obtaining the topological characteristics of the graph nodes, the system integrates the optimization marginal cost-effectiveness ratio, the topological centrality, and the disturbance crosstalk sensitivity to calculate and output the node comprehensive priority score of each single motor operating unit, and allocates optimization computing resources to the single motor operating unit in descending order of the node comprehensive priority score.

[0097] In the operating environment of an embedded multi-core real-time operating system, the computing power of the central control unit (such as the operation cycle of a digital signal processor or field-programmable gate array) is limited and cannot support all motors to execute extremely computationally intensive dual-time-scale optimization and high-frequency compensation logic at the same time. Therefore, a strict priority allocation strategy must be formulated. The process of comprehensive priority scoring of system computing nodes is defined by the following formula 2:

[0098] ;

[0099] In the formula, This represents the overall priority score of a single motor operating unit. This indicates the positive incentive weight for the cost-effectiveness ratio; This represents the normalized optimization marginal cost-effectiveness node feature generated by the optimization marginal cost-effectiveness ratio mapping; Indicates the centrality penalty weight; This represents the topological centrality output by the graph neural network model; Indicates the sensitivity penalty weight; This represents the perturbation crosstalk sensitivity output by the graph neural network model.

[0100] The technical logic above shows that the system encourages those with high optimization potential and good health (i.e., While prioritizing energy efficiency optimization for motors with high energy values, we also optimize the energy efficiency of motors located at topological critical nodes. High values) or motors that are prone to causing global instability ( Higher numerical values ​​are assigned larger penalty weights. This design has important guiding significance in engineering practice.

[0101] For example, although a certain edge-driven motor may have limited energy-saving potential, its position at the end of the mechanical network means that disturbances to it will have little impact on other components. Therefore, its priority score may actually be higher than that of a motor located at the main drive hub. Based on the calculated overall priority scores of the nodes, the central controller, through the underlying task scheduler (such as priority-based preemptive scheduling logic), prioritizes allocating computing cores and communication bandwidth to the motor with the highest score, allowing it to initiate the variable-step-size iterative optimization algorithm disclosed in Example 1. Motors with lower scores are placed in a waiting queue to maintain their current operating point, thus achieving an optimal match between system-level computing resources and expected benefits.

[0102] Specifically, after determining the optimization eligibility of each motor, it is also necessary to address the physical interference problem during optimization. Based on the coupling strength represented by the weight values ​​and the disturbance crosstalk sensitivity, the system executes an asynchronous orthogonal disturbance injection strategy for multiple individual motor operating units. Specifically, the asynchronous orthogonal disturbance injection strategy involves: when the coupling strength between coupled individual motor operating units exceeds a preset decoupling safety threshold, allocating non-overlapping disturbance injection time windows to the coupled individual motor operating units in the time dimension, or allocating disturbance injection frequencies with orthogonal characteristics to them in the frequency dimension, in order to eliminate interference crosstalk conflicts caused by multiple individual motor operating units simultaneously performing optimization control.

[0103] The preset decoupling safety threshold is a safety boundary index generated by the system based on the current bus capacitance margin and the torsional stiffness of the mechanical chassis, calibrated offline and then fine-tuned online. When the coupling strength between the two motors is lower than this threshold, it indicates that they are in a weakly coupled state in physical space, allowing them to perform optimization simultaneously and freely.

[0104] However, once the system determines that the coupling strength between the two is greater than the decoupling safety threshold, such as the left drive motor and the right drive motor on the same axle, which form a strong rigid coupling through the mechanical half shaft and anti-roll bar, the asynchronous orthogonal disturbance injection strategy will be forcibly intervened.

[0105] On one hand, the system employs a time-dimensional decoupling scheme, allocating non-overlapping disturbance injection time windows. The central controller acts as a token ring manager in the time domain, dividing the control cycle into multiple logical time slots. For example, in the first 100-millisecond time slot, only motor A is allowed to inject optimization disturbances. At this time, the lower-level suppression loop of motor B is forcibly activated and blocks the instructions of its upper-level search loop. Motor B is specifically responsible for maintaining the current steady state and absorbing slight mechanical transmission fluctuations from motor A. After motor A completes the extreme value neighborhood determination and triggers steady-state zero-disturbance locking, the system passes the optimization token to motor B in the next time slot. This time-slice-based time-division optimization strategy effectively avoids the risk of superposition resonance of multiple disturbance sources occurring at the same time point from a physical perspective, effectively protecting the mechanical transmission chain from damage caused by low-frequency beat oscillations.

[0106] On the other hand, for coupled motors that must respond to changes in operating conditions simultaneously and cannot accept time delays, the system starts a decoupling scheme in the frequency dimension, namely, allocating a disturbance injection frequency with orthogonal characteristics.

[0107] In signal processing theory, two perturbation signals are said to be orthogonal if their inner product is zero within a given integration period. The system control's underlying inverter injects frequency-differentiated sinusoidal perturbations or specifically coded square wave perturbation sequences into the coupled motor units.

[0108] For example, a 40 Hz low-frequency sinusoidal perturbation is injected into the upper search loop of motor A, while a 65 Hz sinusoidal perturbation is injected into the coupled motor B. Since these two frequencies do not have integer multiples of harmonics, the voltage ripple superposition effect they generate on the DC bus, as well as the vibration beat frequency generated on the mechanical chassis, cancel each other out after integral filtering over one control cycle.

[0109] Meanwhile, the receivers of each motor, through the adaptive Kalman filtering mentioned in Example 1, combined with bandpass filtering characteristics, can extract the energy efficiency gradient feedback signal belonging to its own frequency with high precision, while filtering out the orthogonal frequency signals injected by other motors as background Gaussian white noise that has no impact on the current optimization. This frequency domain orthogonal decoupling technology enables multiple strongly coupled motors to perform optimization detection in parallel at the same time, significantly improving the collaborative optimization efficiency of the entire multi-motor cluster in response to changing operating conditions.

[0110] Based on existing technology analysis, traditional motor optimization control systems, when extended to multi-motor applications, typically employ an isolated parallel operation mechanism. This means that each motor's local controller independently injects disturbances blindly based on its own efficiency targets. This control mode, lacking top-level topology awareness, is highly susceptible to resonance of optimization disturbances from multiple motors within the electrical network and mechanical structure under variable operating conditions such as sudden load changes. This can not only cause transient drops in DC bus voltage, potentially triggering overcurrent protection shutdowns, but also lead to the breakage of mechanical transmission components due to the superimposed alternating stress.

[0111] The multi-motor cooperative scheduling steps provided in this embodiment employ a graph neural network model to accurately mathematically model and quantify the sensitivity of the multi-physics coupled network. This model integrates health status, marginal cost-effectiveness ratio, and topological constraints to construct a scientific comprehensive priority evaluation system. Combined with the underlying time-frequency dual-dimensional orthogonal injection mechanism, this solution successfully eliminates the physical crosstalk bottleneck in the multi-objective concurrent optimization process. Overall, this embodiment, while ensuring the stable operation of the electrical network and mechanical architecture of the multi-motor system, achieves globally optimal planning of the overall energy efficiency optimization trajectory of the system, demonstrating significant application value and industrial implementation significance in the field of highly integrated electric vehicles and high-end manufacturing equipment.

[0112] Example 3:

[0113] like Figure 2As shown, this embodiment, based on the method logic provided in Embodiments 1 and 2, provides a hardware and software architecture for a motor energy efficiency optimization search and control system oriented towards variable operating conditions. This system maps abstract optimization algorithms to specific industrial control hardware, ensuring stable operation under complex electromagnetic environments and harsh mechanical conditions.

[0114] This embodiment provides a motor energy efficiency optimal search and control system for variable operating conditions, which adopts the aforementioned motor energy efficiency optimal search and control method for variable operating conditions. The system includes: a multi-physics sensing and adaptive Kalman filter purification unit, a multi-physics dynamic digital twin unit, an operating condition trend prediction unit, a marginal cost-effectiveness ratio decision unit, a dual time-scale optimization control unit, and a steady-state locking unit.

[0115] In practical industrial applications such as electric vehicle distributed drive control platforms or industrial multi-axis servo drive cabinets, the aforementioned control units are typically integrated into a central motor controller (MCU) with a heterogeneous multi-core architecture. This includes, for example, an ARM microprocessor for performing complex matrix operations and neural network inference, and a digital signal processor (DSP) or field-programmable gate array (FPGA) for performing high-frequency low-level logic operations. Specifically:

[0116] The multi-physics sensing and adaptive Kalman filtering purification unit is used to synchronously acquire the original multi-physics signals of the motor, perform multi-component frequency domain feature stripping through adaptive Kalman filtering, and output the purified actual current, actual temperature and actual effective vibration amplitude.

[0117] At the hardware level, this unit relies on sensor arrays deployed at key nodes of the drive system, including isolated high-frequency Hall current sensors at the inverter output, a rotary transformer at the rotor shaft end, PT100 thermistors in the stator slots, and piezoelectric accelerometers at the end caps. The analog signals from the sensors are converted into discrete digital signals by a high-speed analog-to-digital converter (ADC) inside the controller. The controller performs adaptive Kalman filter matrix operations through a built-in hardware multiplier-accumulator (MAC) to frequency-domain strip the real-time current from mixed high-frequency switching noise and mechanical vibration, thereby outputting the actual current reflecting the true electromagnetic state and providing a clean data source for subsequent modules.

[0118] The multiphysics dynamic digital twin unit is used to construct a dynamic digital twin model coupling electromagnetic-thermal-mechanical multiphysics fields. The purified actual current, actual temperature, and actual effective vibration amplitude are input into the dynamic digital twin model, and real-time dynamic electrical parameters and detailed health indices are output. This unit allocates an independent data storage area in the controller memory (RAM) to maintain the digital twin base composed of discretized partial differential equations.

[0119] Utilizing the high-speed floating-point operation capability of the DSP core, this unit aligns the low-frequency and high-frequency data fed back by the sensor with the time axis, performs rolling calculations on the stator resistance and quadrature-axis inductance under the current temperature and magnetic flux saturation state, and evaluates the degree of fatigue accumulation of insulating materials and bearings to generate a detailed health index.

[0120] The operating condition trend prediction unit is used to predict the operating condition trend within a preset time window based on the real-time dynamic electrical parameters, and outputs the estimated optimal current vector angle and the predicted steady-state duration. In hardware implementation, this unit typically runs in the application layer processor of the controller, retrieving historical torque and speed sequences via direct memory access (DMA). This unit embeds a lightweight neural network inference engine to predict future driver acceleration intentions or periodic changes in industrial load. Based on the prediction results, the unit accesses a multi-dimensional map stored in non-volatile memory (such as Flash) using a lookup table method to extract the estimated optimal current vector angle, providing a forward-looking initial anchor point for optimization.

[0121] The marginal cost-effectiveness ratio decision unit is used to quantify the marginal lifetime cost and expected energy efficiency benefits of a single optimization in real time, calculate the optimization marginal cost-effectiveness ratio, autonomously decide the optimization granularity mode based on the optimization marginal cost-effectiveness ratio, and output the maximum allowable step size and convergence judgment threshold corresponding to the optimization granularity mode.

[0122] This unit constitutes the core evaluation center of the system, primarily relying on the Arithmetic Logic Unit (ALU) to perform multivariate algebraic and exponential operations. This unit calculates a dimensionless cost-effectiveness ratio by converting the expected energy savings into positive gains and the temperature rise and fatigue caused by disturbances into negative costs. Its hardware logic incorporates multiple levels of comparators, dynamically configuring the step size parameters and decision thresholds in the registers required for the next calculation based on the range of the ratio.

[0123] The dual-timescale optimization control unit is used to perform dual-objective optimization control of energy efficiency and lifespan based on the maximum allowable step size and using a dual-timescale hierarchical architecture: the upper search loop of the first control cycle performs variable step size iterative optimization based on a gradient and curvature dual-drive mechanism, and the lower suppression loop of the second control cycle locally performs torque ripple and harmonic dynamic suppression, wherein the first control cycle is longer than the second control cycle.

[0124] In the controller firmware design, these two loops are assigned to hardware interrupts of different priorities. The timer interrupt corresponding to the first control cycle (e.g., 10 milliseconds) is responsible for calculating the gradient and curvature and issuing current vector angle update instructions; the underlying space vector pulse width modulation (SVPWM) interrupt corresponding to the second control cycle (e.g., 100 microseconds) is responsible for executing current closed-loop feedback and, by injecting feedforward compensation voltage, using the high-frequency switching of the inverter power switching transistors to offset transient harmonics and torque jitter without affecting the macroscopic convergence of upper-level energy efficiency.

[0125] The steady-state locking unit uses a multi-dimensional adaptive threshold to determine the extreme value neighborhood. When the convergence threshold is met, it triggers steady-state zero-disturbance locking, stops injecting optimization disturbances, and forms a search control closed loop. This unit includes a set of state monitoring state machines that continuously compare the current first-order gradient, second-order curvature, and objective function volatility with the thresholds in the registers. Once the conditions are met, the state machine switches to locking mode, triggering the controller's write protection logic, temporarily freezing the search loop from sending new disturbance commands to the underlying layer, allowing the system to smoothly transition to a steady-state high-efficiency operating state until the operating condition trend prediction unit detects another significant external load change.

[0126] Based on existing technology analysis, traditional online motor optimization systems are limited by their flat hardware architecture and singular evaluation dimensions, typically employing a method of continuously injecting disturbances with a fixed step size. This type of system not only easily leads to controller overload under transient conditions, but the continuously generated disturbance signals also inevitably increase additional copper and iron losses, exacerbate fatigue wear of transmission bearings, and face significant problems of steady-state oscillation and over-adjustment in variable operating conditions.

[0127] The hardware and software collaborative system provided in this embodiment successfully integrates complex digital twin analysis and marginal cost-effectiveness decision-making onto a microcontroller platform by utilizing hierarchical computing power and multi-sensor data fusion. This system ensures a balance between finding a high-efficiency operating point and protecting the physical lifespan of the motor itself at the hardware execution level, providing a highly industrially feasible device foundation for solving the technical problem of low overall energy efficiency throughout the entire lifecycle of equipment under varying operating conditions.

[0128] Example 4:

[0129] Corresponding to the above embodiments, the present invention also proposes an electronic device.

[0130] like Figure 3The diagram shows a structural schematic of an electronic device according to the present invention. The electronic device 100 includes a processor 101 and a memory 103. The processor 101 and the memory 103 are connected, for example, via a bus 102. Optionally, the electronic device 100 may further include a transceiver 104. It should be noted that in practical applications, the transceiver 104 is not limited to one unit, and the structure of this electronic device 100 does not constitute a limitation on the embodiments of the present invention.

[0131] Processor 101 may be a CPU, a general-purpose processor, a DSP, an ASIC, an FPGA, or other programmable logic device, transistor logic device, hardware component, or any combination thereof. It may implement or execute the various exemplary logic blocks, modules, and circuits described in connection with this disclosure. Processor 101 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0132] Bus 102 may include a pathway for transmitting information between the aforementioned components. Bus 102 may be a PCI bus or an EISA bus, etc. Bus 102 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 3 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0133] The memory 103 stores a computer program corresponding to the motor energy efficiency optimization search and control method for variable operating conditions in the above embodiments of the present invention. This computer program is controlled and executed by the processor 101. The processor 101 executes the computer program stored in the memory 103 to implement the content shown in the aforementioned method embodiments.

[0134] Among them, electronic devices 100 include, but are not limited to: mobile terminals such as laptops and tablets, as well as fixed terminals such as desktop computers. Figure 3 The electronic device 100 shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of the present invention.

[0135] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for optimal energy efficiency search control of a motor under varying operating conditions, characterized in that, include: The original signals of the multi-physics field of the motor are acquired synchronously, and multi-component frequency domain features are stripped by adaptive Kalman filtering to output the purified actual current, actual temperature and actual effective vibration amplitude. A dynamic digital twin model coupling electromagnetic-thermal-mechanical multiphysics fields is constructed. The purified actual current, actual temperature and actual vibration effective amplitude are input into the dynamic digital twin model, and the real-time dynamic electrical parameters and detailed health index are output. Based on the real-time dynamic electrical parameters, the operating condition trend is predicted within a preset time window, and the estimated optimal current vector angle and the predicted steady-state duration are output. The marginal lifetime cost and expected energy efficiency benefits of a single optimization are quantified in real time, the marginal cost-effectiveness ratio of optimization is calculated, the optimization granularity mode is autonomously decided based on the optimization marginal cost-effectiveness ratio, and the maximum allowable step size and convergence judgment threshold corresponding to the optimization granularity mode are output. Based on the maximum allowable step size, a dual-time-scale hierarchical architecture is adopted to perform dual-objective optimization control for energy efficiency and lifespan: the upper search loop of the first control cycle performs variable step size iterative optimization based on gradient and curvature dual-drive mechanism, and the lower suppression loop of the second control cycle performs torque ripple and harmonic dynamic suppression locally, wherein the first control cycle is longer than the second control cycle. A multi-dimensional adaptive threshold is used to determine the extreme value neighborhood. When the convergence determination threshold is met, steady-state zero disturbance locking is triggered, the injection of optimization disturbance is stopped, and a search control closed loop is formed.

2. The method according to claim 1, characterized in that, The synchronous acquisition of the original multi-physics field signal of the motor, followed by multi-component frequency domain feature stripping using adaptive Kalman filtering, includes: The system synchronously collects multiphase stator current, DC bus voltage, rotor position, mechanical speed, stator winding temperature, bearing vibration acceleration, and permanent magnet temperature; it then performs coordinate transformation on the multiphase stator current to obtain a real-time current vector in an orthogonal stationary coordinate system. Construct a discrete state-space model that includes the actual current component, the optimization disturbance current component, the actual winding temperature component, and the effective vibration amplitude component. The adaptive Kalman filter performs filtering iterations, and adjusts the amplitude of the process noise covariance matrix in real time according to the transient and steady-state conditions determined by the current change rate. The output is the actual current, the actual temperature, and the actual effective vibration amplitude after removing the optimization disturbance current component.

3. The method according to claim 1, characterized in that, The constructed dynamic digital twin model, which couples electromagnetic, thermo-mechanical, and multi-physics fields, outputs real-time dynamic electrical parameters and detailed health indices, including: The pre-stored offline design parameters of the motor are obtained and a multi-physics field coupled digital twin base is constructed. The dynamic time warping algorithm is used to align the purified actual current, actual temperature and actual vibration effective amplitude with the timestamp of the multi-physics field coupled digital twin base. Multiphysics state estimation is performed by combining extended Kalman filtering, and the real-time dynamic electrical parameters are output. The real-time dynamic electrical parameters include dynamic stator resistance, dynamic direct-axis inductance, dynamic quadrature-axis inductance, and dynamic permanent magnet flux linkage. The estimation error of the multiphysics state estimation is calculated, and the subdivided health index is output based on the estimation error. The subdivided health index includes the stator insulation health index, the permanent magnet health index, and the mechanical transmission system health index. The comprehensive health index of the motor is obtained by weighted summation of the subdivided health indices.

4. The method according to claim 1, characterized in that, The step of predicting the operating condition trend within a preset time window based on the real-time dynamic electrical parameters, and outputting the estimated optimal current vector angle and the predicted steady-state duration, includes: Using historical speed and torque time-series data within a preset time period as input, a time-series prediction neural network model and a feature regression model are used to generate speed and torque prediction curves for a future preset time window. Identify the steady-state operating condition range in the speed and torque prediction curves and output the predicted steady-state duration. By combining the real-time dynamic electrical parameters with the predicted operating conditions within the steady-state operating range, the estimated optimal current vector angle is obtained by matching from the dynamic maximum torque-current ratio database.

5. The method according to claim 3, characterized in that, The real-time quantification of the marginal lifetime cost of a single optimization specifically includes: Obtain the preset system reference maximum step size and motor body calibration parameters, and calculate the effective value of current ripple and torque pulsation amplitude caused by optimization based on the system reference maximum step size and motor body calibration parameters; Obtain the estimated optimization convergence time; calculate the marginal lifetime cost of optimization for three types, including: calculating the marginal cost of insulation thermal aging based on the effective value of current ripple, the estimated optimization convergence time and the stator insulation health index; and calculating the marginal cost of permanent magnet demagnetization based on the effective value of current ripple and the estimated optimization convergence time and the permanent magnet health index. The marginal cost of mechanical wear is calculated based on the torque pulsation amplitude, the estimated optimization convergence time, and the mechanical transmission system health index. The total marginal lifetime cost is obtained by weighted summing of the marginal cost of insulation thermal aging, the marginal cost of permanent magnet demagnetization, and the marginal cost of mechanical wear.

6. The method according to claim 5, characterized in that, The expected energy efficiency benefits are calculated by seeking the optimal marginal cost-effectiveness ratio, specifically including: Based on the actual current and the real-time dynamic electrical parameters, calculate the total system energy efficiency loss at the current operating point of the motor; Based on the total energy efficiency loss of the system, calculate the reduction in energy efficiency loss per unit time between the current initial operating point and the estimated optimal point. Obtain the preset average transient loss increment of the optimization process; subtract the product of the energy efficiency loss reduction per unit time and the stable operation time difference from the product of the average transient loss increment of the optimization process and the estimated optimization convergence time to obtain the total expected energy efficiency gain of a single optimization, wherein the stable operation time difference is the predicted steady state duration minus the estimated optimization convergence time. The optimal marginal cost-effectiveness ratio is obtained by dividing the total expected energy efficiency benefit by the total marginal lifetime cost.

7. The method according to claim 6, characterized in that, The step of autonomously deciding on the optimization granularity mode based on the optimization marginal cost-effectiveness ratio, and outputting the maximum allowable step size and convergence judgment threshold corresponding to the optimization granularity mode, includes: Set a high-efficiency cost ratio threshold, a medium-efficiency cost ratio threshold, and a low-efficiency cost ratio threshold, with the values ​​of the three decreasing sequentially; when the marginal cost ratio of the optimization is greater than the high-efficiency cost ratio threshold, it is determined to be in full-quantity high-precision mode, and the maximum allowable step size equal to the maximum step size of the system benchmark is output, as well as the corresponding convergence determination threshold and pulsation threshold. When the marginal cost-effectiveness ratio of the optimization is between the medium cost-effectiveness ratio threshold and the high cost-effectiveness ratio threshold, it is determined to be a standard convergence mode. The maximum allowable step size is reduced by a preset first ratio, as well as the corresponding convergence determination threshold and pulsation threshold, wherein the first ratio is less than 1. When the marginal cost-effectiveness ratio of the optimization is between the low cost-effectiveness ratio threshold and the medium cost-effectiveness ratio threshold, it is determined to be a fast approximation mode, and the maximum allowable step size of the system benchmark maximum step size is amplified by a preset second ratio, as well as the corresponding convergence determination threshold and pulsation threshold, wherein the second ratio is greater than 1. When the marginal cost-effectiveness ratio of optimization is less than or equal to the low cost-effectiveness ratio threshold, it is determined that the optimization mode should be abandoned, the current working point should be maintained and no optimization disturbance should be injected.

8. The method according to claim 3, characterized in that, The upper-level search loop of the first control cycle performs variable-step-size iterative optimization based on a dual-driven mechanism of gradient and curvature, specifically including: A dual-objective overall optimization function for energy efficiency and lifespan is constructed, the value of which is a weighted sum of the energy efficiency loss equivalent calculated based on the total energy efficiency loss of the system and the lifespan loss equivalent calculated based on the comprehensive health index of the motor, with preset weights. Calculate the first-order gradient and second-order curvature of the bi-objective total optimization function with respect to the current vector angle command; Based on the maximum allowable step size, the first-order gradient, the second-order curvature, and the health constraint factor, the current optimization step size amplitude is dynamically updated through absolute value scalar operations, wherein the value of the health constraint factor decreases as the comprehensive health index of the motor decreases. Extract the inverse sign component of the first-order gradient, multiply the inverse sign component by the current optimization step size magnitude, update the current vector angle command, and send it to the lower suppression loop of the second control cycle.

9. The method according to claim 7, characterized in that, The lower-level suppression loop of the second control cycle locally performs torque ripple and harmonic dynamic suppression, specifically including: The total harmonic distortion rate of stator current and the amplitude of electromagnetic torque pulsation are detected in real time in the underlying current loop controller. When the total harmonic distortion rate of the stator current or the amplitude of the electromagnetic torque ripple exceeds the ripple threshold corresponding to the current optimization granularity mode, a feedforward compensation current command is generated to counteract the transient ripple caused by optimization and to maintain the steady-state operating point output by the upper search loop unchanged.

10. A motor energy efficiency optimal search and control system for variable operating conditions, characterized in that, The system employs the optimal motor energy efficiency search control method for variable operating conditions as described in any one of claims 1 to 9, wherein the system comprises: The multi-physics sensing and adaptive Kalman filter purification unit is used to synchronously acquire the original multi-physics signals of the motor, perform multi-component frequency domain feature stripping through adaptive Kalman filtering, and output the purified actual current, actual temperature and actual vibration effective amplitude. A multi-physics dynamic digital twin unit is used to construct a dynamic digital twin model of electromagnetic-thermal-mechanical multi-physics coupling. The purified actual current, actual temperature and actual vibration effective amplitude are input into the dynamic digital twin model, and the real-time dynamic electrical parameters and detailed health index are output. The operating condition trend prediction unit is used to predict the operating condition trend within a preset time window based on the real-time dynamic electrical parameters, and output the estimated optimal current vector angle and the predicted steady-state duration. The marginal cost-effectiveness ratio decision unit is used to quantify the marginal lifetime cost and expected energy efficiency benefits of a single optimization in real time, calculate the optimization marginal cost-effectiveness ratio, autonomously decide the optimization granularity mode based on the optimization marginal cost-effectiveness ratio, and output the maximum allowable step size and convergence judgment threshold corresponding to the optimization granularity mode. A dual-timescale optimization control unit is used to perform dual-objective optimization control of energy efficiency and lifespan based on the maximum allowable step size and using a dual-timescale hierarchical architecture: the upper search loop of the first control cycle performs variable step size iterative optimization based on a gradient and curvature dual-drive mechanism, and the lower suppression loop of the second control cycle locally performs torque ripple and harmonic dynamic suppression, wherein the first control cycle is longer than the second control cycle. The steady-state locking unit is used to determine the extreme value neighborhood using a multi-dimensional adaptive threshold. When the convergence determination threshold is met, steady-state zero-disturbance locking is triggered, the injection of optimization disturbances is stopped, and a search control closed loop is formed.

Images