Motor energy-saving and consumption-reducing intelligent management method and system based on cloud control platform

Abstract

The application provides a motor energy-saving and consumption-reducing intelligent management method and system based on a cloud control platform, which comprises the following steps: collecting single-machine operation data and group-machine operation data at an edge end; solving parameter error criteria by using a sparse inversion matrix based on the group-machine operation data on a cloud control platform; calculating real physical parameters of a target motor including real flux linkage values and real inductance matrixes by combining the parameter error criteria and the single-machine operation data; calculating cross-coupling strength between d-axis current and q-axis current based on the real physical parameters, and generating a conformal correction matrix in a non-orthogonal coordinate system according to the cross-coupling strength; constructing a simulation environment on the cloud control platform and backtesting the real physical parameters and the conformal correction matrix by using the single-machine operation data; and driving the target motor to operate based on the real physical parameters and the conformal correction matrix by a motor controller at the edge end if the energy efficiency improvement index and the stability index are verified. The application has the effect of effectively reducing the energy consumption of the group motor.

Description

Technical Field

[0001] This invention belongs to the field of motor energy-saving management technology, specifically relating to a smart management method and system for motor energy saving and consumption reduction based on a cloud control platform. Background Technology

[0002] In the field of energy conservation and consumption reduction, existing cloud control platforms mainly focus on system-level energy optimization. For example, based on production plans and electricity pricing strategies, they macroscopically schedule the operating time and load distribution of motor groups to reduce overall energy consumption. Current cloud control platforms treat each motor as a standard component with fixed performance parameters, and the control commands issued by the platform are based on this universal assumption. However, since each motor is an independently evolving individual, its key internal physical parameters (such as rotor flux linkage and stator inductance) will have initial differences due to manufacturing tolerances and will undergo irreversible errors due to material aging and wear during long-term operation. Although cloud control platforms collect massive amounts of operating data from motor groups, they find it difficult to use this massive data to make individual corrections. Furthermore, individual edge motor controllers, due to limited computing power and a lack of a global data perspective, cannot autonomously identify their own parameter errors. This deviation between the actual parameters and the controller's fixed model will, over time, lead to additional copper and iron losses, causing the actual operating energy efficiency of the motor to deviate significantly from its theoretical optimal value, thus continuously generating energy waste that could have been avoided. Summary of the Invention

[0003] This invention provides a cloud-based intelligent management method and system for energy saving and consumption reduction of motors to solve the above-mentioned technical problems.

[0004] In a first aspect, the present invention provides an intelligent management method for energy saving and consumption reduction of motors based on a cloud control platform, the method comprising the following steps: Collect single-machine operation data of the target motor and group operation data of all reference motors of the same model as the target motor at the edge, and upload the single-machine operation data and group operation data to the cloud control platform; Based on the group machine operation data, the cloud control platform calculates the parameter error benchmark that reflects the flux attenuation and inductance change of the reference motor using a sparse inversion matrix; By combining parameter error benchmarks and single-machine operation data, the true physical parameters of the target motor, including the true flux linkage value and the true inductance matrix, are calculated. The cross-coupling strength between the d-axis current and the q-axis current is calculated based on the real physical parameters. Based on the cross-coupling strength and using the conformal mapping principle, a conformal correction matrix in a non-orthogonal coordinate system is generated. A simulation environment was built on the cloud control platform, and the real physical parameters and conformal correction matrix were backtested using single-machine running data. The energy efficiency improvement index and stability index of the simulation backtest were calculated and verified. If both the energy efficiency improvement index and the stability index are verified, the target motor will be driven by the motor controller at the edge based on the actual physical parameters and the conformal correction matrix.

[0005] Optionally, the step of calculating the parameter error benchmark reflecting the flux attenuation and inductance change of the reference motor based on the group machine operation data and using a sparse inversion matrix on the cloud control platform includes the following steps: On the cloud control platform, the steady-state and transient components of the motor stator current are extracted from the group machine operation data. The steady-state component is used as the observation feature for identifying flux linkage parameters, and the transient component is used as the observation feature for identifying inductance parameters. Based on the voltage balance equation of the reference motor, a Jacobian matrix is ​​constructed to characterize the sensitivity relationship between flux decay and inductance change and current response deviation, and the elements in the Jacobian matrix are used as non-zero elements of the sparse inversion matrix. Combining group machine operation data, sparse inversion matrix and regularization constraint terms, a global objective function is constructed with the goal of minimizing the L2 norm of the current response deviation of the reference motor. The regularization constraint terms include the smoothness constraint of parameter error in the time dimension and the Gaussian distribution constraint of parameter error in the group dimension. The global objective function is solved iteratively using the preprocessed conjugate gradient method, and the optimal solution vector is extracted from the sparse inversion matrix. The flux attenuation value and inductance change value after decomposing the optimal solution vector are used as the parameter error benchmark.

[0006] Optionally, the step of iteratively solving the global objective function using the preprocessed conjugate gradient method and extracting the optimal solution vector from the sparse inversion matrix, and using the flux linkage attenuation value and inductance change value after decomposing the optimal solution vector as the parameter error benchmark, includes the following steps: Initialize the initial values ​​of the preprocessing conjugate gradient method, set the factory standard parameters of the reference motor as the solution vector of the zeroth iteration, and calculate the initial gradient direction of the global objective function; The sparse inversion matrix is ​​transformed by preprocessing with incomplete Cholesky decomposition. In each iteration step, a linear search is performed along the conjugate gradient direction of the current iteration step to determine the step size that makes the global objective function decrease the fastest and update the solution vector; Calculate the current response deviation residual corresponding to the updated solution vector, statistically analyze the probability density distribution of the current response deviation residual, and use the Huber loss function to perform weighted suppression on outliers in the current response deviation residual based on the probability density distribution. If the gradient magnitude of the global objective function is less than the preset convergence threshold, the iteration stops and the finally converged solution vector is decomposed into flux decay and inductance change values ​​as parameter error benchmarks.

[0007] Optionally, the step of calculating the true physical parameters of the target motor, including the true flux linkage value and the true inductance matrix, by combining the parameter error benchmark and single-machine operating data includes the following steps: Based on the characteristics of the speed signal amplitude and current change rate in the single-machine operation data, the single-machine operation data is divided into steady-state data dominated by rotating back electromotive force and transient data dominated by inductive voltage drop. The residual observation equation is constructed based on the motor control principle. The residual observation equation includes a rotating back electromotive force term and an inductive voltage drop term. The inductance change value in the parameter error reference is used as a fixed parameter of the residual observation equation. The steady-state interval data is substituted into the residual observation equation. The goal is to minimize the projection error of the rotating back electromotive force term on the d-axis and q-axis. The true flux linkage value of the target motor is calculated using the least squares method. The true flux linkage value is used as a fixed parameter in the residual observation equation. The transient interval data is substituted into the residual observation equation. The incremental inductance of the target motor under different currents is identified with the goal of minimizing the voltage vector deviation caused by the inductive voltage drop term. The true inductance matrix of the target motor is then fitted and generated.

[0008] Optionally, the step of calculating the cross-coupling strength between the d-axis current and the q-axis current based on real physical parameters, and generating the conformal correction matrix in the non-orthogonal coordinate system based on the cross-coupling strength and using the conformal mapping principle, includes the following steps: A nonlinear flux linkage model is constructed by combining the real flux linkage value and the real inductance matrix and using the magnetic co-energy function. The partial derivatives of the nonlinear flux linkage model with respect to the d-axis current and the q-axis current are calculated to obtain the incremental inductance matrix containing self-inductance and mutual inductance terms. The magnitude of the mutual inductance term in the incremental inductance matrix is ​​extracted as the cross-coupling strength. Based on the ratio of cross-coupling strength to self-inductance, calculate the equivalent deflection angle between the d-axis magnetic flux path and the q-axis magnetic flux path, and define a virtual non-orthogonal coordinate system determined by the equivalent deflection angle. Establish the complex function mapping relationship from the standard orthogonal coordinate system to the virtual nonorthogonal coordinate system, and solve the derivative matrix of the complex function mapping relationship as the conformal transformation tensor; The conformal transformation tensor is inversely operated on and a resistance asymmetry correction factor is added to generate a conformal correction matrix.

[0009] Optionally, the inverse operation of the conformal transformation tensor and the addition of a resistance asymmetry correction factor to generate the conformal correction matrix includes the following steps: Find the inverse matrix of the conformal transformation tensor to obtain the decoupling transformation kernel that maps non-orthogonal physical quantities back to orthogonal control quantities; The unbalance of the three-phase resistance is calculated based on the resistance values ​​of each phase winding in the single-machine operation data, and a voltage drop compensation vector for resistance asymmetry is constructed. The voltage drop compensation vector for resistor asymmetry is superimposed onto the corresponding elements of the decoupling transformation kernel to form a composite matrix that includes static resistance compensation and dynamic inductance decoupling functions. The composite matrix is ​​discretized into a conformal correction matrix that adapts to the PWM update frequency of the edge motor controller.

[0010] Optionally, the step of constructing a simulation environment on the cloud control platform and using single-machine running data to backtest the real physical parameters and conformal correction matrix, and calculating and verifying the energy efficiency improvement index and stability index of the simulation backtest, includes the following steps: A simulation environment was built on the cloud control platform and single-machine operation data was loaded as the excitation source. A benchmark model with the factory standard parameters of the applied benchmark motor and a test model with the real physical parameters and conformal correction matrix were established respectively. Run the test model, collect the stator current waveform output by the test model, and subtract it from the actual current waveform in the single-machine operation data to obtain the current prediction residual sequence in the time domain. Bode plot analysis was performed on the open-loop transfer function of the test model to extract the amplitude-frequency response curve and the phase-frequency response curve, and the system phase margin at the cutoff frequency was calculated. The reduction in the absolute integral error of the test model relative to the benchmark model within a unit period is quantified as an energy efficiency improvement indicator. The maximum Lyapunov exponent of the state trajectory of the test model was calculated using Lyapunov stability theory. The system determines whether the root mean square value of the current prediction residual sequence is lower than the preset residual limit, whether the system phase margin is greater than the minimum stable phase angle, whether the energy efficiency improvement index is positive, and whether the maximum Lyapunov exponent is less than zero. Only when all the judgment conditions are met simultaneously is it determined that the energy efficiency improvement index and stability index have been verified.

[0011] Optionally, the process of running the test model, acquiring the stator current waveform output by the test model, and subtracting it from the actual current waveform in the single-machine operation data to obtain the current prediction residual sequence in the time domain includes the following steps: The voltage command sequence from the single-machine operation data is input into the test model, and the differential equations of the test model are solved by numerical integration using the Runge-Kutta method to obtain the simulated output current. Align the simulated output current with the actual sampled current in the single-machine operation data on the time axis through cross-correlation. The difference between the aligned simulated output current and the actual sampled current is calculated to generate the original current prediction residual sequence. A fast Fourier transform is performed on the original current prediction residual sequence to separate the fundamental frequency component and the switching frequency harmonic component. The fundamental frequency component after filtering out the switching frequency harmonic component is then reconstructed into the current prediction residual sequence.

[0012] In a second aspect, the present invention also provides a cloud-based intelligent management system for energy saving and consumption reduction of motors, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the cloud-based intelligent management method for energy saving and consumption reduction of motors as described in any one of the first aspects.

[0013] Thirdly, the present invention also provides a computer-readable storage medium storing instructions, characterized in that, when executed by a processor, the instructions cause the processor to be configured to perform the intelligent management method for energy saving and consumption reduction of motors based on a cloud control platform according to any one of the first aspects.

[0014] The beneficial effects of this invention are: This invention utilizes massive amounts of operating data from a group of benchmark motors of the same model and employs a sparse inversion matrix for calculation to construct a high-precision parameter error benchmark. This benchmark profoundly reflects the universal error patterns of key physical parameters such as flux linkage and inductance caused by manufacturing tolerances and service aging. Based on this, the true physical parameters of any in-service target motor can be calculated over time. Then, based on these true physical parameters, the dq-axis cross-coupling strength caused by parameter mismatch in the motor's vector control can be quantitatively calculated. This cross-coupling is the fundamental cause of additional copper and iron losses and deteriorated energy efficiency in the motor. The conformal correction matrix generated using the conformal mapping principle is equivalent to adding a dynamic correction patch to the controller algorithm of this motor, thereby actively counteracting the cross-coupling effect at the control level. After being verified by simulation backtesting in the cloud, the personalized control strategy, which includes real physical parameters and conformal correction matrices, is sent to the edge for execution. This ensures that the motor can maintain the optimal decoupled control state under various operating conditions, fundamentally suppressing parasitic energy loss caused by model mismatch, and bringing the actual operating energy efficiency of the motor as close as possible to its theoretical optimum. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating one embodiment of the intelligent management method for motor energy saving and consumption reduction based on a cloud control platform in this application.

[0016] Figure 2 This is a flowchart illustrating the process of calculating the parameter error reference reflecting the flux linkage decay and inductance change of a reference motor in one embodiment of this application. Detailed Implementation

[0017] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0018] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0019] Figure 1 This is a flowchart illustrating a cloud-based intelligent management method for motor energy saving and consumption reduction in one embodiment. It should be understood that, although... Figure 1 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figure 1 At least some steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed alternately or in turn with other steps or at least a portion of the sub-steps or stages of other steps. For example Figure 1 As shown, the intelligent management method for motor energy saving and consumption reduction based on a cloud control platform disclosed in this invention specifically includes the following steps: S101. Collect single-machine operation data of the target motor and group operation data of all reference motors of the same model as the target motor at the edge, and upload the single-machine operation data and group operation data to the cloud control platform.

[0020] In the Industrial Internet of Things (IIoT) architecture, edge computing nodes are first responsible for establishing real-time communication links with the target motor in the field, and frequently collecting single-machine operating data, including three-phase current, terminal voltage, rotor mechanical speed, and winding temperature. Simultaneously, the edge nodes also aggregate group operating data from all benchmark motors of the same model, specifications, batch, and even operating conditions as the target motor, either from the production line or in the database. This data covers the entire lifecycle characteristics from new machine manufacturing to different aging stages. During the data collection process, the raw analog signals are filtered, denoised, and standardized to ensure that the data uploaded to the cloud meets the requirements of high-precision algorithms. Then, using a high-bandwidth transmission protocol, the processed single-machine unique features and group statistical features are securely transmitted to the big data storage center of the cloud control platform.

[0021] S102. Based on the group machine operation data and using the sparse inversion matrix, the parameter error benchmark reflecting the flux attenuation and inductance change of the reference motor is calculated on the cloud control platform.

[0022] After receiving the uploaded data, the cloud control platform addresses the unavoidable issues of flux linkage attenuation due to magnet demagnetization and inductance parameter errors caused by winding deformation during long-term motor operation. At the algorithm level, sparse inversion theory is introduced to solve this poorly positioned inverse problem. First, the steady-state and transient components of the stator current in the group machine data are extracted, corresponding to the observation windows of flux linkage and inductance characteristics, respectively. A Jacobian matrix reflecting the sensitivity of small parameter changes to the current response is constructed. Using this matrix as the core, a global objective function containing data fitting terms and regularization constraints is constructed. This objective function is iteratively optimized through multiple rounds using the preprocessing conjugate gradient method. While ensuring the sparsity of the solution vector, the residual between the theoretical model and actual observations is minimized, ultimately outputting a set of parameter error benchmarks that represent the common parameter errors of this motor model at the current stage of its life cycle. This process effectively quantifies common aging trends and avoids identification divergence caused by excessive noise in individual machine data.

[0023] S103. Calculate the true physical parameters of the target motor, including the true flux linkage value and the true inductance matrix, by combining the parameter error benchmark and single-machine operation data.

[0024] After obtaining a universally applicable parameter error benchmark, the data stream was precisely segmented into a steady-state interval dominated by the rotating back electromotive force (EMF) and a transient interval dominated by the inductive voltage drop, utilizing the characteristics of speed amplitude and current change rate in the single-machine operating data. This time-domain segmentation ensured the decoupling of the observability of different physical parameters. Based on the motor voltage balance principle, a residual observation equation was constructed. First, the inductance change in the parameter error benchmark was substituted as a known term. Using the steady-state interval data, the objective function that minimizes the projection error of the rotating back EMF term was solved using the least squares method, thus locking in the true flux linkage value of the target motor. Next, the true flux linkage value was substituted back into the equation, and the vector deviation caused by the inductive voltage drop was analyzed using the transient interval data. The incremental inductance under different current conditions was identified, and a complete true inductance matrix was generated through fitting.

[0025] S104. Calculate the cross-coupling strength between the d-axis current and the q-axis current based on the real physical parameters, and generate the conformal correction matrix in the non-orthogonal coordinate system according to the cross-coupling strength and the conformal mapping principle.

[0026] This study constructs a nonlinear flux linkage model based on the actual flux linkage value and inductance matrix. Through differentiation analysis, an incremental inductance matrix containing self-inductance and mutual inductance components is obtained, from which the mutual inductance amplitude is extracted as a quantitative indicator of cross-coupling strength. Since traditional control strategies based on ideal orthogonal coordinate systems cannot effectively handle this coupling, a conformal mapping principle is introduced to construct a virtual non-orthogonal coordinate system to counteract the physical coupling effect. The equivalent deflection angle is calculated based on the ratio of coupling strength to self-inductance, establishing a complex variable function mapping from the standard orthogonal space to the virtual non-orthogonal space. The derivative matrix of this mapping is then solved to obtain the conformal transformation tensor. Subsequently, the tensor is inversely operated on and a resistance asymmetry correction factor is superimposed to generate the final conformal correction matrix. Ultimately, the originally complex nonlinear coupled system is straightened into a linear decoupled system, enabling the controller to control a real motor with severe cross-coupling as if it were an ideal motor. This significantly reduces current fluctuations during high-dynamic operation and improves the linearity and accuracy of torque response.

[0027] S105. Construct a simulation environment on the cloud control platform and use single-machine running data to backtest the real physical parameters and conformal correction matrix, calculate and verify the energy efficiency improvement index and stability index of the simulation backtest.

[0028] Before distributing the optimized control parameters to the edge, rigorous closed-loop verification must be performed in the virtual environment of the cloud control platform. Specifically, using previously collected single-machine operating data as the simulation excitation source, two models are run in parallel in the digital twin environment: a baseline model applying factory standard parameters and a test model loaded with real physical parameters and conformal correction matrices. The Runge-Kutta method is used to numerically integrate and solve the differential equations to obtain the predicted current waveform of the test model, which is then time-domain aligned and residual analyzed with the actual sampled current. Simultaneously, frequency domain Bode plot analysis is performed on the system's open-loop transfer function to calculate stability indices such as phase margin, and the maximum exponent is calculated using Lyapunov theory to quantify the system's dynamic stability. The energy efficiency improvement index is determined by comparing the reduction in the absolute integral error of the test model and the baseline model within a unit period. Verification is considered successful only when the current prediction residual is below the limit, the system phase margin meets the minimum stability angle requirement, the maximum Lyapunov exponent is less than zero, and the energy efficiency index is positive.

[0029] S106. If both the energy efficiency improvement index and the stability index are verified, the target motor will be driven to run by the motor controller at the edge based on the real physical parameters and the conformal correction matrix.

[0030] After rigorous simulation verification and performance evaluation in the cloud, the confirmed effective real physical parameters and conformal correction matrices were packaged into a control firmware update package and remotely sent to the motor controller at the edge via a secure channel. Upon receiving and parsing these parameters, the controller no longer uses the factory-preset fixed parameters but instead updates its internal observer model in real time using the real physical parameters. It then utilizes the conformal correction matrix to perform feedforward decoupling compensation on the PI regulator output of the current loop. Within each PWM (Pulse Width Modulation) control cycle, the controller calculates a precise voltage vector command based on real-time sampled current feedback and the correction matrix, driving the inverter switching action. Because the correction matrix has pre-eliminated the cross-coupling components between coordinate axes and compensated for the effects of flux linkage attenuation, the motor can maintain the optimal angle between the current vector and the rotor magnetic field during operation, significantly reducing the reactive current component used to generate unwanted coupled magnetic fields. The final implementation results in a significant improvement in motor operating efficiency across the entire speed range, reduced heat generation, and more agile dynamic response.

[0031] In one embodiment, reference is made to Figure 2 The steps for calculating the parameter error benchmark reflecting the flux attenuation and inductance change of the reference motor on the cloud control platform based on the group machine operation data and using the sparse inversion matrix include the following: S201. Extract the steady-state and transient components of the motor stator current from the group machine operation data on the cloud control platform, use the steady-state component as the observation feature for identifying flux linkage parameters, and use the transient component as the observation feature for identifying inductance parameters. S202. Based on the voltage balance equation of the reference motor, a Jacobian matrix is ​​constructed to characterize the sensitivity relationship between flux linkage decay and inductance change and current response deviation, and the elements in the Jacobian matrix are used as non-zero elements of the sparse inversion matrix. S203 combines group machine operation data, sparse inversion matrix and regularization constraints, and constructs a global objective function with the goal of minimizing the L2 norm of the current response deviation of the reference motor. The regularization constraints include the smoothness constraint of parameter error in the time dimension and the Gaussian distribution constraint of parameter error in the group dimension. S204. The global objective function is solved iteratively using the preprocessed conjugate gradient method, and the optimal solution vector in the sparse inversion matrix is ​​extracted. The flux attenuation value and inductance change value after the optimal solution vector is decomposed are used as the parameter error benchmark.

[0032] In this embodiment, since the influence mechanism of the physical parameters of the motor on the current waveform varies significantly in the frequency domain under different operating conditions, wavelet transform or high-order filter techniques can be used to accurately separate the acquired stator current sequence into steady-state and transient components. The steady-state component is mainly dominated by the motor's fundamental frequency, reflecting the energy conversion efficiency of the motor under constant speed and load. This signal has a very strong correlation with the attenuation of the flux linkage amplitude, and is therefore designated as a dedicated observation feature for identifying flux linkage parameters. In contrast, the transient component contains high-frequency harmonics and step responses of the current during sudden load changes or speed adjustments. It is mainly determined by the energy storage characteristics of the windings and directly maps the dynamic changes of the stator inductance, thus being used as an observation feature for identifying inductance parameters. Signal decoupling reduces mutual interference during parameter identification, ensuring that subsequent algorithms can accurately capture the independent evolution patterns of flux linkage and inductance from the chaotic data.

[0033] A Jacobian matrix capable of quantifying this causal relationship is constructed based on the voltage balance physical equations of a benchmark motor. Specifically, the voltage equations of the motor... After discretization, its residuals are related to the parameters to be identified (magnetic flux linkage). The partial derivatives of the inductance L and the inductance L form the Jacobian matrix J. This Jacobian matrix is ​​essentially a multidimensional sensitivity mapping table, where the elements of the matrix... The numerical value represents a specific physical parameter. Even small perturbations (such as changes in inductance) can cause deviations in the current response. The degree of generation. Based on this, in order to solve this inverse problem mathematically, a problem of the form of is constructed. The sparse inversion matrix (approximately Hessian matrix) is derived from the fact that the current response at a specific moment is strongly correlated only with the parameter state at that moment and within a very short time window before and after it. Most elements in the matrix are naturally zero. This sparsity not only reduces the computational and storage requirements but also physically reflects the locality of the influence of motor parameter changes on the current.

[0034] After constructing the sensitivity matrix, to obtain a unique and stable solution from the noisy swarm data, a global objective function integrating data-driven approaches and prior knowledge constraints needs to be built. This function optimizes the objective by minimizing the L2 norm of the deviation between the current calculated from the benchmark motor theoretical model and the actual observed current, ensuring that the identification results are numerically closest to the real physical phenomenon. However, relying solely on data fitting can easily lead to overfitting or unstable solutions; therefore, regularization constraints must be introduced: on the one hand, considering that the aging of motor physical parameters over time is a slow and continuous process without abrupt changes, a smoothness constraint in the time dimension is introduced; on the other hand, based on the statistical similarity of the aging patterns of motors in the same batch, a constraint is introduced that the parameter errors follow a Gaussian distribution in the swarm dimension. The mathematical expression is as follows: ,in Let H be the global objective function, H be the sparse inversion matrix, x be the parameter vector to be solved, and y be the observation bias vector. and These are the time smoothing weight coefficient and the group constraint weight coefficient, respectively. The time difference operator matrix, This represents the mean vector of the population parameters. This multi-objective fusion optimization architecture not only effectively suppresses the interference of measurement noise on identification accuracy but also forces the solution process to follow physical evolution laws and population statistical characteristics, thereby outputting parameter error estimates with high confidence and robustness.

[0035] Finally, for the constructed large-scale sparse linear equation system, a preprocessed conjugate gradient method is used for efficient iterative solution to obtain the optimal solution vector in the sparse inversion matrix. Since the Jacobian matrix may have ill-conditioned characteristics, direct solution can lead to slow convergence or even divergence. The preprocessing step significantly improves the condition number of the matrix by performing an incomplete decomposition transformation on the coefficient matrix, making the iterative search direction more accurately point to the global minimum. During the iteration process, the algorithm continuously corrects the solution vector until the gradient magnitude of the objective function drops below the preset convergence threshold. The final converged solution vector contains a mixture of various parameter deviations, which need to be decomposed and mapped back to the physical domain according to a predefined parameter arrangement rule to extract the flux linkage decay value and inductance change value. The iterative update formula follows... In the formula Let be the solution vector for the k-th iteration. To calculate the optimal search step size, The direction is the conjugate gradient direction.

[0036] In one implementation, the global objective function is solved iteratively using the preprocessed conjugate gradient method, and the optimal solution vector in the sparse inversion matrix is ​​extracted. The flux linkage attenuation value and inductance change value after decomposing the optimal solution vector are used as parameter error benchmarks. The steps include: Initialize the initial values ​​of the preprocessing conjugate gradient method, set the factory standard parameters of the reference motor as the solution vector of the zeroth iteration, and calculate the initial gradient direction of the global objective function; The sparse inversion matrix is ​​transformed by preprocessing with incomplete Cholesky decomposition. In each iteration step, a linear search is performed along the conjugate gradient direction of the current iteration step to determine the step size that makes the global objective function decrease the fastest and update the solution vector; Calculate the current response deviation residual corresponding to the updated solution vector, statistically analyze the probability density distribution of the current response deviation residual, and use the Huber loss function to perform weighted suppression on outliers in the current response deviation residual based on the probability density distribution. If the gradient magnitude of the global objective function is less than the preset convergence threshold, the iteration stops and the finally converged solution vector is decomposed into flux decay and inductance change values ​​as parameter error benchmarks.

[0037] In this embodiment, given that the physical parameter errors of industrial motors in actual operation typically exhibit gradual changes relative to their factory design values, rather than sudden and irregular abrupt changes, the factory standard parameters of the benchmark motor can be set as the solution vector for the zeroth iteration, assuming the motor is in an ideal lossless state. This setting utilizes high-confidence prior design data to limit the search range to the vicinity of the convex domain of the global optimum, avoiding the local minima trap that might result from blind random initialization. After determining the initial value, the deviation between the theoretical output generated by substituting this initial value into the system model and the actual observed data is calculated, and this deviation is mapped inversely to the gradient direction of the global objective function at the current position. This gradient direction mathematically points precisely to the direction of the fastest error growth; therefore, its opposite direction constitutes the initial steepest descent path, providing clear guidance for the first iteration. The calculation process follows the formula... ,in Let H represent the initial gradient vector, and H be the sparse inversion matrix. The initial solution vector is set. This represents the actual observed vector.

[0038] Faced with sparse inversion matrices constructed from massive amounts of data from multiple machines, their huge dimensionality and often ill-conditioned characteristics—namely, the extremely uneven distribution of eigenvalues—lead to elongated valley-shaped contour lines in the objective function. Direct solution easily causes the iterative path to oscillate repeatedly at the bottom of these valleys, hindering rapid progress. Therefore, before entering the main iteration loop, an incomplete Joleskine decomposition technique is introduced to preprocess the coefficient matrix. The core principle of this technique is to find a preprocessing matrix that approximates the inverse of the original matrix. Without destroying the sparse structure of the original matrix, this preprocessing matrix performs a transformation on the linear equations, thereby improving the condition number of the coefficient matrix. Unlike complete decomposition, which generates a large number of non-zero padding elements that exhaust computational memory, incomplete decomposition strictly preserves the sparsity of the matrix by discarding padding elements smaller than a certain threshold, achieving an excellent balance between computational accuracy and storage efficiency. Geometrically, this transformation is equivalent to scaling the coordinate axes of the stretched parameter space, reshaping the elongated error valleys into nearly circular basins, allowing the gradient direction to more directly point to the minimum point. In specific implementation, the preprocessing matrix M is constructed to satisfy… , where L is a lower triangular sparse matrix, and the original equation is transformed into a new set of equations with better condition numbers.

[0039] Upon entering the core iterative solution phase, the algorithm no longer uses simple negative gradient directions as search paths. Instead, it employs a conjugate gradient strategy, ensuring that each search direction is orthogonal (conjugate) to the previous direction with respect to the coefficient matrix. This orthogonality guarantees that the optimization results of the current step will not destroy the error components eliminated in previous steps, thus achieving efficient approximation by systematically removing errors from the multidimensional space. In each determined conjugate gradient direction, a precise linear search operation is performed, i.e., finding the local minimum of the objective function on the cross-section of that dimension. By calculating the optimal step size that makes the objective function decrease fastest along that direction, the solution vector is pushed from the current position to the optimal position in that direction, thus completing one iterative update. The update logic follows the formula... In the formula For the updated solution vector, This is the current solution vector. The calculated optimal step size scalar. This is the current conjugate gradient direction vector.

[0040] After each solution vector update, to prevent outliers caused by random noise in the measurement data or sensor malfunctions from devastatingly interfering with the parameter identification results, a robustness assessment of the current fitting state must be performed. This involves calculating the current response deviation residual sequence corresponding to the updated solution vector and introducing the Huber loss function. The residual r is processed. In this embodiment, the specific mathematical form of the Huber loss function is defined as follows: , in, This is the robust threshold. During the iteration process, the standard deviation of the residual sequence is calculated. And set the threshold to (The optimal value for k is 1.345), which dynamically adjusts the sensitivity to outliers. For normal residuals smaller than the threshold, mean square error is used to maintain high accuracy; while for abnormally large residuals exceeding the threshold, the absolute value linear error calculation is automatically switched. This mechanism integrates weighted suppression into the iterative gradient calculation of the conformal gradient method, which is equivalent to installing a smart filter for the optimization process, ensuring that the final identified parameter benchmark is dominated by the physical characteristics of the motor itself.

[0041] After each iteration, the magnitude (Euclidean norm) of the current gradient vector of the global objective function is calculated in real time. This magnitude intuitively reflects the inclination of the current solution on the error surface. When the gradient magnitude continuously decreases and eventually falls below the preset minimum convergence threshold, it indicates that the solution vector has reached the flat valley of the error surface, meaning that the optimal solution that meets the accuracy requirements has been found. At this point, the algorithm immediately stops iterating to avoid overcomputation. Subsequently, the physical meaning of the finally converged mathematical solution vector is restored and decomposed. Since the parameter arrangement structure has been preset when constructing the sparse inversion matrix, it is only necessary to decompose the solution vector into independent values ​​representing the flux attenuation degree and the inductance change amplitude according to the corresponding index rules. The output flux attenuation value and inductance change value constitute a high-precision parameter error benchmark, quantifying the average aging level of the benchmark motor group.

[0042] In one implementation, calculating the true physical parameters of the target motor, including the true flux linkage value and the true inductance matrix, by combining parameter error benchmarks and single-machine operating data includes the following steps: Based on the characteristics of the speed signal amplitude and current change rate in the single-machine operation data, the single-machine operation data is divided into steady-state data dominated by rotating back electromotive force and transient data dominated by inductive voltage drop. The residual observation equation is constructed based on the motor control principle. The residual observation equation includes a rotating back electromotive force term and an inductive voltage drop term. The inductance change value in the parameter error reference is used as a fixed parameter of the residual observation equation. The steady-state interval data is substituted into the residual observation equation. The goal is to minimize the projection error of the rotating back electromotive force term on the d-axis and q-axis. The true flux linkage value of the target motor is calculated using the least squares method. The true flux linkage value is used as a fixed parameter in the residual observation equation. The transient interval data is substituted into the residual observation equation. The incremental inductance of the target motor under different currents is identified with the goal of minimizing the voltage vector deviation caused by the inductive voltage drop term. The true inductance matrix of the target motor is then fitted and generated.

[0043] In this embodiment, the residual observation equation is as follows: For the collected single-machine operation data, it is first necessary to divide it into time segments with different physical meanings through intelligent feature analysis, so as to facilitate subsequent targeted decoupling and identification of different physical parameters. The analysis process is closely based on the amplitude characteristics of the speed signal and the rate of change characteristics of the current signal: data segments exhibiting high speed and minimal current fluctuations, and approximately constant current, are marked as steady-state intervals. In this interval, the electromagnetic physical processes inside the motor are mainly dominated by the back electromotive force generated by the rotation of the permanent magnet cutting the magnetic field lines. Conversely, data segments accompanied by large and rapid changes in current, and in the process of acceleration or deceleration, are marked as transient intervals. At this time, the inductive voltage drop effect of the inductive element resisting the current change is dominant. This time-domain segmentation strategy based on signal characteristics achieves the separation of the observation dimensions of flux linkage parameters and inductance parameters from a physical principle, avoiding the mathematical non-convergence or non-uniqueness of solutions that may be caused by solving multiple coupled variables simultaneously in a single mixed data stream. The mathematical solution of physical parameters depends on establishing a mathematical model that accurately describes the electromagnetic balance relationship inside the motor. Here, the residual observation equation is constructed based on the voltage equation of the synchronous motor in the rotating coordinate system. This set of equations aims to reconstruct the estimated voltages along the d-axis and q-axis through mathematical calculations and compare them with the voltage commands issued by the actual controller. The specific residual observation equations are constructed as follows: ; The output residual of the residual observation equation is defined as: ; in, and Estimate the voltage along the d-axis and q-axis respectively. and These are the d-axis and q-axis voltage commands recorded in the single-machine operation data. and The sampling current for the d-axis and q-axis in the single-machine operation data. The electrical angular velocity in the single-machine operation data. This is the actual flux linkage value. For the rotating back electromotive force term, For stator resistance, and These are the d-axis and q-axis inductance components in the actual inductance matrix. and Together they constitute the inductive voltage drop term.

[0044] In the first stage of parameter identification, the algorithm focuses on accurately determining the true flux linkage value of the target motor using steady-state data. This is because, under steady-state high-speed operating conditions, the rotating back electromotive force term... It plays a dominant role in the voltage balance equation. To eliminate the local minima problem that may be caused by simultaneous optimization of multiple variables, this step fully utilizes the calculation results of the previous steps: the flux attenuation value in the parameter error benchmark obtained from the group machine data is used as the prior initial value (or search center point) for the actual flux optimization, and the inductance change value is temporarily fixed in the residual observation equation as a known quantity. This approach not only utilizes the statistical regularity of the group machine's large data to constrain the search range of the single machine, but also eliminates the interference caused by inductance parameter fluctuations. Subsequently, the selected steady-state interval data stream is substituted into the equation to fine-tune it around the prior initial value. The value is calculated until the mean square error between the estimated voltage and the actual voltage command reaches the global minimum.

[0045] After determining the true flux linkage value, the identification process enters its second stage, which aims to identify the true inductance matrix of the target motor under different current conditions using transient interval data. At this point, the inductance matrix calculated in the previous stage is... By locking these parameters into the residual observation equations, the uncertainty introduced by the unknown flux linkage is eliminated. The algorithm leverages the rapid current changes in transient data, focusing on the inductive voltage drop term in the equations. and The resulting voltage vector deviation. Because the inductance parameters exhibit nonlinear changes due to the magnetic saturation effect caused by the current magnitude, the algorithm optimizes at multiple different current operating points, using the minimization of the objective function J as the criterion to identify the incremental inductance value at each operating point. Subsequently, these discrete inductance observations are numerically fitted to generate a true inductance matrix that continuously describes the magnetic circuit saturation characteristics of the motor. This step effectively digitally reconstructs the dynamic energy storage characteristics of the motor windings; the generated inductance matrix not only contains self-inductance information but also implicitly includes preliminary characteristics of the cross-coupling effect.

[0046] In one implementation, calculating the cross-coupling strength between the d-axis current and the q-axis current based on real physical parameters, and generating a conformal correction matrix in a non-orthogonal coordinate system based on the cross-coupling strength and using the conformal mapping principle, includes the following steps: A nonlinear flux linkage model is constructed by combining the real flux linkage value and the real inductance matrix and using the magnetic co-energy function. The partial derivatives of the nonlinear flux linkage model with respect to the d-axis current and the q-axis current are calculated to obtain the incremental inductance matrix containing self-inductance and mutual inductance terms. The magnitude of the mutual inductance term in the incremental inductance matrix is ​​extracted as the cross-coupling strength. Based on the ratio of cross-coupling strength to self-inductance, calculate the equivalent deflection angle between the d-axis magnetic flux path and the q-axis magnetic flux path, and define a virtual non-orthogonal coordinate system determined by the equivalent deflection angle. Establish the complex function mapping relationship from the standard orthogonal coordinate system to the virtual nonorthogonal coordinate system, and solve the derivative matrix of the complex function mapping relationship as the conformal transformation tensor; The conformal transformation tensor is inversely operated on and a resistance asymmetry correction factor is added to generate a conformal correction matrix.

[0047] In this embodiment, in order to accurately describe the electromagnetic characteristics of the motor under high magnetic saturation and complex cross-coupling conditions, a fourth-order polynomial magnetic co-energy function model including cross-coupling terms was constructed. : , in, The self-inductance coefficient is linear. The key mutual inductance coefficient characterizes the cross-coupling effect. This is the saturation coefficient. By incorporating real physical parameters into this function, an analytical model can be constructed that continuously describes the nonlinear variation of magnetic flux with current. as well as Next, partial derivatives of the flux linkage function with respect to the d-axis current and the q-axis current are taken, resulting in a... Incremental inductance matrix Elements on the diagonal of the matrix and These represent the self-inductance terms of each axis, reflecting the contribution of the current on that axis to the magnetic flux linkage on that axis; while the elements on the anti-diagonal... and This represents the mutual inductance term, revealing how changes in the d-axis current cause q-axis flux linkage fluctuations, and conversely, the cross-coupling phenomenon. The amplitude of the mutual inductance term is directly extracted as a quantitative indicator of the cross-coupling strength. .

[0048] After quantifying the cross-coupling strength, it was found that a strong cross-coupling effect is mathematically equivalent to a distortion of the magnetic flux path, causing the originally orthogonal d-axis and q-axis magnetic circuits to no longer maintain a 90-degree phase difference. Based on the extracted ratio of cross-coupling strength to self-inductance, the equivalent deflection angle of the magnetic flux path relative to the ideal orthogonal axis was calculated using the arctangent function. A new virtual non-orthogonal coordinate system is defined based on the equivalent deflection angle, with the axis of this system aligned with the actual magnetic flux flow path. In this virtual coordinate system, the originally entangled magnetic flux components are naturally separated, transforming the physical coupling relationship into the non-orthogonality of the coordinate axes in the geometric projection. Then, the current vector is treated as a point on the complex plane, and an analytic function is established that conformally maps a point on the standard dq plane to the distorted virtual coordinate plane. Solving for the derivative matrix of this complex function mapping yields the conformal transformation tensor. This tensor is a... The matrix, whose elements contain information about the scaling and rotation of the coordinate system, can accurately describe the geometric deformation patterns within a local space. The calculation formula is embodied in... Therefore, by utilizing the conformal mapping property, the relative phase relationship between the voltage and current vectors remains unchanged before and after the transformation. This maintains the physical intuitiveness of the control system while decoupling, ensuring that no new distortions are introduced during the transformation process.

[0049] Finally, obtain the conformal transformation tensor. inverse matrix In practical motor control systems, this inverse matrix is ​​not used to change the coordinate transformation (such as the Park transformation) itself, but is incorporated into the core algorithm of the feedforward decoupling controller. Specifically, it is implemented by connecting the correction matrix in series after the output voltage command of the PI regulator in the current loop and before the SVPWM module. (Including resistance asymmetry correction) The reference voltage vector directly acts on the output of the current controller. During each PWM interrupt cycle of the controller, real-time sampling is utilized. The current value is queried or the current correction matrix element is calculated, and matrix multiplication is performed: , In this way, conformal mapping theory is concretized into a voltage feedforward compensation strategy. Without changing the original vector control architecture (FOC), the nonlinear voltage distortion caused by magnetic circuit saturation and cross-coupling inside the motor is offset at the control port by real-time correction of the voltage command. This makes the motor appear as an ideally decoupled linear object from the controller's perspective.

[0050] In one implementation, generating a conformal correction matrix by inverse operation of the conformal transformation tensor and adding a resistance asymmetry correction factor includes the following steps: Find the inverse matrix of the conformal transformation tensor to obtain the decoupling transformation kernel that maps non-orthogonal physical quantities back to orthogonal control quantities; The unbalance of the three-phase resistance is calculated based on the resistance values ​​of each phase winding in the single-machine operation data, and a voltage drop compensation vector for resistance asymmetry is constructed. The voltage drop compensation vector for resistor asymmetry is superimposed onto the corresponding elements of the decoupling transformation kernel to form a composite matrix that includes static resistance compensation and dynamic inductance decoupling functions. The composite matrix is ​​discretized into a conformal correction matrix that adapts to the PWM update frequency of the edge motor controller.

[0051] In this embodiment, the inverse matrix of the conformal transformation tensor is obtained through linear algebraic operations. This inverse matrix physically acts as a decoupling transformation kernel. It can receive the complex current response generated by the actual motor in a non-orthogonal magnetic circuit and project it back into a standard orthogonal coordinate system. This ensures that, from the controller's perspective, the current components of the d-axis and q-axis are no longer entangled but rather appear as two independent controlled objects. The calculation process follows the matrix inversion rule. ,in For decoupling transformation kernel, The determinant of a tensor, This is the adjoint matrix. Addressing inductive coupling is only one part of parameter correction; the asymmetry in the actual motor winding resistance also causes voltage vector distortion, leading to torque ripple. Therefore, a detailed unbalance analysis is needed using the measured winding resistance values ​​of each phase from the collected single-machine operating data.

[0052] Specifically, the average and variance of the three-phase resistances are first calculated to quantify the resistance differences between the three phases. Then, a voltage drop compensation vector for resistance mismatch is constructed based on this difference. This vector aims to simulate and cancel the additional voltage drop component on the dq axis caused by resistance mismatch, ensuring that the current loop regulator does not generate erroneous integral accumulation due to resistance deviation. The calculation formula is as follows: The dq-axis components after the Park transform, where For the compensation vector, For the resistance of each phase, The average resistance, This represents the phase current.

[0053] The elements of the voltage drop compensation vector for resistive asymmetry are injected into the corresponding positions of the decoupling transformation kernel or added as bias terms in an additive manner, according to the corresponding d-axis and q-axis influence relationship, thus forming a fully functional composite matrix. In essence, it's an inverse model that incorporates the system's main nonlinear and asymmetric characteristics, capable of simultaneously performing dynamic decoupling and static compensation in a single matrix operation. Mathematically, it is expressed as... Specifically, an equivalent impedance method is used to integrate the voltage compensation term into the matrix structure. This represents the current vector. Considering that edge motor controllers are discrete-time systems based on digital signal processors or microcontrollers, the continuous-domain composite matrix generated in the cloud cannot be used directly and must be discretized. Therefore, based on the controller's underlying PWM carrier frequency and current sampling period, a zero-order hold or bilinear transform method is used to convert the continuous-domain composite matrix into a conformal correction matrix in the form of difference equations. This involves not only the discretization and sampling of numerical values ​​but also the fixed-point scaling of matrix elements to adapt to the finite word-length operation characteristics of embedded processors. The final generated conformal correction matrix can be quickly invoked within each interrupt cycle of the controller for real-time correction of voltage commands.

[0054] In one implementation, the process of building a simulation environment on a cloud control platform and using single-machine running data to backtest the actual physical parameters and conformal correction matrix, and calculating and verifying the energy efficiency improvement and stability indicators of the simulation backtest, includes the following steps: A simulation environment was built on the cloud control platform and single-machine operation data was loaded as the excitation source. A benchmark model with the factory standard parameters of the applied benchmark motor and a test model with the real physical parameters and conformal correction matrix were established respectively. Run the test model, collect the stator current waveform output by the test model, and subtract it from the actual current waveform in the single-machine operation data to obtain the current prediction residual sequence in the time domain. Bode plot analysis was performed on the open-loop transfer function of the test model to extract the amplitude-frequency response curve and the phase-frequency response curve, and the system phase margin at the cutoff frequency was calculated. The reduction in the absolute integral error of the test model relative to the benchmark model within a unit period is quantified as an energy efficiency improvement indicator. The maximum Lyapunov exponent of the state trajectory of the test model was calculated using Lyapunov stability theory. The system determines whether the root mean square value of the current prediction residual sequence is lower than the preset residual limit, whether the system phase margin is greater than the minimum stable phase angle, whether the energy efficiency improvement index is positive, and whether the maximum Lyapunov exponent is less than zero. Only when all the judgment conditions are met simultaneously is it determined that the energy efficiency improvement index and stability index have been verified.

[0055] In this implementation, a digital twin simulation environment is first built on the cloud control platform. This environment aims to completely reproduce the operating conditions of the target motor in the physical world. Previously collected and uploaded single-machine operating data, especially the voltage command sequence and load torque information contained therein, are used as the excitation source to drive the simulation model, ensuring the consistency between the virtual test scenario and the real historical scenario. Two sets of comparative models are built in parallel within this environment: the first is a baseline model, whose internal parameters are strictly set to the standard design values ​​of the motor at the factory, representing the original control state without personalized modifications; the second is a test model, which not only loads the real physical parameters identified in the cloud, including flux linkage decay and inductance changes, but also incorporates a newly generated conformal correction matrix as a feedforward compensation element. After the simulation starts, the stator current waveform data output by the test model at each moment is automatically recorded. To evaluate the model's fit to the real physical object, this simulated output current is compared point-by-point with the actual sampled current synchronously recorded in the single-machine operating data. Through simple subtraction, a time-domain current prediction residual sequence is generated. This residual sequence intuitively reflects the deviation between the model's predicted value and the actual value. If the residual sequence remains at an extremely low level and there is no obvious systematic deviation, it indicates that the parameters and correction matrix in the test model have captured the true physical characteristics of the motor with extremely high accuracy; conversely, if the residual is large or there are periodic fluctuations, it means that there are still errors in the identification results.

[0056] Besides time-domain fit analysis, frequency-domain stability analysis is equally crucial. Using frequency response analysis from control theory, the open-loop transfer function of the closed-loop control system constructed from the test model is derived and calculated. Based on this transfer function, a Bode plot is plotted, and the amplitude-frequency response curve reflecting system gain changes and the phase-frequency response curve reflecting signal delay are extracted. Special attention is paid to the cutoff frequency point where the system open-loop gain crosses the 0dB line, and the difference between the phase angle at this frequency and -180 degrees is calculated, i.e., the system phase margin. In the formula, PM represents the phase margin. Cutoff frequency The phase margin is a core indicator for measuring the relative stability of a feedback control system, quantifying the additional phase lag the system can withstand before oscillations occur. This analysis allows us to predict whether a control system with new parameters possesses sufficient damping to suppress oscillations and prevent system instability caused by parameter mismatch when facing high-frequency disturbances or dynamic load changes.

[0057] To quantify the energy-saving effect, it is necessary to compare the differences in current tracking performance between the old and new models. The energy efficiency improvement index is defined as the degree of improvement in current control accuracy of the test model relative to the baseline model. Specifically, the absolute values ​​of the current tracking errors (i.e., the difference between the model output current and the ideal reference current) of both the baseline and test models are integrated over a unit control cycle to obtain their respective integral absolute errors (IAE). The energy efficiency improvement index is calculated as the difference between the integral error of the baseline model and the integral error of the test model. Since the current tracking error is directly related to the motor's torque ripple and reactive power loss, a reduction in error means that more electrical energy is effectively converted into mechanical energy, rather than being consumed in heat and vibration. Therefore, a positive index, and a larger value, indicates that the new control strategy requires a more stable and precise current to maintain the same torque output, thus directly confirming the actual potential for energy saving and consumption reduction.

[0058] Furthermore, Lyapunov stability theory in nonlinear dynamics is used to rigorously examine the dynamic behavior of the test model. The maximum Lyapunov exponent of the state-space trajectory of the test model is calculated. This index describes the average exponential separation rate of two adjacent trajectories in the system's phase space over time. Its calculation is based on the formula... ,in Let be the state deviation vector at time t. This means that after a small disturbance, the system's state trajectory will eventually converge back to the equilibrium point, indicating that the system is asymptotically stable; conversely, if... This means that small deviations will amplify exponentially over time, indicating the risk of chaos or divergence.

[0059] Finally, based on the multi-dimensional verification results, a rigorous logical judgment is performed. Four key thresholds are set: the upper limit of the root mean square value of the current prediction residual, the minimum allowable angle of the system phase margin, the positive threshold of the energy efficiency improvement index, and the zero point of the maximum Lyapunov exponent. The logical judgment expression is as follows: Only when all four conditions are met simultaneously will the system determine that the real physical parameters and conformal correction matrix generated in the cloud have passed all verifications. The new control strategy not only reproduces the real physical process in terms of accuracy and has the ability to resist disturbances in terms of stability, but also achieves a substantial improvement in energy efficiency and ensures long-term convergence in terms of dynamic characteristics.

[0060] In one implementation, running a test model, acquiring the stator current waveform output by the test model, and subtracting it from the actual current waveform in the single-machine operation data to obtain the current prediction residual sequence in the time domain includes the following steps: The voltage command sequence from the single-machine operation data is input into the test model, and the differential equations of the test model are solved by numerical integration using the Runge-Kutta method to obtain the simulated output current. Align the simulated output current with the actual sampled current in the single-machine operation data on the time axis through cross-correlation. The difference between the aligned simulated output current and the actual sampled current is calculated to generate the original current prediction residual sequence. A fast Fourier transform is performed on the original current prediction residual sequence to separate the fundamental frequency component and the switching frequency harmonic component. The fundamental frequency component after filtering out the switching frequency harmonic component is then reconstructed into the current prediction residual sequence.

[0061] In this embodiment, the voltage command sequence recorded in the single-machine operation data is extracted. This sequence represents the voltage excitation applied to the motor terminals by the actual controller at a certain point in the past. This voltage sequence, along with the initial state of the motor (such as speed and position), is loaded as an input signal into the test model. The test model is essentially a set of nonlinear ordinary differential equations describing the electromagnetic transient process of the motor. To solve these equations, the algorithm uses the fourth-order Runge-Kutta method (RK4) for numerical integration. This method approximates the state value at the next time step by calculating the slope-weighted average of four different points within each time step, exhibiting extremely high computational accuracy and numerical stability. The calculation process is executed iteratively. Where y represents the current state variable and h is the simulation step size. to These are four intermediate slope values.

[0062] Because unavoidable time synchronization errors often exist between the actual data acquisition system and the cloud simulation system, such as transmission delays or clock drift, directly comparing the simulated current with the actual current may lead to human-induced phase errors. Therefore, time axis alignment must be performed before calculating the difference. The simulated output current sequence is calculated using cross-correlation function analysis techniques. With the actual sampled current sequence The sequence of cross-correlation coefficients between The lag time that allows the cross-correlation coefficient to reach its maximum value through searching. This allows us to determine the optimal time offset between the two sequences. Subsequently, the simulated current sequence is shifted and corrected based on this offset to ensure that it perfectly coincides with the actual current sequence on the time axis.

[0063] After time alignment, the difference between the aligned simulated output current and the actual sampled current is calculated point by point using vector subtraction to generate the original current prediction residual sequence. The original residual sequence contains all the deviation information caused by model parameter errors, but it is also mixed with a large amount of measurement noise and high-frequency interference. Therefore, a fast Fourier transform is used to convert the original residual sequence in the time domain to the frequency domain for spectral analysis. In the frequency domain spectrum, based on the carrier frequency characteristics of the motor controller, the switching frequency harmonic components concentrated in the high-frequency band and the fundamental frequency components distributed in the low-frequency band can be clearly identified and located. The algorithm designs a digital low-pass filter or frequency domain mask to accurately filter out the energy spectrum near the switching frequency and its harmonics, retaining only the fundamental frequency components that reflect parameter drift and model mismatch. Subsequently, the inverse fast Fourier transform is used to restore the processed frequency domain data back to the time domain, reconstructing the final current prediction residual sequence.

[0064] The present invention also discloses an intelligent management system for energy saving and consumption reduction of motors based on a cloud control platform, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the intelligent management method for energy saving and consumption reduction of motors based on a cloud control platform as described above.

[0065] The processor can be a central processing unit (CPU). Of course, depending on the actual use, it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), off-the-shelf programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc., and this application does not limit it.

[0066] The memory can be an internal storage unit of a computer device, such as a hard disk or RAM, or an external storage device, such as a plug-in hard disk, smart memory card (SMC), secure digital card (SD), or flash memory card (FC) provided on the computer device. Furthermore, the memory can be a combination of internal storage units and external storage devices of a computer device. The memory is used to store computer programs and other programs and data required by the computer device. The memory can also be used to temporarily store data that has been output or will be output. This application does not limit this.

[0067] The present invention also discloses a computer-readable storage medium storing instructions that, when executed by a processor, cause the processor to be configured to perform the intelligent motor energy-saving and consumption-reducing management method based on a cloud control platform as described in any of the above embodiments.

[0068] The computer program can be stored in a machine-readable medium. The computer program includes computer program code, which can be in the form of source code, object code, executable file, or certain middleware. The machine-readable medium includes any entity or device capable of carrying computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the machine-readable medium includes, but is not limited to, the above-mentioned components.

[0069] The above-described intelligent management method for energy saving and consumption reduction of motors based on cloud control platform is stored in the computer-readable storage medium and loaded and executed on the processor to facilitate the storage and application of the above method.

[0070] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of protection of this application is limited to these examples; within the framework of this application, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of one or more embodiments of this application as described above, which are not provided in detail for the sake of brevity.

[0071] One or more embodiments in this application are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of this application. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of one or more embodiments in this application should be included within the protection scope of this application.

Claims

1. A smart management method for energy saving and consumption reduction of motors based on a cloud control platform, characterized in that, Includes the following steps: Collect single-machine operation data of the target motor and group operation data of all reference motors of the same model as the target motor at the edge, and upload the single-machine operation data and group operation data to the cloud control platform; Based on the group machine operation data, the cloud control platform calculates the parameter error benchmark that reflects the flux attenuation and inductance change of the reference motor using a sparse inversion matrix; By combining parameter error benchmarks and single-machine operation data, the true physical parameters of the target motor, including the true flux linkage value and the true inductance matrix, are calculated. The cross-coupling strength between the d-axis current and the q-axis current is calculated based on the real physical parameters. Based on the cross-coupling strength and using the conformal mapping principle, a conformal correction matrix in a non-orthogonal coordinate system is generated. A simulation environment was built on the cloud control platform, and the real physical parameters and conformal correction matrix were backtested using single-machine running data. The energy efficiency improvement index and stability index of the simulation backtest were calculated and verified. If both the energy efficiency improvement index and the stability index are verified, the target motor will be driven by the motor controller at the edge based on the actual physical parameters and the conformal correction matrix.

2. The intelligent management method for motor energy saving and consumption reduction based on a cloud control platform according to claim 1, characterized in that, The step of calculating the parameter error benchmark reflecting the flux attenuation and inductance change of the reference motor based on the group machine operation data and using a sparse inversion matrix on the cloud control platform includes the following steps: On the cloud control platform, the steady-state and transient components of the motor stator current are extracted from the group machine operation data. The steady-state component is used as the observation feature for identifying flux linkage parameters, and the transient component is used as the observation feature for identifying inductance parameters. Based on the voltage balance equation of the reference motor, a Jacobian matrix is ​​constructed to characterize the sensitivity relationship between flux decay and inductance change and current response deviation, and the elements in the Jacobian matrix are used as non-zero elements of the sparse inversion matrix. Combining group machine operation data, sparse inversion matrix and regularization constraint terms, a global objective function is constructed with the goal of minimizing the L2 norm of the current response deviation of the reference motor. The regularization constraint terms include the smoothness constraint of parameter error in the time dimension and the Gaussian distribution constraint of parameter error in the group dimension. The global objective function is solved iteratively using the preprocessed conjugate gradient method, and the optimal solution vector is extracted from the sparse inversion matrix. The flux attenuation value and inductance change value after decomposing the optimal solution vector are used as the parameter error benchmark.

3. The intelligent management method for motor energy saving and consumption reduction based on a cloud control platform according to claim 2, characterized in that, The process of iteratively solving the global objective function using the preprocessed conjugate gradient method and extracting the optimal solution vector from the sparse inversion matrix, using the flux linkage attenuation value and inductance change value after decomposing the optimal solution vector as the parameter error benchmark, includes the following steps: Initialize the initial values ​​of the preprocessing conjugate gradient method, set the factory standard parameters of the reference motor as the solution vector of the zeroth iteration, and calculate the initial gradient direction of the global objective function; The sparse inversion matrix is ​​transformed by preprocessing with incomplete Cholesky decomposition. In each iteration step, a linear search is performed along the conjugate gradient direction of the current iteration step to determine the step size that makes the global objective function decrease the fastest and update the solution vector; Calculate the current response deviation residual corresponding to the updated solution vector, statistically analyze the probability density distribution of the current response deviation residual, and use the Huber loss function to perform weighted suppression on outliers in the current response deviation residual based on the probability density distribution. If the gradient magnitude of the global objective function is less than the preset convergence threshold, the iteration stops and the finally converged solution vector is decomposed into flux decay and inductance change values ​​as parameter error benchmarks.

4. The intelligent management method for motor energy saving and consumption reduction based on a cloud control platform according to claim 1, characterized in that, The process of calculating the true physical parameters of the target motor, including the true flux linkage value and the true inductance matrix, by combining parameter error benchmarks and single-machine operation data includes the following steps: Based on the characteristics of the speed signal amplitude and current change rate in the single-machine operation data, the single-machine operation data is divided into steady-state data dominated by rotating back electromotive force and transient data dominated by inductive voltage drop. The residual observation equation is constructed based on the motor control principle. The residual observation equation includes a rotating back electromotive force term and an inductive voltage drop term. The inductance change value in the parameter error reference is used as a fixed parameter of the residual observation equation. The steady-state interval data is substituted into the residual observation equation. The goal is to minimize the projection error of the rotating back electromotive force term on the d-axis and q-axis. The true flux linkage value of the target motor is calculated using the least squares method. The true flux linkage value is used as a fixed parameter in the residual observation equation. The transient interval data is substituted into the residual observation equation. The incremental inductance of the target motor under different currents is identified with the goal of minimizing the voltage vector deviation caused by the inductive voltage drop term. The true inductance matrix of the target motor is then fitted and generated.

5. The intelligent management method for motor energy saving and consumption reduction based on a cloud control platform according to claim 1, characterized in that, The process of calculating the cross-coupling strength between the d-axis current and the q-axis current based on real physical parameters, and generating the conformal correction matrix in the non-orthogonal coordinate system based on the cross-coupling strength and the conformal mapping principle, includes the following steps: A nonlinear flux linkage model is constructed by combining the real flux linkage value and the real inductance matrix and using the magnetic co-energy function. The partial derivatives of the nonlinear flux linkage model with respect to the d-axis current and the q-axis current are calculated to obtain the incremental inductance matrix containing self-inductance and mutual inductance terms. The magnitude of the mutual inductance term in the incremental inductance matrix is ​​extracted as the cross-coupling strength. Based on the ratio of cross-coupling strength to self-inductance, calculate the equivalent deflection angle between the d-axis magnetic flux path and the q-axis magnetic flux path, and define a virtual non-orthogonal coordinate system determined by the equivalent deflection angle. Establish the complex function mapping relationship from the standard orthogonal coordinate system to the virtual nonorthogonal coordinate system, and solve the derivative matrix of the complex function mapping relationship as the conformal transformation tensor; The conformal transformation tensor is inversely operated on and a resistance asymmetry correction factor is added to generate a conformal correction matrix.

6. The intelligent management method for motor energy saving and consumption reduction based on a cloud control platform according to claim 5, characterized in that, The inverse operation of the conformal transformation tensor and the addition of a resistance asymmetry correction factor to generate the conformal correction matrix includes the following steps: Find the inverse matrix of the conformal transformation tensor to obtain the decoupling transformation kernel that maps non-orthogonal physical quantities back to orthogonal control quantities; The unbalance of the three-phase resistance is calculated based on the resistance values ​​of each phase winding in the single-machine operation data, and a voltage drop compensation vector for resistance asymmetry is constructed. The voltage drop compensation vector for resistor asymmetry is superimposed onto the corresponding elements of the decoupling transformation kernel to form a composite matrix that includes static resistance compensation and dynamic inductance decoupling functions. The composite matrix is ​​discretized into a conformal correction matrix that adapts to the PWM update frequency of the edge motor controller.

7. The intelligent management method for motor energy saving and consumption reduction based on a cloud control platform according to claim 1, characterized in that, The process of constructing a simulation environment on a cloud control platform and using single-machine running data to backtest real physical parameters and conformal correction matrices, and calculating and verifying the energy efficiency improvement and stability indicators of the simulation backtest, includes the following steps: A simulation environment was built on the cloud control platform and single-machine operation data was loaded as the excitation source. A benchmark model with the factory standard parameters of the applied benchmark motor and a test model with the real physical parameters and conformal correction matrix were established respectively. Run the test model, collect the stator current waveform output by the test model, and subtract it from the actual current waveform in the single-machine operation data to obtain the current prediction residual sequence in the time domain. Bode plot analysis was performed on the open-loop transfer function of the test model to extract the amplitude-frequency response curve and the phase-frequency response curve, and the system phase margin at the cutoff frequency was calculated. The reduction in the absolute integral error of the test model relative to the benchmark model within a unit period is quantified as an energy efficiency improvement indicator. The maximum Lyapunov exponent of the state trajectory of the test model was calculated using Lyapunov stability theory. The system determines whether the root mean square value of the current prediction residual sequence is lower than the preset residual limit, whether the system phase margin is greater than the minimum stable phase angle, whether the energy efficiency improvement index is positive, and whether the maximum Lyapunov exponent is less than zero. Only when all the judgment conditions are met simultaneously is it determined that the energy efficiency improvement index and stability index have been verified.

8. The intelligent management method for motor energy saving and consumption reduction based on a cloud control platform according to claim 7, characterized in that, The process of running the test model involves acquiring the stator current waveform output by the test model and subtracting it from the actual current waveform in the single-machine operation data to obtain the current prediction residual sequence in the time domain. This includes the following steps: The voltage command sequence from the single-machine operation data is input into the test model, and the differential equations of the test model are solved by numerical integration using the Runge-Kutta method to obtain the simulated output current. Align the simulated output current with the actual sampled current in the single-machine operation data on the time axis through cross-correlation. The difference between the aligned simulated output current and the actual sampled current is calculated to generate the original current prediction residual sequence. A fast Fourier transform is performed on the original current prediction residual sequence to separate the fundamental frequency component and the switching frequency harmonic component. The fundamental frequency component after filtering out the switching frequency harmonic component is then reconstructed into the current prediction residual sequence.

9. A cloud-based intelligent management system for motor energy saving and consumption reduction, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the intelligent management method for energy saving and consumption reduction of motors based on a cloud control platform as described in any one of claims 1 to 8.

10. A computer-readable storage medium storing instructions thereon, characterized in that, When executed by the processor, the instruction causes the processor to be configured to perform the intelligent management method for energy saving and consumption reduction of motors based on a cloud control platform according to any one of claims 1 to 8.

Images