A method and device for controlling a flux self-balancing three-phase transformer

Abstract

This invention discloses a control method and device for a flux self-balancing three-phase transformer. The method employs a hierarchical control architecture, integrating model predictive control, adaptive inverse model control, and inter-phase cross-coupling compensation. By collecting data such as three-phase flux, load current, and temperature, and after filtering and preprocessing, control weights are adaptively allocated based on operating conditions. The model predictive control layer performs multi-step look-ahead optimization and outputs the target flux trajectory; the adaptive inverse model control layer tracks and identifies parameters online; and the cross-coupling layer achieves coordinated three-phase adjustment. The system also includes mechanisms for zero-sequence flux suppression, parameter calibration, and thermal protection. Compared with traditional methods, this invention can improve operating performance under unbalanced load conditions.

Description

Technical Field

[0001] This invention relates to the field of power transformer control technology, and more specifically to a method and device for controlling a three-phase transformer with magnetic flux self-balancing. Background Technology

[0002] Three-phase transformers are core equipment in power systems and are widely used in industrial power distribution, commercial buildings, rail transit, new energy grid connection, and electric vehicle charging stations. However, in actual operation, three-phase transformers often face unbalanced load conditions, which adversely affect their performance and efficiency.

[0003] Traditional three-phase transformers have the following technical problems under unbalanced load conditions: First, magnetic flux imbalance leads to efficiency degradation. When the three-phase load is unbalanced, the imbalance causes a large deviation in the three-phase magnetic flux, which triggers local magnetic saturation in the core. Magnetic saturation causes nonlinear changes in reluctance, further exacerbating inter-phase coupling mismatch. Under unbalanced load conditions, the transformer's iron losses increase, and efficiency decreases.

[0004] Secondly, zero-sequence flux circulation causes systemic losses. Three-phase unbalanced operation generates a zero-sequence flux component, which seeks a closed path in the core and induces eddy current losses in the tank and structural components. Simultaneously, the zero-sequence flux leads to neutral current overload, causing additional copper losses and thermal stress accumulation, affecting the transformer's reliability and lifespan.

[0005] Secondly, traditional control systems lack active adjustment mechanisms. Traditional transformers employ passive magnetic circuit designs, which cannot dynamically respond to load changes. Magnetic circuit parameters exhibit strong nonlinearity and time-varying characteristics, including factors such as magnetic saturation, temperature drift, and long-term aging, all of which increase the difficulty of control. Physical coupling exists between the three phases, and independent control of a single phase can cause adjustments to one phase to affect the state of the other phases. Furthermore, traditional control methods struggle to coordinate multiple control objectives.

[0006] To address these issues, existing technologies have proposed several improvements. For example, some solutions employ single-phase compensation based on PID control, but this method cannot handle the coupling relationship between the three phases, resulting in limited compensation effectiveness. Other solutions utilize feedforward control based on magnetic circuit models, but establishing these models requires precise parameter identification, and the model accuracy is difficult to guarantee under the influence of nonlinear factors such as magnetic saturation and temperature variations. Still other solutions employ adaptive control methods, but these do not consider the cross-coupling between the three phases and zero-sequence flux suppression, resulting in limited overall performance improvement.

[0007] Recognizing the above problems, the inventors of this invention developed a control scheme that integrates multi-domain algorithm fusion, combining model predictive control algorithm, adaptive inverse model control algorithm, and phase-to-phase cross-coupling compensation algorithm to achieve coordinated self-balancing control of three-phase magnetic flux and improve the performance of transformers under unbalanced load conditions. Summary of the Invention

[0008] The purpose of this invention is to provide a control method and device for a flux self-balancing three-phase transformer. By integrating model predictive control, adaptive inverse model control, and phase-to-phase cross-coupling compensation through a hierarchical fusion control architecture, the invention achieves coordinated self-balancing regulation of three-phase flux and improves the operating performance of traditional transformers under unbalanced load conditions.

[0009] To achieve the above objectives, the present invention provides the following technical solution: A flux self-balancing three-phase transformer control method is proposed, which integrates multi-domain algorithms through a hierarchical control architecture. The first layer is a model predictive control look-ahead optimization layer, responsible for multi-step prediction and trajectory planning over a longer timescale, while handling multi-dimensional constraints and multi-objective optimization. The second layer is an adaptive inverse model control fast tracking layer, responsible for tracking the target trajectory given by model predictive control over a short timescale, and achieving adaptation to changes in magnetic circuit parameters through online identification. The third layer is a weighted adaptive layer, responsible for dynamically adjusting the interphase coupling weight matrix based on the correlation between flux saturation, temperature, and interphase error, achieving coordinated three-phase regulation.

[0010] In the look-ahead optimization layer of model predictive control, this method performs rolling time-domain prediction of the three-phase flux state for multiple future control cycles based on a dynamic flux prediction model. The prediction model includes a magnetic circuit time constant decay term, a position-to-fluid gain term, and a load current-to-fluid coupling term. In the prediction time domain, this method adaptively adjusts the weights of multiple performance indicators based on the predicted flux imbalance coefficient, including flux balance weight, harmonic distortion rate weight, and mechanical regulation energy consumption weight. This method evaluates the weighted comprehensive performance indicators corresponding to multiple candidate compensated control sequences, while checking multi-dimensional constraints such as flux saturation constraints, temperature protection constraints, mechanical velocity constraints, and phase symmetry constraints. From the candidate control sequences that meet the constraints, it selects the control sequence with better performance indicators and outputs the target flux trajectory for multiple future steps.

[0011] In the fast tracking layer of adaptive inverse model control, this method executes inverse model tracking control in each control cycle. First, it calculates the tracking error between the target flux and the actual flux at the current moment. Then, based on the current flux and flux change rate, it predicts the flux value at the next moment and calculates the one-step prediction error. It extracts the target flux value at the next moment from the future trajectory output by the model predictive control and calculates the trajectory look-ahead error. This method weights and sums the current tracking error, the one-step prediction error, and the trajectory look-ahead error according to preset weighting coefficients to obtain a combined error signal. This method constructs an enhanced regression vector, which includes historical compensation current data, historical flux data, the product of the flux deviation of the other two phases and the interphase coupling weights, and the future trajectory predicted by the model predictive control. Based on the combined error signal and the enhanced regression vector, this method uses a recursive least squares algorithm to update the inverse transfer function parameters from flux to compensation current online.

[0012] One innovation of this method lies in its cross-coupling feedforward compensation mechanism. For each phase, the method obtains the deviation between the target flux and the actual flux of the other two phases. It then multiplies the flux deviation of each other phase by the coupling weight of the current phase relative to that other phase and the inverse transfer function parameter of that other phase. Finally, it sums the cross-coupling contributions of all other phases relative to the current phase to obtain the cross-coupling feedforward compensation current of the current phase. This mechanism utilizes the identified inverse model parameters of the other phases and quantifies the magnetic circuit coupling strength between the three phases through inter-phase coupling weights, forming a three-phase interconnected feedforward network.

[0013] This method also includes a zero-sequence flux suppression mechanism. In the model predictive control domain, the method calculates the predicted zero-sequence components of the three-phase flux and outputs the zero-sequence suppression target flux trajectory. In the adaptive inverse model control domain, the method calculates the actual zero-sequence components of the three-phase flux and calculates the zero-sequence suppression compensation current based on the actual zero-sequence components, inverse transfer function parameters, and zero-sequence suppression gain coefficient. The zero-sequence suppression from the two domains is then superimposed and fused.

[0014] This method also incorporates a parameter calibration mechanism between model predictive control and adaptive inverse model control. The method periodically calculates the average value of the three-phase inverse transfer function parameters and uses the reciprocal of this average value to calibrate the predictive model parameters of the model predictive control. Simultaneously, the method uses the reciprocal of the model predictive control predictive model parameters to correct the initial values ​​of the inverse transfer function of the adaptive inverse model control.

[0015] This invention also provides a flux self-balancing three-phase transformer control device, which includes a data acquisition module, an operating condition identification module, a model predictive control module, an adaptive inverse model control module, a cross-coupling feedforward module, a zero-sequence suppression module, a fusion decision module, a thermal protection module, an actuator drive module, a parameter calibration module, and a weighted adaptive module. These modules work collaboratively to achieve all the functions of the aforementioned control method.

[0016] Compared with the prior art, the present invention has the following beneficial effects: 1. Improved operational performance. Through a hierarchical fusion control architecture and a three-phase coordinated regulation mechanism, this invention can improve operational efficiency and flux balance under unbalanced load conditions.

[0017] 2. Zero-sequence flux suppression. A collaborative suppression mechanism combining model predictive control feedforward and adaptive inverse model control feedback can reduce zero-sequence flux and neutral line current.

[0018] 3. Fast dynamic response. By combining look-ahead control with adaptive inverse model control, the recovery time for load step changes can be shortened.

[0019] 4. Good parameter adaptability. Through parameter calibration mechanism and online identification of adaptive inverse model, the system has a high tolerance for parameter deviation.

[0020] 5. Reasonable computational load. Through a two-layer control structure, model predictive control is executed over a longer period, while adaptive inverse model control is executed over a shorter period, resulting in a moderate total computational load.

[0021] 6. Constraint handling capability. Through the explicit constraint checking mechanism of model predictive control, it can simultaneously handle multi-dimensional constraints such as magnetic flux saturation, temperature protection, mechanical velocity, and phase symmetry. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the overall architecture of the control method of the present invention; Figure 2 This is a schematic diagram of the hierarchical fusion control framework of the present invention; Figure 3 This is a schematic diagram of the workflow of the look-ahead optimization layer of the model predictive control in this invention; Figure 4 This is a schematic diagram of the workflow of the adaptive inverse model control fast tracking layer of the present invention; Figure 5 This is a schematic diagram illustrating the principle of the cross-coupling feedforward compensation mechanism of the present invention. Figure 6 This is a schematic diagram illustrating the principle of the parameter calibration mechanism of the present invention; Figure 7 This is a timing diagram illustrating the complete control flow of the present invention; Figure 8 This is a schematic diagram of the module composition of the control device of the present invention; Figure 9 This is a diagram illustrating the performance comparison. Detailed Implementation

[0023] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of protection of the present invention.

[0024] Example 1: Overall Scheme of Control Method See Figure 1 This embodiment provides a control method for a three-phase transformer with self-balancing magnetic flux. This method addresses the problems of magnetic flux distribution imbalance, zero-sequence magnetic flux circulation loss, and insufficient adaptability of the control system under unbalanced load conditions in three-phase transformers by proposing a hierarchical fusion control architecture.

[0025] The hierarchical control architecture comprises: a top-level Model Predictive Control (MPC) look-ahead optimization layer, a middle-level Adaptive Inverse Model Control (AIMC) fast tracking layer, and a bottom-level weighted adaptive layer. These three control layers form an organic whole through parameter interaction, fusion decision-making, and cross-coupling.

[0026] The implementation of this control method includes the following main steps: Step S1: Data Acquisition and Preprocessing Collect operating data from the three-phase transformer, including three-phase magnetic flux data Φ A Φ B Φ C Three-phase load current data I load,A I load,B I load,C Temperature data T and compensation winding position data P A P B P C To obtain accurate magnetic flux data, Hall effect flux sensors or a search coil method can be used. Load current data is acquired through a current transformer. Temperature data is acquired through temperature sensors located at key positions in the core and hot spots in the windings. Compensation winding position data is acquired through a position encoder.

[0027] The acquired data is filtered and preprocessed to reduce measurement noise. Kalman filters or other digital filters can be used to effectively suppress sensor noise and improve control accuracy.

[0028] Step S2: Operating Condition Identification and Adaptive Weight Allocation Based on the preprocessed three-phase magnetic flux data, the average value Φ of the three-phase magnetic flux is calculated. avg Then, the standard deviation normalized index of the three-phase magnetic flux deviation from the average value is calculated and used as the magnetic flux imbalance coefficient σ. ΦThis indicator quantifies the degree of imbalance in three-phase magnetic flux, providing a basis for subsequent operating condition identification and control strategy selection. It also calculates the flux change rate dΦ / dt, which reflects the speed of flux change and serves as an indicator of the severity of load disturbances.

[0029] Based on the magnetic flux imbalance coefficient σ Φ Operating conditions are identified using the rate of change of magnetic flux dΦ / dt. The specific judgment rules are as follows: When σ Φ When the values ​​exceed the first threshold and dΦ / dt exceed the high rate of change threshold, the condition is identified as a severe imbalance. Under this condition, the three-phase magnetic flux deviation is large and changes rapidly, requiring model predictive control to play its look-ahead optimization role.

[0030] When σ Φ When the values ​​are less than the second threshold and dΦ / dt are less than the low rate of change threshold, the condition is determined to be a mild imbalance condition. Under this condition, the three-phase magnetic flux is basically balanced and changes slowly, allowing the adaptive inverse model control to perform its tracking function.

[0031] For other situations, which are classified as moderately unbalanced operating conditions, a strategy of fusion of model predictive control and adaptive inverse model control is adopted. The first and second thresholds represent the actual flux changes under the given operating conditions.

[0032] Based on the operating condition identification results, the adaptive allocation model predicts the control weights α. MPC Adaptive inverse model control weights α AIMC and cross-coupling control weight α cross .

[0033] Step S3: Model predictive control look-ahead optimization See Figure 2 and Figure 3 The model predictive control look-ahead optimization layer performs rolling optimization every preset period. Based on the magnetic flux dynamic prediction model, this layer makes multi-step predictions of the three-phase magnetic flux state for multiple control cycles in the future, and determines the optimal compensation control sequence through optimization algorithms, outputting the target magnetic flux trajectory for the next multiple steps.

[0034] The model predictive control uses a flux dynamic prediction model that includes three core terms: the magnetic circuit time constant decay term A. mag。 Describe the natural decay process of magnetic flux, located at the flux gain term B. mag Describe the effect of compensating winding position adjustment on magnetic flux, and the load current to the magnetic flux coupling term E. mag Describe the effect of load current disturbance on magnetic flux.

[0035] Within the prediction time domain, model predictive control needs to adaptively adjust the weights of several performance metrics based on the predicted flux imbalance coefficient. These performance metrics include the flux balance weight w. balance,MPC Harmonic distortion rate weight w THD and mechanical regulation energy consumption weight w move .

[0036] The weighting adjustment strategy is as follows: when the predicted magnetic flux imbalance coefficient σ Φ,pred When the value is increased, the magnetic flux balance weight w is increased. balance,MPC And reduce the harmonic distortion rate weight w THD Prioritize restoring magnetic flux balance; when the predicted magnetic flux imbalance coefficient σ Φ,pred When decreasing, reduce the magnetic flux balance weight w. balance,MPC And increase the harmonic distortion rate weight w THD and mechanical regulation energy consumption weight w move .

[0037] Model predictive control (MDC) evaluates the weighted composite performance index corresponding to multiple candidate compensated control sequences. For each candidate control sequence in a finite control set, MDC recursively calculates the predicted three-phase flux values ​​for multiple future steps based on the prediction model, and then calculates various performance indices, including flux tracking error, harmonic distortion rate, and mechanical adjustment. These performance indices are weighted and summed using dynamic weights to obtain the composite performance index.

[0038] A key feature of this invention is its explicit constraint handling mechanism. Model predictive control simultaneously examines multidimensional constraints, including: (1) Magnetic flux saturation constraint: Check whether the predicted magnetic flux value of each future step is less than the set ratio of the saturation magnetic flux value to prevent the iron core from being oversaturated.

[0039] (2) Temperature protection constraint: Check whether the product of the temperature-based compensation control gain limit and the flux error exceeds the maximum allowable compensation current.

[0040] (3) Mechanical speed constraint: Check whether the adjustment amount of the compensation winding position in each future step exceeds the product of the maximum allowable speed and the control cycle.

[0041] (4) Phase symmetry constraint: Check whether the sum of the position adjustment of the three-phase compensation windings meets the symmetry requirement.

[0042] Only candidate control sequences that meet the constraints are considered feasible. Model predictive control selects the control sequence with the best overall performance from the set of feasible control sequences and outputs the target flux trajectory Φ for the next several steps. target,MPC (k+1:k+N p Simultaneously, the model predictive control calculates and predicts the zero-sequence component Φ of the three-phase magnetic flux. zero,predIt outputs the zero-sequence suppressed target magnetic flux trajectory.

[0043] Step S4: Adaptive Inverse Model Control for Fast Tracking See Figure 4 The adaptive inverse model control fast tracking layer executes once per control cycle. This layer predicts the target magnetic flux trajectory for the tracking model and achieves adaptation to changes in magnetic circuit parameters through online identification.

[0044] Adaptive inverse model control first calculates the target magnetic flux Φ at the current time. target,MPC (k) and actual magnetic flux Φ actual,i The tracking error e of (k) Φ,i (k). This tracking error reflects the current control deviation.

[0045] Then, the adaptive inverse model control predicts the flux value Φ at the next moment based on the current flux and the rate of change of flux. predict,i (k+1), and calculate the one-step prediction error e Φ,pred,i (k+1).

[0046] Adaptive inverse model control also extracts the target magnetic flux value Φ at the next moment from the future trajectory of the model predictive control output. target,i,MPC (k+1), and calculate the trajectory look-ahead error.

[0047] A key aspect of this invention is the construction of the combined error signal. Adaptive inverse model control weights and sums the current tracking error, one-step prediction error, and trajectory look-ahead error according to preset weighting coefficients to obtain the combined error signal e. combined,i (k).

[0048] Adaptive inverse model control constructs enhanced regression vector φ mag,i (k). This vector contains the following components: (1) Historical Compensation Current Data I comp,i (k-1:kn): Compensation current value for the most recent n control cycles.

[0049] (2) Historical magnetic flux data Φ i (k:km): The magnetic flux value in the most recent m control cycles.

[0050] (3) The product of the other two phase flux deviations and the phase coupling weights.

[0051] (4) Model predicts and controls future trajectories.

[0052] Based on the combined error signal e combined,i (k) and enhanced regression vector φ mag,i (k) Adaptive inverse model control uses the recursive least squares (RLS) algorithm to update the inverse transfer function parameters H of the magnetic flux to the compensation current online.inv,mag,i (k).

[0053] This invention also considers the nonlinear characteristics of the magnetic circuit. Based on the current magnetic flux value, the magnetic flux saturation region is determined, and piecewise nonlinear corrections are applied to the inverse transfer function. (1) When the magnetic flux value is less than the first proportion of the saturation magnetic flux, it is in the linear region, and the linear inverse transfer function H is used. inv,linear,i .

[0054] (2) When the magnetic flux value is between the first and second proportions of the saturation magnetic flux, it is in the transition region. The linear inverse transfer function is multiplied by the nonlinear correction factor of the transition region.

[0055] (3) When the magnetic flux value is greater than the second ratio of the saturation magnetic flux, it is in the saturation region. The linear inverse transfer function is multiplied by the nonlinear correction factor of the saturation region.

[0056] The first ratio and the second ratio are the saturation magnetic flux ratios under actual operating conditions, which are adjusted according to the design of the actual transformer.

[0057] Step S5: Cross-coupling feedforward compensation See Figure 5 Cross-coupling feedforward compensation is one of the core features of this invention.

[0058] For each phase, the cross-coupling feedforward module obtains the deviation between the target flux and the actual flux of the other two phases. Then, the flux deviation of each other phase is multiplied by the coupling weight of the current phase relative to that other phase and the inverse transfer function parameter of that other phase, and the cross-coupling contributions of all other phases relative to the current phase are summed to obtain the cross-coupling feedforward compensation current of the current phase.

[0059] The innovation of this cross-coupled feedforward compensation lies in the following: it avoids redundant modeling by utilizing the inverse model parameters of other phases that have already been identified; it quantifies the magnetic circuit coupling strength between the three phases through inter-phase coupling weights; and it forms a three-phase linked feedforward network.

[0060] Step S6: Zero-sequence flux suppression Suppressing zero-sequence magnetic flux is another technical objective of this invention. Zero-sequence magnetic flux is the same-direction component of the three-phase magnetic flux, which is generated when the load is unbalanced. It will form eddy current loss cycles in the iron core and oil tank, and cause excessive neutral current.

[0061] In the adaptive inverse model control domain, the actual zero-sequence component of the three-phase magnetic flux is calculated. The zero-sequence component is calculated by summing the actual measured values ​​of the three-phase magnetic flux and then dividing by 3, as follows: Φ zero,actual (k)=[Φ A (k)+Φ B (k)+Φ C[(k)] / 3. The zero-sequence suppression compensation current is calculated based on the actual zero-sequence component, inverse transfer function parameters, and zero-sequence suppression gain coefficient. In the model predictive control domain, the predicted future zero-sequence component is calculated, and the zero-sequence suppression target flux trajectory is output. The zero-sequence suppression target flux trajectory of the model predictive control is superimposed and fused with the flux generated by the zero-sequence suppression compensation current of the adaptive inverse model control.

[0062] Step S7: Fusion Decision and Final Compensation Current Generation See Figure 6 The fusion decision is a step in this invention to achieve multi-algorithm collaboration. This step fuses the outputs of the model predictive control layer, the adaptive inverse model control layer, and the cross-coupling layer to generate the final compensation current command.

[0063] First, based on the difference between the target flux in the next future step and the current flux in the model predictive control output, the current inverse transfer function parameters, and the prediction time-domain step number, the compensation current I of the model predictive control pilot term is calculated. comp,i,MPCguided (k).

[0064] Secondly, based on the current inverse transfer function parameters and the weighted sum of the current tracking error and the one-step prediction error, the adaptive inverse model control tracking term compensation current I is calculated. comp,i,AIMCtracked (k).

[0065] Finally, the compensation current of the model predictive control guidance term, the compensation current of the adaptive inverse model control tracking term, the compensation current of the cross-coupling feedforward term, and the compensation current of the zero-sequence suppression term are multiplied by their respective dynamic fusion weights and then summed to obtain the final compensation current command. The final compensation current command is then saturated and limited to prevent the compensation current from exceeding the actuator's safe range.

[0066] Step S8: Thermal protection and compensation gain limiting The temperature data is used to determine whether the temperature is close to the maximum allowable temperature. When the temperature is close to the maximum allowable temperature, thermal protection limits are applied to the compensation control gain to reduce the compensation gain and prevent overcompensation from exacerbating heat generation.

[0067] Step S9: Actuator driving and output filtering The final compensation current command is low-pass filtered and rate-limited to match the bandwidth limitations of the mechanical actuator.

[0068] The compensation current command, after low-pass filtering and rate limiting, is output to the power electronic actuator. The power electronic actuator uses PWM technology to convert the compensation current command into the actual current driving the compensation winding.

[0069] Step S10: Parameter Calibration See Figure 6Parameter calibration is the mechanism by which this invention achieves the fusion of model predictive control and adaptive inverse model control. This step is performed periodically.

[0070] Parameter calibration includes two directions: The first direction is the reverse calibration from adaptive inverse model control to model predictive control. The average value of the three-phase inverse transfer function parameters is calculated, and the reciprocal of this average value is used as the estimate of the model predictive control flux to position gain parameters, and then a weighted average is performed for updating.

[0071] The second direction is the positive correction from model predictive control to adaptive inverse model control. The reciprocal of the model predictive control flux to the position gain parameter is used as the initial value for the correction of the inverse transfer function of the adaptive inverse model control, and then weighted average fusion is performed.

[0072] Step S11: Adaptive update of interphase coupling weight matrix Interphase coupling weight matrix W ij The adaptive update embodies the invention's control over the physical coupling relationship between the three phases. This step is performed periodically. The update of the interphase coupling weight matrix takes into account the effects of the magnetic saturation domain, thermal domain, and crossover error domain.

[0073] Example 2: System startup self-check and fault tolerance mechanism To improve system reliability, this invention performs hardware self-tests during system startup and implements fault-tolerance mechanisms during operation.

[0074] Step S12: System startup self-test During the system startup phase, a hardware self-test is first performed to check the sensor communication status, actuator response status, and memory integrity.

[0075] Step S13: Parameter Loading and Topology Recognition Read the previously saved inverse transfer function parameters, model predictive control prediction model parameters, and phase coupling weight matrix parameters from non-volatile memory.

[0076] Zero-point offset calibration is performed on the sensor under no-load conditions.

[0077] By applying a small-signal excitation and observing the three-phase magnetic flux response, the core topology type is identified and the initial basic weights of the interphase coupling weight matrix are determined.

[0078] Step S14: Saving runtime parameters and fault tolerance During operation, the current parameters are periodically saved to non-volatile memory.

[0079] When abnormal fluctuations in the inverse transfer function parameters are detected exceeding a preset threshold, the inverse transfer function parameter update is frozen and restored to the backup parameter value.

[0080] When a model predictive control failure is detected, the system can relax some constraints and resolve, or enter a degraded mode.

[0081] When an excessively high temperature is detected, thermal protection is triggered, reducing the compensation gain.

[0082] When the position of the compensation winding is detected to be close to the mechanical boundary, the mechanical limit protection is triggered.

[0083] Example 3: Hardware Implementation of the Control Device See Figure 8 This embodiment provides a flux self-balancing three-phase transformer control device. The device includes the following modules: Data acquisition module: This module includes a magnetic flux sensor, a current sensor, a temperature sensor, and a position sensor, as well as corresponding signal conditioning circuits and analog-to-digital converters (ADCs).

[0084] Operating condition identification module: This module calculates the flux imbalance coefficient and flux change rate based on the preprocessed three-phase flux data, identifies the operating condition based on the flux imbalance coefficient and flux change rate, and adaptively allocates control weights.

[0085] Model Predictive Control Module: This module performs multi-step prediction of the three-phase flux state for multiple control cycles based on the flux dynamic prediction model. It adaptively adjusts the performance index weights according to the predicted flux imbalance coefficient, evaluates the weighted comprehensive performance index corresponding to multiple candidate compensation control sequences, checks multi-dimensional constraints, and outputs the future multi-step target flux trajectory and zero-sequence suppression target flux trajectory corresponding to the control sequence that satisfies the constraints.

[0086] Adaptive Inverse Model Control Module: This module calculates the current tracking error, one-step prediction error, and trajectory look-ahead error, and performs a weighted summation to obtain a combined error signal. It then constructs an enhanced regression vector and updates the parameters of the inverse transfer function from magnetic flux to compensation current online based on the combined error signal and the enhanced regression vector. Finally, it performs piecewise nonlinear correction on the inverse transfer function based on the magnetic flux saturation region.

[0087] Cross-coupling feedforward module: This module obtains the flux deviation of the other two phases for each phase, multiplies the flux deviation of each other phase by the coupling weight of the current phase relative to that other phase and the inverse transfer function parameter of that other phase, and then sums them to obtain the cross-coupling feedforward compensation current of the current phase.

[0088] Zero-sequence suppression module: This module calculates the actual zero-sequence component of the three-phase magnetic flux, calculates the zero-sequence suppression compensation current based on the actual zero-sequence component, inverse transfer function parameters and zero-sequence suppression gain coefficient, and superimposes and fuses the zero-sequence suppression target magnetic flux trajectory of the model predictive control with the magnetic flux generated by the zero-sequence suppression compensation current of the adaptive inverse model control.

[0089] Fusion Decision Module: This module calculates the compensation current of the model predictive control guidance term and the compensation current of the adaptive inverse model control tracking term. It then multiplies each compensation current by its corresponding dynamic fusion weight and sums them to obtain the final compensation current command.

[0090] Thermal protection module: This module determines whether the temperature is close to the maximum allowable temperature based on the temperature data and performs thermal protection limit on the compensation control gain.

[0091] Actuator drive module: This module performs saturation limiting, low-pass filtering and rate limiting on the final compensation current command, and drives the power electronic actuator of the compensation winding according to the processed compensation current command.

[0092] Parameter calibration module: This module periodically performs parameter calibration between model predictive control and adaptive inverse model control.

[0093] Weight Adaptive Module: This module periodically calculates correction factors based on the correlation between magnetic flux saturation, temperature, and interphase error, and updates the interphase coupling weight matrix.

[0094] The control unit also includes a memory module and a non-volatile memory module for storing data and saving control algorithm parameters. The control unit also includes a communication interface module to implement communication protocols and data logging functions.

[0095] See Figure 9 This embodiment verifies the technical effects of the present invention through simulation and experimentation, and compares and analyzes it with traditional methods: The simulation platform uses MATLAB / Simulink + PLECS for magnetic circuit simulation. MATLAB / Simulink is used to implement the control algorithm, and PLECS is used to accurately simulate the magnetic circuit characteristics of the transformer, including nonlinear effects such as magnetic saturation, hysteresis, and eddy current losses.

[0096] Five typical operating conditions were set up for the test: Operating condition 1: 30% step imbalance (phase A 100% load, phase B 70%, phase C 70%); Operating Condition 2: Sinusoidal ripple imbalance (load amplitude ±20%, frequency 0.5Hz); Operating Condition 3: Random fluctuation imbalance (Gaussian white noise, standard deviation σ=15%); Operating Condition 4: Temperature shock (ambient temperature jumps from 25°C to 80°C); Operating Condition 5: Parameter Drift (20% linear decay of magnetic permeability, simulating 5 years of aging).

[0097] The step response test results for operating condition 1 show that at the instant the load of phase A jumps from 70% to 100%, the magnetic flux imbalance coefficient σ ΦThe curve of the change of (t) is as follows: Traditional transformers: 0→0.22 (peak value)→0.18 (steady-state value), with no active adjustment capability; Single MPC: 0 → 0.12 (peak) → 0.08 (steady-state, reached in 4 seconds); Single AIMC: 0 ​​→ 0.10 (peak) → 0.06 (steady state, reached in 3 seconds); Traditional algorithm (cross-coupling only): 0 → 0.08 (peak) → 0.04 (steady state, reached in 2.5 seconds); This invention: 0→0.05 (peak value)→0.025 (steady state, reached in less than 2 seconds); The overshoot of this invention is only 42% of that of a single MPC and 50% of that of a single AIMC, demonstrating the advantages of multi-algorithm fusion.

[0098] The long-term drift test under operating condition 5 simulates a 20% gradual decrease in permeability over 5 years of operation. The performance degradation rate comparison is as follows: Traditional transformer: 15% (linear extrapolation, no adaptive capability); Single MPC: 12% (model mismatch accumulates gradually); Single AIMC: 6% (adaptive compensation for partial attenuation); Traditional algorithm: 5% (compensation for weight adjustment); This invention: 2.8% (bidirectional calibration achieves parameter self-consistency).

[0099] The invention exhibits minimal long-term performance degradation, demonstrating the effectiveness of the bidirectional parameter calibration mechanism.

[0100] The following is a comprehensive comparison of various performance indicators under a 30% load imbalance condition: Efficiency: Traditional transformers 85-88%, single MPC 92%, single AIMC 91%, traditional algorithms 93%, this invention ≥95%. This invention improves efficiency by 7-10 percentage points.

[0101] Flux imbalance coefficient: Traditional transformer 0.15-0.25, single MPC 0.08, single AIMC 0.06, traditional algorithm 0.04, this invention <0.03. The imbalance coefficient of this invention is reduced by 80-88%.

[0102] Neutral current ratio: Traditional transformers 80-120%, single MPC 45%, single AIMC 40%, traditional algorithm 35%, this invention <30%. This invention reduces neutral current by 70%.

[0103] Dynamic response time: Traditional transformers require 8-10 cycles, a single MPC requires 4 cycles, a single AIMC requires 3 cycles, traditional algorithms require 2.5 cycles, and this invention requires less than 2 cycles. This invention improves the dynamic response speed by 60%.

[0104] Parameter robustness: Traditional transformer ±5%, single MPC ±8%, single AIMC ±12%, traditional algorithm ±15%, this invention ±20%. The parameter robustness of this invention is improved by 300%.

[0105] Computational load: Traditional transformer 5% (basic PID only), single MPC 85% (evaluating a large number of candidate sequences per cycle), single AIMC 25% (inverse model identification), traditional algorithm 20% (cross-coupled computation), this invention 30% (two-layer structure to distribute the load). This invention reduces the computational load by 60% compared to single MPC.

[0106] Constraint violation rate: Traditional transformer 20 times / hour, single MPC 5 times / hour, single AIMC 3 times / hour, traditional algorithm 1 time / hour, this invention 0 times / hour. This invention achieves zero constraint violations.

[0107] The above experimental results fully verify the technical effects of the present invention, demonstrate its superiority, and show broad application prospects.

[0108] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method of controlling a flux self-balancing three-phase transformer, characterized in that, include: Collect three-phase magnetic flux data, load current data, temperature data, and compensation winding position data of the three-phase transformer, and perform filtering preprocessing on the collected data; The flux imbalance coefficient and flux change rate are calculated based on the preprocessed three-phase flux data. The operating condition is identified based on the flux imbalance coefficient and flux change rate. The model predictive control weight, adaptive inverse model control weight, and cross-coupling control weight are adaptively allocated based on the operating condition identification results. The model predictive control rolling optimization is performed. Based on the magnetic flux dynamic prediction model, the three-phase magnetic flux state of multiple control cycles in the future is predicted in multiple steps. The performance index weights are adaptively adjusted in the prediction time domain. The weighted comprehensive performance index of the candidate compensation control sequence is evaluated and the constraint conditions are checked. The target magnetic flux trajectory and the zero-sequence suppression target magnetic flux trajectory are output. The adaptive inverse model control tracking includes: calculating the tracking error between the target magnetic flux and the actual magnetic flux at the current moment; predicting the magnetic flux value at the next moment based on the current magnetic flux and the rate of change of magnetic flux and calculating the one-step prediction error; extracting the target magnetic flux value at the next moment from the target magnetic flux trajectory output by the model predictive control and calculating the trajectory look-ahead error; weighting and summing the current tracking error, the one-step prediction error, and the trajectory look-ahead error according to preset weight coefficients to obtain a combined error signal; constructing an enhanced regression vector containing historical compensation current data, historical magnetic flux data, the product of the magnetic flux deviation of the other two phases and the phase coupling weight, and the future trajectory of the model predictive control; updating the inverse transfer function parameters online using a recursive least squares algorithm based on the combined error signal and the enhanced regression vector; and performing piecewise nonlinear correction based on the magnetic flux saturation region. For each phase, the flux deviation of the other two phases is obtained. The flux deviation is multiplied by the coupling weight and the inverse transfer function parameter and then summed to obtain the cross-coupled feedforward compensation current. The actual zero-sequence component of the three-phase magnetic flux is calculated. The actual zero-sequence component is obtained by summing the actual measured values ​​of the three-phase magnetic flux and dividing by three. The zero-sequence suppression compensation current is calculated based on the actual zero-sequence component, the inverse transfer function parameters, and the zero-sequence suppression gain coefficient. The model predictive control (MMC) pilot term compensation current is calculated based on the difference between the target flux in the next step and the current flux, the current inverse transfer function parameters, and the number of predicted time-domain steps. The adaptive inverse model control (AMM) tracking term compensation current is calculated based on the current inverse transfer function parameters and the weighted sum of the current tracking error and the one-step prediction error. The MMC pilot term compensation current, the adaptive inverse model control (AMM) tracking term compensation current, the cross-coupling feedforward compensation current, and the zero-sequence suppression compensation current are summed according to dynamic fusion weights to obtain the final compensation current command.

2. The method according to claim 1, characterized in that: It also includes: thermal protection limits based on temperature; The power electronic actuator is driven after the compensation current command is saturated limited, low-pass filtered, and rate limited. Periodically perform bidirectional parameter calibration between model predictive control and adaptive inverse model control; Regularly update the interphase coupling weight matrix; The operating condition identification includes: calculating the average value of the three-phase magnetic flux, and calculating the normalized index of the standard deviation of the three-phase magnetic flux from the average value as the magnetic flux imbalance coefficient; when the magnetic flux imbalance coefficient is greater than a first threshold and the magnetic flux change rate is greater than a high change rate threshold, it is determined to be a severe imbalance condition; when the magnetic flux imbalance coefficient is less than a second threshold and the magnetic flux change rate is less than a low change rate threshold, it is determined to be a mild imbalance condition; otherwise, it is determined to be a moderate imbalance condition. The model predictive control rolling optimization includes: adaptively adjusting the magnetic flux balance weight, harmonic distortion rate weight, and mechanical regulation energy consumption weight based on the predicted magnetic flux imbalance coefficient in the prediction time domain; simultaneously checking magnetic flux saturation constraints, temperature protection constraints, mechanical speed constraints, and phase symmetry constraints, and selecting the control sequence with better performance indicators from the candidate control sequences that meet the constraints; calculating the predicted three-phase magnetic flux zero-sequence component and outputting the zero-sequence suppression target magnetic flux trajectory.

3. The method according to claim 1, characterized in that: The piecewise nonlinear correction includes: using a linear inverse transfer function when the magnetic flux value is less than a first proportion of the saturation magnetic flux; multiplying the linear inverse transfer function by a transition region nonlinear correction factor when the magnetic flux value is between a first proportion and a second proportion of the saturation magnetic flux; and multiplying the linear inverse transfer function by a saturation region nonlinear correction factor when the magnetic flux value is greater than a second proportion of the saturation magnetic flux.

4. The method according to claim 1, characterized in that: The calculation of the cross-coupling feedforward compensation current includes: for each phase, obtaining the deviation between the target flux and the actual flux of the other two phases; multiplying the flux deviation of each other phase by the current coupling weight relative to that other phase and the inverse transfer function parameter of that other phase; and summing the cross-coupling contributions of all other phases relative to the current phase. The calculation of the zero-sequence suppression compensation current also includes: multiplying the future zero-sequence component predicted by the model predictive control by the zero-sequence suppression gain coefficient of the model predictive control to obtain the zero-sequence suppression target magnetic flux trajectory; and superimposing and fusing the zero-sequence suppression target magnetic flux trajectory of the model predictive control with the magnetic flux generated by the zero-sequence suppression compensation current of the adaptive inverse model control.

5. The method according to claim 1, characterized in that: The dynamic fusion weights are allocated as follows: when the condition is determined to be severely unbalanced, the weight of the model predictive control guidance term is set to 0.5-0.7, the weight of the adaptive inverse model control tracking term is set to 0.2-0.4, and the weight of the cross-coupling term is set to 0.1-0.2; when the condition is determined to be slightly unbalanced, the weight of the model predictive control guidance term is set to 0.1-0.3, the weight of the adaptive inverse model control tracking term is set to 0.4-0.6, and the weight of the cross-coupling term is set to 0.2-0.4; when the condition is determined to be moderately unbalanced, the weight of the model predictive control guidance term is set to 0.3-0.5, the weight of the adaptive inverse model control tracking term is set to 0.3-0.5, and the weight of the cross-coupling term is set to 0.1-0.

3.

6. The method of claim 2, wherein, The bidirectional parameter calibration includes: A parameter calibration process is performed every set control cycle. Calculate the average value of the three-phase inverse transfer function parameters, and use the reciprocal of this average value as the estimate of the model predictive control flux to position gain parameters; The estimated values ​​of the control flux to position gain parameters predicted by the model are updated by weighted averaging with the current values ​​according to the fusion coefficient; Calculate the error between the actual measured magnetic flux and the model predicted magnetic flux, and use this error as the disturbance estimate; The load disturbance gain parameters of the model predictive control are updated based on the average value of the disturbance estimates. The inverse of the model-predicted control flux to the position gain parameter is used as the initial value for the correction of the adaptive inverse model control inverse transfer function, and then weighted and fused with the currently identified inverse transfer function value according to the preset learning weights.

7. The method of claim 2, wherein, The updated interphase coupling weight matrix includes: The phase coupling weight matrix update process is performed once every set control cycle; For each phase, a saturation correction factor is calculated based on the ratio of the current magnetic flux value to the saturation magnetic flux value of that phase. The saturation correction factor increases when the magnetic flux value is close to the saturation magnetic flux value. The temperature correction factor is calculated based on the difference between the current temperature and the reference temperature. The temperature correction factor increases as the temperature rises. Obtain the flux error sign of the current phase and the flux error sign of the other phases. When the flux error signs of the two phases are the same, the cross-error coupling factor is positive to enhance the cooperative correction. When the flux error signs of the two phases are opposite, the cross-error coupling factor is negative to avoid overcompensation. The updated interphase coupling weight matrix is ​​obtained by multiplying the basic interphase coupling weights determined based on the core topology by the saturation correction factor, temperature correction factor, and cross-error coupling factor.

8. The method according to claim 1, characterized in that: The recursive least squares algorithm includes: calculating the update amount of the inverse transfer function parameters based on the combined error signal, the enhanced regression vector, and the inner product of the transpose of the enhanced regression vector and itself; adding the current inverse transfer function parameters to the update amount to obtain the inverse transfer function parameters at the next time step; comparing the current flux tracking error with the flux tracking error at the previous time step, increasing the learning rate when the current flux tracking error decreases, and decreasing the learning rate when the current flux tracking error increases; and determining that the inverse transfer function parameters have converged when the change in the inverse transfer function parameters is less than a preset convergence threshold over multiple consecutive update cycles, and reducing the learning rate to enter steady-state tracking mode. The multi-step prediction includes: setting the prediction time domain as multiple control cycles, assuming that the load current remains unchanged within the prediction time domain; for each candidate compensation control sequence in the finite control set, recursively calculating the predicted three-phase magnetic flux values ​​for the next multiple steps based on the magnetic circuit time constant decay term, position-to-magnetic flux gain term, and load current-to-magnetic flux coupling term of the magnetic flux dynamic prediction model; calculating the magnetic flux tracking error, harmonic distortion rate, and mechanical adjustment amount within the prediction time domain for each candidate control sequence; and obtaining a comprehensive performance index by weighting and summing the magnetic flux tracking error, harmonic distortion rate, and mechanical adjustment amount using dynamic weights.

9. The method of claim 2, wherein, Also includes: During the system startup phase, a hardware self-test is performed to check the sensor communication status, actuator response status, and memory integrity. Read the previously saved inverse transfer function parameters, model predictive control prediction model parameters, and interphase coupling weight matrix parameters from non-volatile memory; Zero-point offset calibration is performed on a three-phase flux sensor, load current sensor, and temperature sensor under no-load conditions. By applying a small signal excitation and observing the three-phase magnetic flux response, the core topology type is identified and the initial basic weights of the interphase coupling weight matrix are determined. During operation, the current inverse transfer function parameters, model predictive control predictive model parameters, and interphase coupling weight matrix parameters are periodically saved to non-volatile memory. When it is detected that the inverse transfer function parameter fluctuates abnormally beyond the preset threshold within a short period of time, the inverse transfer function parameter update is frozen and restored to the backup parameter value.

10. A flux self-balancing three-phase transformer control device, characterized in that, include: The data acquisition module is used to collect three-phase magnetic flux data, three-phase load current data, temperature data and compensation winding position data of the three-phase transformer, and to perform filtering preprocessing on the collected data to reduce measurement noise. The operating condition identification module is used to calculate the flux imbalance coefficient and flux change rate based on the preprocessed three-phase flux data, identify the operating condition based on the flux imbalance coefficient and flux change rate, and adaptively allocate model predictive control weights, adaptive inverse model control weights, and cross-coupling control weights. The model predictive control module is used to make multi-step predictions of the three-phase flux state for multiple control cycles based on the flux dynamic prediction model. It adaptively adjusts the performance index weights according to the predicted flux imbalance coefficient, evaluates the weighted comprehensive performance indexes corresponding to multiple candidate compensation control sequences, checks multi-dimensional constraints, and outputs the future multi-step target flux trajectory and zero-sequence suppression target flux trajectory corresponding to the control sequence that satisfies the constraints. An adaptive inverse model control module is used to calculate the tracking error between the target magnetic flux and the actual magnetic flux at the current moment, the one-step prediction error obtained by predicting the magnetic flux value at the next moment based on the current magnetic flux and the rate of change of magnetic flux, and the trajectory look-ahead error obtained by extracting the target magnetic flux value at the next moment from the target magnetic flux trajectory output by the model prediction control module; and to obtain a combined error signal by weighted summation of the current tracking error, the one-step prediction error and the trajectory look-ahead error, and to construct an enhanced regression vector containing historical compensation current data, historical magnetic flux data, phase-to-phase magnetic flux deviation weighted data and model prediction control future trajectory, and to update the inverse transfer function parameters from magnetic flux to compensation current online according to the combined error signal and the enhanced regression vector, and to perform piecewise nonlinear correction on the inverse transfer function according to the magnetic flux saturation region; The cross-coupling feedforward module is used to obtain the magnetic flux deviation of the other two phases for each phase, and then multiply the magnetic flux deviation of each other phase by the coupling weight of the current phase relative to that other phase and the inverse transfer function parameter of that other phase, and sum them to obtain the cross-coupling feedforward compensation current of the current phase. The zero-sequence suppression module is used to calculate the actual zero-sequence component of the three-phase magnetic flux. Based on the actual zero-sequence component, the inverse transfer function parameters, and the zero-sequence suppression gain coefficient, the zero-sequence suppression compensation current is calculated. The zero-sequence suppression target magnetic flux trajectory of the model predictive control is superimposed and fused with the magnetic flux generated by the zero-sequence suppression compensation current of the adaptive inverse model control. The fusion decision module is used to calculate the model predictive control guidance term compensation current based on the difference between the target magnetic flux in the next step and the current magnetic flux output by the model predictive control module, the current inverse transfer function parameters, and the prediction time-domain step number; and to calculate the adaptive inverse model control tracking term compensation current based on the current inverse transfer function parameters and the weighted sum of the current tracking error and the one-step prediction error. The model predictive control guidance term compensation current, the adaptive inverse model control tracking term compensation current, the cross-coupling feedforward compensation current, and the zero-sequence suppression compensation current are multiplied by their respective dynamic fusion weights and then summed to obtain the final compensation current command. The thermal protection module is used to determine whether the temperature is close to the maximum allowable temperature based on the temperature data and to limit the compensation control gain for thermal protection. The actuator drive module is used to perform saturation limiting, low-pass filtering and rate limiting on the final compensation current command, and drive the power electronic actuator of the compensation winding according to the processed compensation current command.

11. The apparatus according to claim 10, characterized in that, Also includes: The parameter calibration module is used to periodically perform bidirectional parameter calibration between model predictive control and adaptive inverse model control. The reciprocal of the average value of the three-phase inverse transfer function parameters is used to calibrate the predictive model parameters of model predictive control, and the reciprocal of the predictive model parameters of model predictive control is used to correct the initial value of the inverse transfer function of adaptive inverse model control. The weighted adaptive module is used to periodically calculate the saturation correction factor, temperature correction factor, and cross-error coupling factor based on the correlation between magnetic flux saturation, temperature, and interphase error, and update the interphase coupling weight matrix.

Images