Transformer on-load voltage regulation control method

Abstract

The invention relates to the technical field of on-load voltage regulation, in particular to a transformer on-load voltage regulation control method. According to the technical scheme, the transformer on-load voltage regulation control method comprises the following steps that S1, a transformer cluster running in parallel is decomposed into distributed intelligent nodes, and each node executes local voltage prediction and constraint condition management; s2, adopting a distributed model prediction control framework, and cooperatively solving a voltage regulation optimization target through an edge calculation unit; s3, a harmonic disturbance signal with a preset amplitude is injected into the transformer, equivalent impedance is dynamically identified based on response data, and virtual impedance compensation is triggered; and S4, establishing a hierarchical cooperative control architecture, and realizing clock synchronization, dynamic weight distribution and multi-objective optimization decision among nodes. Through a cooperative mechanism of distributed predictive control and dynamic impedance compensation, the response speed and the regulation precision of the on-load voltage regulation system to the power grid voltage fluctuation are remarkably improved, and the voltage out-of-limit problem under the complex working conditions of new energy grid connection, load sudden change and the like is effectively solved.

Description

Technical Field

[0001] This invention relates to the field of on-load tap changer technology, and more particularly to an on-load tap changer control method for transformers. Background Technology

[0002] Traditional on-load tap changer control of transformers often employs a centralized PID control strategy, directly driving the tap changer by detecting bus voltage deviation. This method relies on fixed parameter tuning, and control commands are issued to the parallel transformer group through a single master station. In the impedance parameter matching stage, a static impedance model is typically set based on the equipment nameplate values, lacking dynamic adaptation to changes in operating conditions.

[0003] Traditional on-load tap changer control methods for transformers largely rely on centralized control strategies, which have significant deficiencies in response speed and dynamic coordination capabilities. In scenarios involving new energy grid integration or sudden load changes, fixed-parameter control modes struggle to track voltage fluctuations in a timely manner, easily leading to transient voltage instability or even power outages. For transformer clusters operating in parallel, centralized control strategies lack real-time impedance matching mechanisms, resulting in the continuous accumulation of interphase circulating currents. This not only accelerates tap changer contact wear but may also cause localized overheating faults, significantly increasing the maintenance burden. Furthermore, static impedance models cannot adapt to changes in transformer operating conditions such as aging and winding deformation, often leading to tap changer failure due to parameter mismatch, severely impacting power supply reliability. Summary of the Invention

[0004] This invention proposes an on-load tap changer control method for transformers, which solves the problems of existing centralized control strategies having slow response and lack of dynamic impedance matching, leading to circulating current accumulation and equipment overload during the tap changer process of parallel transformer groups. At the same time, the static impedance model cannot adapt to changes in operating conditions, causing tap changer failure.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0006] A method for on-load tap changing control of a transformer includes the following steps:

[0007] S1. Decompose the parallel-operating transformer cluster into distributed intelligent nodes, with each node performing local voltage prediction and constraint management;

[0008] S2. A distributed model predictive control framework is adopted, and the voltage regulation optimization objective is solved collaboratively by edge computing units;

[0009] S3. Inject a harmonic disturbance signal of preset amplitude into the transformer, dynamically identify the equivalent impedance based on the response data and trigger virtual impedance compensation;

[0010] S4. Establish a hierarchical collaborative control architecture to achieve clock synchronization between nodes, dynamic weight allocation, and multi-objective optimization decision-making;

[0011] S5. Drive the tap changer to perform voltage regulation operation according to the optimization instructions, and collect circulating current data in real time for closed-loop feedback.

[0012] Furthermore, the local voltage prediction described in S1 specifically includes:

[0013] A prediction model based on autoregressive moving average exogenous variable ARMAX is constructed. The input variables include historical voltage data of the current node, circulating current feedback of adjacent transformers, and load rate change gradient.

[0014] The prediction time domain is updated on a rolling basis within each control cycle to generate the voltage deviation prediction envelope for multiple future control cycles;

[0015] The prediction results are used to perform conflict detection with the tap changer action limit in the local constraint library.

[0016] Furthermore, the virtual impedance compensation described in S3 includes:

[0017] A harmonic test signal with an amplitude lower than 1% of the rated voltage is injected into the three-phase windings of the transformer, and the voltage and current response waveforms are collected.

[0018] The equivalent impedance matrix is ​​solved online by recursive least squares method, and the interphase impedance coupling relationship is identified in real time.

[0019] When the impedance deviation between any two phases exceeds the preset threshold, dynamic virtual impedance compensation is activated. The compensation amount is calculated proportionally and integrally based on the historical cumulative value of the impedance deviation and the real-time rate of change.

[0020] Furthermore, the clock synchronization described in S4 is achieved in the following way:

[0021] The master-slave clock architecture is established using the IEEE 1588v2 Precision Time Protocol (PTP), with synchronization error compensation down to the microsecond level.

[0022] In a Time-Sensitive Network (TSN), a dedicated communication channel is allocated to transmit voltage regulation commands, ensuring that end-to-end transmission jitter is less than 10% of a single control cycle;

[0023] Clock offset calibration is performed at the beginning of each control cycle, and the calibration data is embedded in the header of the control command frame.

[0024] Furthermore, the dynamic weight allocation described in S4 includes:

[0025] Real-time data collection of load rate, temperature rise, and cumulative number of operations for each transformer is used to construct a health assessment index.

[0026] Dynamic weight coefficients are generated based on the Sigmoid function, where the weight of transformers with load rates higher than the cluster average is increased by 20%-50%.

[0027] A weight matrix is ​​introduced into the global optimization objective function to give higher priority to the voltage deviation of high-weight nodes in the optimization calculation.

[0028] Furthermore, the multi-objective optimization decision described in S4 specifically refers to:

[0029] An improved non-dominated sorting genetic algorithm with an elite strategy, NSGA-II, is used to simultaneously optimize the following in the decision space: the minimum quadratic norm of voltage deviation, the suppression rate of circulating current peak, and the daily cumulative upper limit constraint of tap changer operation.

[0030] A dynamic crossover and mutation operator is introduced. When Pareto front convergence stall is detected, the mutation probability is automatically increased to break through the local optimum.

[0031] Furthermore, the voltage regulation process of the tap changer described in S5 includes:

[0032] The load switching is completed in the vacuum interrupter chamber, and the soft switching of current transfer is achieved through the IGBT module during the switching process;

[0033] After each operation, the winding impedance data is collected and compared with the virtual impedance compensation target value;

[0034] If the impedance matching error is still greater than 2% after three consecutive voltage adjustments, the local prediction model parameters will be reset.

[0035] Furthermore, the ARMAX prediction model also integrates:

[0036] Statistical characteristics of the sliding window of the tap changer's historical adjustment step size;

[0037] The distortion rates of the 3rd and 5th harmonic components in the system background harmonics;

[0038] The actual voltage regulation direction identifier of the adjacent node in the previous control cycle.

[0039] Furthermore, the proportional-integral operation employs a parameter adaptive mechanism:

[0040] When the rate of change of impedance deviation exceeds 5% / s, the integral time constant is automatically shortened to 30% of the reference value;

[0041] When the impedance deviation does not decrease for two consecutive control cycles, an enhanced compensation mode with increasing proportional coefficient is triggered.

[0042] Furthermore, it also includes digital twin verification steps:

[0043] Construct an electromagnetic-thermal coupling model of a transformer cluster in the real-time digital simulation system RTDS;

[0044] Injection includes a composite test condition involving short-circuit faults, load step changes, and mixed harmonic interference;

[0045] When the simulated circulating current value exceeds 80% of the physical system's safety threshold, the control parameter self-learning module is activated to optimize the weight allocation strategy and prediction model hyperparameters through deep reinforcement learning.

[0046] The positive effects of this invention are as follows: through the synergistic mechanism of distributed predictive control and dynamic impedance compensation, the response speed and regulation accuracy of the on-load tap changer system to grid voltage fluctuations are significantly improved, effectively solving the voltage limit exceeding problem under complex operating conditions such as new energy grid connection and load change, and ensuring the power supply quality of the power system.

[0047] The innovative hierarchical collaborative architecture eliminates the circulating current oscillation risk present in traditional centralized control when implementing multi-transformer group control. At the same time, it optimizes equipment utilization and reduces the loss of key equipment through dynamic weight allocation, providing a highly reliable solution for intelligent voltage regulation in large-scale substations. Detailed Implementation

[0048] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0049] Example

[0050] A method for on-load tap changing control of a transformer includes the following steps:

[0051] S1. Decompose the parallel-operating transformer cluster into distributed intelligent nodes, with each node performing local voltage prediction and constraint management;

[0052] S2. A distributed model predictive control framework is adopted, and the voltage regulation optimization objective is solved collaboratively by edge computing units;

[0053] S3. Inject a harmonic disturbance signal of preset amplitude into the transformer, dynamically identify the equivalent impedance based on the response data and trigger virtual impedance compensation;

[0054] S4. Establish a hierarchical collaborative control architecture to achieve clock synchronization between nodes, dynamic weight allocation, and multi-objective optimization decision-making;

[0055] S5. Drive the tap changer to perform voltage regulation operation according to the optimization instructions, and collect circulating current data in real time for closed-loop feedback.

[0056] By decomposing the parallel transformer cluster into distributed intelligent nodes, each node independently performs local voltage prediction and constraint management. Distributed Model Predictive Control (DMPC) is employed at the edge computing layer, where each node generates optimization targets based on its local predictions and uploads them to the coordinating controller for global conflict resolution. By injecting specific subharmonic disturbance signals, the equivalent impedance matrix of the transformers is identified online, triggering dynamic virtual compensation when impedance imbalance is detected. The hierarchical control architecture ensures multi-node command synchronization through clock synchronization, dynamically allocates control weights based on load rate, and ultimately drives the tap changer to perform voltage regulation and feed back circulating current data to form a closed loop.

[0057] It achieves millisecond-level dynamic response, solving the voltage over-limit problem caused by response lag in traditional methods. It eliminates the risk of circulating current oscillation in parallel transformers through dynamic impedance compensation. The hierarchical collaborative mechanism takes into account both global optimization and local fast control, improving system stability.

[0058] The local voltage prediction mentioned in S1 specifically includes:

[0059] A prediction model based on autoregressive moving average exogenous variable ARMAX is constructed. The input variables include historical voltage data of the current node, circulating current feedback of adjacent transformers, and load rate change gradient.

[0060] The prediction time domain is updated on a rolling basis within each control cycle to generate the voltage deviation prediction envelope for multiple future control cycles;

[0061] The prediction results are used to perform conflict detection with the tap changer action limit in the local constraint library.

[0062] Local voltage forecasting employs an autoregressive moving average exogenous variable (ARMAX) model. Input variables include historical voltage deviation, circulating current data from adjacent transformers, and load change rate. The model generates voltage change trends for multiple future control cycles through rolling time-domain prediction. The prediction formula is as follows:

[0063]

[0064] Among them, circulation data I cir The prediction results are obtained through real-time communication and then matched and verified against the voltage regulation number limits in the local constraint library.

[0065] The prediction model integrating multi-source data significantly improves the accuracy of voltage change trend prediction, and the rolling time-domain mechanism enhances the adaptability to load fluctuations.

[0066] The virtual impedance compensation described in S3 includes:

[0067] A harmonic test signal with an amplitude lower than 1% of the rated voltage is injected into the three-phase windings of the transformer, and the voltage and current response waveforms are collected.

[0068] The equivalent impedance matrix is ​​solved online by recursive least squares method, and the interphase impedance coupling relationship is identified in real time.

[0069] When the impedance deviation between any two phases exceeds the preset threshold, dynamic virtual impedance compensation is activated. The compensation amount is calculated proportionally and integrally based on the historical cumulative value of the impedance deviation and the real-time rate of change.

[0070] Virtual impedance compensation involves injecting low-amplitude harmonic test signals into the transformer and acquiring voltage and current response data. The equivalent impedance matrix is ​​then calculated online using the recursive least squares method.

[0071] Z eq =arg min||V meas -Z·I inj || 2

[0072] When the phase-to-phase impedance deviation exceeds the threshold, the virtual impedance value is dynamically adjusted using a proportional-integral algorithm to make the equivalent impedance tend to be balanced.

[0073] Accurate identification of impedance parameters solves the problem that traditional static models cannot adapt to changes in operating conditions, and the dynamic compensation mechanism effectively suppresses circulating current caused by impedance mismatch.

[0074] The clock synchronization described in S4 is achieved in the following way:

[0075] The master-slave clock architecture is established using the IEEE 1588v2 Precision Time Protocol (PTP), with synchronization error compensation down to the microsecond level.

[0076] In a Time-Sensitive Network (TSN), a dedicated communication channel is allocated to transmit voltage regulation commands, ensuring that end-to-end transmission jitter is less than 10% of a single control cycle;

[0077] Clock offset calibration is performed at the beginning of each control cycle, and the calibration data is embedded in the header of the control command frame.

[0078] Clock synchronization employs the Precision Time Protocol (PTP), achieving microsecond-level synchronization by embedding high-precision timestamps in control command frames and combining this with hardware clock calibration circuitry. A Time-Sensitive Network (TSN) allocates dedicated communication channels and uses flow shaping technology to ensure deterministic transmission of control commands.

[0079] Eliminate regulation conflicts caused by asynchronous control commands from multiple nodes and ensure the consistency of control timing in large-scale parallel systems.

[0080] The dynamic weight allocation described in S4 includes:

[0081] Real-time data collection of load rate, temperature rise, and cumulative number of operations for each transformer is used to construct a health assessment index.

[0082] Dynamic weight coefficients are generated based on the Sigmoid function, where the weight of transformers with load rates higher than the cluster average is increased by 20%-50%.

[0083] A weight matrix is ​​introduced into the global optimization objective function to give higher priority to voltage deviations at high-weight nodes in the optimization calculation. Dynamic weight allocation is based on real-time load rate calculations of control weight coefficients, using the Sigmoid function.

[0084]

[0085] Where L i Let L be the load factor of the i-th transformer. avg The average value is the cluster average; high-load nodes receive greater weight, and their voltage regulation needs are prioritized in optimization calculations. This prevents thermal runaway caused by overloaded transformers due to regulation lag, optimizes control resource allocation, and improves the regulation accuracy of critical nodes.

[0086] The multi-objective optimization decision-making described in S4 specifically refers to:

[0087] An improved non-dominated sorting genetic algorithm with an elite strategy (NSGA-II) is adopted to simultaneously optimize the following in the decision space: the minimum quadratic norm of voltage deviation, the suppression rate of circulating current peak, and the daily cumulative upper limit constraint of tap changer operation.

[0088] A dynamic crossover and mutation operator is introduced. When Pareto front convergence stall is detected, the mutation probability is automatically increased to break through the local optimum.

[0089] The multi-objective optimization algorithm employs an improved NSGA-II algorithm, introducing a dynamic crossover and mutation strategy: when the Pareto front converges and stalls, the mutation probability is adaptively increased.

[0090] P m =P m0 ·(1+α·Δ front )

[0091] Here, Δfront is the leading edge stagnation algebra, which achieves breakthrough of local optimum solution; and simultaneously optimizes voltage deviation, circulating current suppression and action number constraints.

[0092] To obtain the optimal balance between voltage quality and equipment lifespan, a dynamic mutation strategy is employed to avoid premature convergence issues common in traditional algorithms.

[0093] The voltage regulation process of the tap changer described in S5 includes:

[0094] The load switching is completed in the vacuum interrupter chamber, and the soft switching of current transfer is achieved through the IGBT module during the switching process;

[0095] After each operation, the winding impedance data is collected and compared with the virtual impedance compensation target value;

[0096] If the impedance matching error is still greater than 2% after three consecutive voltage adjustments, the local prediction model parameters will be reset.

[0097] When the tap changer performs voltage regulation, it achieves soft current switching through the IGBT module, completing arc-free switching within the vacuum interrupter. After each operation, the winding impedance data is collected and compared with the virtual impedance target value. If multiple consecutive matching failures occur, the prediction model parameters are reset.

[0098] Reduce arc loss, extend the mechanical life of tap changers, and ensure the effectiveness of voltage regulation operation through impedance feedback mechanism.

[0099] The ARMAX prediction model also integrates: the sliding window statistical characteristics of the tap changer's historical adjustment step size; the distortion rate of the 3rd and 5th harmonic components in the system background harmonics; and the actual voltage regulation direction identifier of the adjacent node in the previous control cycle.

[0100] The ARMAX prediction model expands its input variables by integrating the sliding window statistical features of the tap changer's historical adjustment step size. The calculation formula is as follows:

[0101]

[0102] By combining background harmonic components with the action direction identifiers of adjacent nodes, the model's adaptability to complex working conditions is enhanced; the prediction reliability under harmonic pollution scenarios is improved, and multi-node adjustment conflicts are reduced by using cooperative action identifiers.

[0103] The proportional-integral operation employs a parameter adaptive mechanism:

[0104] When the rate of change of impedance deviation exceeds 5% / s, the integral time constant is automatically shortened to 30% of the reference value;

[0105] When the impedance deviation does not decrease for two consecutive control cycles, an enhanced compensation mode with increasing proportional coefficient is triggered.

[0106] The proportional-integral parameter adaptive mechanism dynamically adjusts according to the rate of change of impedance deviation: when the deviation increases rapidly, the integral time constant is shortened to prevent overcompensation; when the deviation does not improve, the proportional coefficient is increased to strengthen the adjustment.

[0107] To avoid the contradiction between response speed and stability in fixed-parameter PI control, and to enhance robustness to sudden load changes.

[0108] It also includes digital twin verification steps:

[0109] Construct an electromagnetic-thermal coupling model of a transformer cluster in the real-time digital simulation system RTDS;

[0110] Injection includes a composite test condition involving short-circuit faults, load step changes, and mixed harmonic interference;

[0111] When the simulated circulating current value exceeds 80% of the physical system's safety threshold, the control parameter self-learning module is activated to optimize the weight allocation strategy and prediction model hyperparameters through deep reinforcement learning.

[0112] Digital twin verification constructs an electromagnetic-thermal coupling model and injects a test control strategy for complex fault conditions. When the simulated circulating current exceeds a threshold, a deep reinforcement learning algorithm is used to optimize the control parameters and update the formula:

[0113]

[0114] R(θ) is a reward function that includes voltage, circulating current, and number of actions, enabling closed-loop verification and self-optimization of the control strategy, and significantly reducing the trial-and-error risk of the physical system.

[0115] The above-described embodiments are detailed and specific, illustrating preferred embodiments of the present invention. They are only used to illustrate the technical ideas and features of the present invention, with the aim of enabling those skilled in the art to understand the content of the present invention and implement it accordingly. However, they are not limited to the present invention, and the patent scope of the present invention cannot be limited by this embodiment alone. That is, any equivalent changes or modifications made to the spirit disclosed in the present invention, without departing from the structure of the present invention, such as local improvements within the system and modifications or transformations between subsystems, are still within the patent scope of the present invention.

Claims

1. A method for on-load tap changer control of a transformer, characterized in that, Includes the following steps: S1. Decompose the parallel-operating transformer cluster into distributed intelligent nodes, with each node performing local voltage prediction and constraint management; S2. A distributed model predictive control framework is adopted, and the voltage regulation optimization objective is solved collaboratively by edge computing units; S3. Inject a harmonic disturbance signal of preset amplitude into the transformer, dynamically identify the equivalent impedance based on the response data and trigger virtual impedance compensation; S4. Establish a hierarchical collaborative control architecture to achieve clock synchronization between nodes, dynamic weight allocation, and multi-objective optimization decision-making; S5. Drive the tap changer to perform voltage regulation operation according to the optimization instructions, and collect circulating current data in real time for closed-loop feedback.

2. The on-load tap changer control method for a transformer according to claim 1, characterized in that, The local voltage prediction mentioned in S1 specifically includes: A prediction model based on autoregressive moving average exogenous variable ARMAX is constructed. The input variables include historical voltage data of the current node, circulating current feedback of adjacent transformers, and load rate change gradient. The prediction time domain is updated on a rolling basis within each control cycle to generate the voltage deviation prediction envelope for multiple future control cycles; The prediction results are used to perform conflict detection with the tap changer action limit in the local constraint library.

3. The on-load tap changer control method for a transformer according to claim 1, characterized in that, The virtual impedance compensation described in S3 includes: A harmonic test signal with an amplitude lower than 1% of the rated voltage is injected into the three-phase windings of the transformer, and the voltage and current response waveforms are collected. The equivalent impedance matrix is ​​solved online by recursive least squares method, and the interphase impedance coupling relationship is identified in real time. When the impedance deviation between any two phases exceeds the preset threshold, dynamic virtual impedance compensation is activated. The compensation amount is calculated proportionally and integrally based on the historical cumulative value of the impedance deviation and the real-time rate of change.

4. The on-load tap changer control method for a transformer according to claim 1, characterized in that, The clock synchronization described in S4 is achieved in the following way: A master-slave clock architecture is established using the IEEE 1588v2 Precision Time Protocol (PTP), with synchronization error compensation down to the microsecond level. In a Time-Sensitive Network (TSN), a dedicated communication channel is allocated to transmit voltage regulation commands, ensuring that end-to-end transmission jitter is less than 10% of a single control cycle. Clock offset calibration is performed at the beginning of each control cycle, and the calibration data is embedded in the header of the control command frame.

5. The on-load tap changer control method for a transformer according to claim 1, characterized in that, The dynamic weight allocation described in S4 includes: Real-time data collection of load rate, temperature rise, and cumulative number of operations for each transformer is used to construct a health assessment index. Dynamic weight coefficients are generated based on the Sigmoid function, where the weight of transformers with load rates higher than the cluster average is increased by 20%-50%. A weight matrix is ​​introduced into the global optimization objective function to give higher priority to the voltage deviation of high-weight nodes in the optimization calculation.

6. The on-load tap changer control method for a transformer according to claim 1, characterized in that, The multi-objective optimization decision-making described in S4 specifically refers to: An improved non-dominated sorting genetic algorithm with an elite strategy, NSGA-II, is used to simultaneously optimize the following in the decision space: the minimum quadratic norm of voltage deviation, the suppression rate of circulating current peak, and the daily cumulative upper limit constraint of tap changer operation. A dynamic crossover and mutation operator is introduced. When Pareto front convergence stall is detected, the mutation probability is automatically increased to break through the local optimum.

7. The on-load tap changer control method for a transformer according to claim 1, characterized in that, The voltage regulation process of the tap changer described in S5 includes: The load switching is completed in the vacuum interrupter chamber, and the soft switching of current transfer is achieved through the IGBT module during the switching process; After each operation, the winding impedance data is collected and compared with the virtual impedance compensation target value; If the impedance matching error is still greater than 2% after three consecutive voltage adjustments, the local prediction model parameters will be reset.

8. The on-load tap changer control method for a transformer according to claim 2, characterized in that, The ARMAX prediction model also integrates: Statistical characteristics of the sliding window of the tap changer's historical adjustment step size; The distortion rates of the 3rd and 5th harmonic components in the system background harmonics; The actual voltage regulation direction identifier of the adjacent node in the previous control cycle.

9. A transformer on-load tap changer control method according to claim 3, characterized in that, The proportional-integral operation employs a parameter adaptive mechanism: When the rate of change of impedance deviation exceeds 5% / s, the integral time constant is automatically shortened to 30% of the reference value; When the impedance deviation does not decrease for two consecutive control cycles, an enhanced compensation mode with increasing proportional coefficient is triggered.

10. A transformer on-load tap changer control method according to claim 1, characterized in that, It also includes digital twin verification steps: Construct an electromagnetic-thermal coupling model of a transformer cluster in the real-time digital simulation system RTDS; Injection includes a composite test condition involving short-circuit faults, load step changes, and mixed harmonic interference; When the simulated circulating current value exceeds 80% of the physical system's safety threshold, the control parameter self-learning module is activated to optimize the weight allocation strategy and prediction model hyperparameters through deep reinforcement learning.

Images