A series hybrid active power filter based on light-weight GRU predictive control and a harmonic treatment method thereof

Abstract

The present application relates to the technical field of power electronic harmonic control.A kind of harmonic control method of series type hybrid active power filter based on light GRU prediction control, it is characterized by including the following steps: 1), through the signal acquisition module of series connection between power grid and nonlinear load, real-time acquisition time series data of grid side harmonic current, grid voltage, and the time series data collected are preprocessed;2), using light gated recurrent unit GRU model;3) using prediction feedforward combined with proportional-resonant control algorithm;4), control series type hybrid active power filter according to the compensation voltage instruction executes harmonic compensation operation, and the compensation voltage instruction is converted into actual compensation voltage by pulse width modulation and drive unit, and is injected into power grid by coupling transformer to realize harmonic advance control.The present application has the characteristics of simple architecture, strong real-time performance and good feasibility.

Description

Technical Field

[0001] This invention relates to the field of power electronics harmonic mitigation technology, and in particular to a series hybrid active power filter based on lightweight GRU predictive control and its harmonic mitigation method. Background Technology

[0002] With the rapid development of industrial power electronics technology, nonlinear loads such as frequency converters, rectifiers, and electric arc furnaces are widely used in industrial production scenarios. These nonlinear loads inject a large amount of harmonic current into the power grid, causing a series of problems such as grid voltage distortion, power factor decline, increased equipment losses, and relay protection device malfunctions. These problems seriously affect the safe and stable operation of the power system and the service life of electrical equipment. Therefore, grid harmonic mitigation has become an important research direction in the field of power electronics.

[0003] Series-connected hybrid active power filters (SHAPFs) are one of the mainstream harmonic mitigation devices in medium- and high-voltage power grids. They combine the high-precision, flexible compensation characteristics of active power filters with the low-cost, low-loss advantages of passive filters. By connecting in series between the power grid and the nonlinear load, they can effectively suppress harmonic voltage distortion in the power grid and mitigate harmonic currents generated by the load. Compared to parallel-connected active power filters, they are more suitable for the harmonic mitigation needs of medium- and high-voltage power grids and have broad application prospects in industrial production and power transmission. However, traditional series-connected SHAPFs generally adopt a passive control mode of "detect first, compensate later." That is, they first acquire power grid harmonic data through a sampling unit, then generate compensation commands through control calculations, and finally drive power devices to perform compensation. The entire process suffers from inherent control lag caused by sampling, calculation, and power device switching. Since power harmonics have nonlinear and non-stationary time-series characteristics, this control lag directly leads to the compensation commands being out of sync with real-time harmonics, resulting in decreased compensation accuracy. Especially in scenarios with rapid load fluctuations, the harmonic mitigation effect will be significantly reduced, making it difficult to meet the requirements of high-quality power grid operation.

[0004] In existing technologies, some research has attempted to introduce artificial intelligence (AI) to address the lag problem in harmonic mitigation. However, most existing solutions are designed for parallel active power filter clusters and generally employ complex architectures involving multi-source data fusion, long-period trend prediction, and reinforcement learning scheduling. On the one hand, this requires the collection of multi-dimensional non-electrical data such as equipment operation and business scenarios, resulting in high deployment costs and significant data processing difficulties. On the other hand, long-period trend prediction cannot meet the millisecond-level real-time control requirements of series-type hybrid active power filters, and the complex algorithm architecture is difficult to implement in embedded controllers, making it unsuitable for direct application in engineering scenarios of series-type hybrid active power filters. Furthermore, existing harmonic prediction schemes based on Long Short-Term Memory (LSTM) networks suffer from drawbacks such as large parameter count, high computational complexity, and slow inference speed in embedded deployment due to the multiple gating structures in the LSTM model. These shortcomings make it difficult to meet the stringent real-time control requirements of series-type hybrid active power filters, thus limiting the application of AI in these filters. Currently, there is no harmonic mitigation solution specifically designed for series-type hybrid active power filters, based on lightweight GRU short-time prediction for advanced compensation, and capable of direct embedded deployment, making it difficult to effectively address the pain points of traditional solutions. Summary of the Invention

[0005] The purpose of this invention is to provide a series hybrid active power filter based on lightweight GRU predictive control and its harmonic mitigation method, which features a simple architecture, strong real-time performance, and good feasibility.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: On one hand, the present invention provides a harmonic mitigation method for a series hybrid active power filter based on lightweight GRU predictive control, characterized by comprising the following steps: 1) By connecting the signal acquisition module between the power grid and the nonlinear load in series, the time-series data of harmonic current and voltage on the power grid side are collected in real time, and the collected time-series data are preprocessed. 2) Using the lightweight gated cyclic unit (GRU) model, based on the preprocessed harmonic timing data, predict the dynamic change trend of harmonic current amplitude and phase within the control window of the next 1-5ms, and output the corresponding harmonic prediction value. 3) Using a predictive feedforward combined with a quasi-proportional resonant control algorithm, compensation voltage commands for a series hybrid active power filter are generated based on the predicted harmonic current amplitude and phase. 4) The series-connected hybrid active power filter performs harmonic compensation operation according to the compensation voltage command. The compensation voltage command is converted into the actual compensation voltage through the pulse width modulation and drive unit, and injected into the power grid through the coupling transformer to achieve harmonic advance control.

[0007] Further, in step 1), a signal acquisition module connected in series between the power grid and the nonlinear load is used to collect real-time time-series data of harmonic current and voltage on the power grid side, and the collected time-series data is preprocessed, specifically including the following steps: (1) By deploying Hall current sensors and Hall voltage sensors on the series line between the power grid and the nonlinear load, instantaneous waveform data of three-phase harmonic current and three-phase power grid voltage on the power grid side are collected in real time, and the sampling frequency is set to 10kHz~20kHz. (2) Perform automated anomaly detection and missing value processing on the collected raw time series data, remove high-frequency noise and outlier data in the sampling process, and generate a clean time series dataset; (3) Perform MinMax normalization on the clean time series dataset to scale the data to the [0,1] interval and eliminate the influence of the units on the model prediction accuracy.

[0008] Further, in step 2), a lightweight gated recurrent unit (GRU) model is used to predict the dynamic change trend of harmonic current amplitude and phase within the next 1-5ms control window based on the preprocessed harmonic timing data, and the corresponding harmonic prediction value is output. This specifically includes the following steps: (1) Organize the preprocessed harmonic time series data into a time series in chronological order and arrange it into an input format of [number of samples, time step, number of features], where the time step is set to 20 and the number of features is 1. The time series contains time series data used to characterize the amplitude and phase information of harmonic current, for example, it can be an orthogonal component sequence obtained after coordinate transformation. (2) Input the cleaned univariate time series into the lightweight GRU model, and learn the short-term time series dependency of harmonic time series data through the update gate and reset gate structure in the model; (3) Based on the learned time-series dependency, the forward propagation is used to calculate and output the predicted values ​​of harmonic current amplitude and phase within the control window of 1~5ms in the future. The predicted values ​​are then processed by inverse normalization to optimize the output data and output the final harmonic prediction parameters.

[0009] Further, in step 3), a predictive feedforward combined with a quasi-proportional resonant control algorithm is used to generate a compensation voltage command for the series-type hybrid active power filter based on the predicted harmonic current amplitude and phase. This specifically includes the following steps: (1) The harmonic prediction value output by the GRU model is used as the feedforward quantity and superimposed with the output of the quasi-proportional resonant controller to form a composite control combining prediction feedforward and feedback closed loop, so as to offset the control lag caused by sampling, calculation and switching action. (2) Based on the grid voltage synchronization signal, the amplitude and phase of the target compensation voltage are calculated by combining the harmonic prediction value. The quasi-proportional resonant controller is used to realize the zero steady-state error tracking of the main characteristic subharmonics such as the 3rd, 5th and 7th harmonics, and generate the instantaneous compensation voltage command signal in the time domain. (3) Verify the validity of the compensation voltage command to determine that the compensation voltage command does not exceed the DC side voltage limit of the series hybrid active power filter and the rated capacity of the IGBT device, and output the final compensation voltage command.

[0010] Further, in step 4), the series-connected hybrid active power filter performs harmonic compensation operation according to the compensation voltage command. The compensation voltage command is converted into an actual compensation voltage through pulse width modulation and a drive unit, and then injected into the power grid through a coupling transformer to achieve advanced harmonic mitigation. Specifically, this includes the following steps: (1) Input the compensation voltage command into the pulse width modulation controller, compare the compensation voltage command with the actual output voltage feedback signal of the series hybrid active power filter, calculate the error signal and generate a pulse width modulation waveform with a fixed switching frequency. (2) The pulse width modulation waveform is amplified and electrically isolated by the driving circuit to generate a gate drive signal, and the gate drive signal is applied to the main circuit of the series hybrid active power filter to control the turn-on and turn-off actions of the insulated gate bipolar transistor IGBT switching device, and convert the electrical energy stored in the DC side capacitor into the actual compensation voltage. (3) By connecting the coupling transformer in series with the power grid line, the generated compensation voltage is injected into the power grid to offset the harmonic voltage distortion in the power grid. In conjunction with the passive filter branch, fixed harmonics are filtered out to complete the harmonic advance treatment.

[0011] Furthermore, in step 2), the lightweight GRU model adopts a dual hidden layer structure, with the number of nodes in both the first and second hidden layers being 32 to 64. Compared to the Long Short-Term Memory (LSTM) network model with similar performance, the GRU model has the advantages of significantly reduced parameter quantity and fast inference speed due to the simplified gating structure. The model is trained offline using harmonic time series data measured in industrial scenarios. After training, it is exported in ONNX format and deployed on embedded controllers such as DSPTMS320F28335. The single-step inference time is controlled within 0.5ms, which meets the real-time control requirements of power electronics.

[0012] The harmonic mitigation method for a series hybrid active power filter based on lightweight GRU predictive control described above, after step 4), also includes the following steps: (1) Periodically acquire grid harmonic monitoring data and real-time operation data of series hybrid active power filter, and calculate harmonic mitigation effect index based on the acquired data. The harmonic mitigation effect index includes total grid harmonic distortion rate, single harmonic suppression rate, and voltage distortion improvement rate. (2) Based on the harmonic control effect index, the GRU model is trained incrementally once every fixed control period using the latest collected harmonic time series data, and the model weight parameters are dynamically updated. (3) Adjust the harmonic prediction output according to the updated GRU model and optimize the compensation voltage command in sync to ensure the stability of the harmonic mitigation effect under load fluctuation scenarios.

[0013] On the other hand, the present invention also provides a series hybrid active power filter based on lightweight GRU predictive control, characterized in that it includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the harmonic control method of the series hybrid active power filter based on lightweight GRU predictive control described above.

[0014] On the other hand, the present invention also provides a series hybrid active power filter based on lightweight GRU predictive control, including a harmonic signal acquisition module, a GRU timing prediction module, a predictive feedforward control module, a drive circuit, and a series hybrid active power filter main circuit connected in sequence. The harmonic signal acquisition module includes a Hall current sensor, a Hall voltage sensor, and a high-speed AD sampling unit, which are connected in series between the power grid and the nonlinear load to acquire real-time time-series data of harmonic current and grid voltage on the power grid side. The GRU timing prediction module includes a data preprocessing unit and a lightweight GRU prediction model, which is used to process harmonic timing data and output the predicted values ​​of harmonic current amplitude and phase for the next 1 to 5 ms. The predictive feedforward control module employs a predictive feedforward combined with a quasi-proportional resonant control algorithm to generate compensation voltage commands based on the predicted harmonic values. The main circuit of the series-type hybrid active power filter includes a series active filter unit and a passive filter branch. The series active filter unit is connected in series to the power grid line through a coupling transformer. The passive filter branch is connected in parallel with the series active filter unit and is used to filter out fixed harmonics. The driving circuit is connected between the predictive feedforward control module and the series active filter unit, and is used to amplify the compensation voltage command to drive the IGBT device to turn on and off.

[0015] On the other hand, the present invention also provides a non-volatile computer-readable storage medium storing computer-executable instructions, which, when executed, implement the above-described harmonic mitigation method for a series hybrid active power filter based on lightweight GRU predictive control.

[0016] The beneficial effects of this invention are: 1. It completely solves the inherent control lag problem of traditional series-type hybrid active power filters. By using a lightweight GRU model to make short-term advance predictions of harmonic timing data of 1~5ms, it upgrades the traditional passive control mode of "detection-compensation" to an active advance control mode of "prediction-compensation". This effectively offsets the control delay caused by sampling, calculation and switching actions, greatly improves the accuracy of harmonic compensation and the adaptability to load fluctuations, and realizes the fundamental transformation of harmonic management from lag compensation to advance intervention, ensuring the stability and safety of power grid operation.

[0017] 2. The model boasts significant advantages in lightweight design and enhanced real-time embedded performance. It employs a lightweight GRU model with only two gate structures: update gate and reset gate. Compared to the LSTM model with the same structure, it has the advantages of fewer parameters and lower computational complexity. The single-step inference time can be controlled within 0.5ms. While ensuring the accuracy of harmonic prediction, it significantly reduces the computational load of the model, making it more suitable for the real-time control requirements of embedded controllers such as DSPs. It does not require host computer or server support, making engineering implementation easier and deployment costs lower, and facilitating large-scale application in industrial settings.

[0018] 3. The solution features a simple and highly targeted architecture, specifically designed for the topology characteristics of series-type hybrid active power filters. It uses only single-variable time-series data of grid harmonic current and voltage as model input, eliminating the need for multi-source heterogeneous data acquisition and fusion, thus avoiding complex data processing procedures and reducing data acquisition and processing costs. Simultaneously, it employs a composite control algorithm combining predictive feedforward and quasi-proportional resonance, enabling both advance compensation through GRU prediction and zero-steady-state-error tracking of characteristic subharmonics through the quasi-proportional resonance controller. Combined with low-cost harmonic suppression by passive filter branches, it stably meets national standards for grid harmonic mitigation, balancing mitigation effectiveness and economy.

[0019] 4. The solution has good scalability and adaptability. The GRU model weights can be dynamically updated through the incremental training mechanism to adapt to the harmonic characteristic changes under different operating conditions and avoid the decay of the mitigation effect due to fluctuations in operating conditions. At the same time, the model structure, prediction window and control parameters can be flexibly adjusted according to the actual application scenario. It takes into account the application requirements of series hybrid active power filters of different voltage levels such as low voltage and medium and high voltage, and is suitable for the harmonic mitigation needs of various nonlinear load scenarios such as industrial frequency converters, rectifiers, and electric arc furnaces. It has a wide range of applications. Attached Figure Description

[0020] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings: Figure 1 A schematic flowchart of a harmonic mitigation method for a series hybrid active power filter based on lightweight GRU predictive control, provided for an embodiment of the present invention; Figure 2 This is a structural block diagram of a series hybrid active power filter based on lightweight GRU predictive control, provided for an embodiment of the present invention.

[0021] Figure 3 This is a schematic diagram of the internal structure of a series hybrid active power filter device based on lightweight GRU predictive control, provided for an embodiment of the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0023] The technical solutions provided by the various embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0024] Figure 1 This is a schematic flowchart illustrating a harmonic mitigation method for a series hybrid active power filter based on lightweight GRU predictive control, provided as an embodiment of the present invention. Figure 1 As shown, the harmonic mitigation method provided in this embodiment of the invention mainly includes the following steps: Step S101: By connecting the signal acquisition module between the power grid and the nonlinear load in series, the time-series data of harmonic current and voltage on the power grid side are acquired in real time, and the acquired time-series data is preprocessed. Step S102: Using the lightweight gated cyclic unit (GRU) model, based on the preprocessed harmonic timing data, predict the dynamic change trend of harmonic current amplitude and phase within the control window of the next 1~5ms, and output the corresponding harmonic prediction value. Step S103: Using a predictive feedforward combined with a quasi-proportional resonant control algorithm, a compensation voltage command for a series hybrid active power filter is generated based on the predicted harmonic current amplitude and phase. Step S104: Control the series-type hybrid active power filter to perform harmonic compensation operation according to the compensation voltage command. Convert the compensation voltage command into the actual compensation voltage through the pulse width modulation and drive unit, and inject it into the power grid through the coupling transformer to achieve harmonic advance control.

[0025] The harmonic mitigation method involved in this invention is based on a four-layer closed-loop architecture design of "acquisition-prediction-control-execution". All functional modules are integrated into the local controller of a single series-type hybrid active power filter, without the need for additional host computer or multi-device collaboration. It has a simple structure, high integration, and is fully adapted to the embedded deployment requirements of industrial sites, which can effectively reduce deployment costs and maintenance difficulty.

[0026] In one embodiment of the present invention, in step S101, the signal acquisition module is deployed on the series line between the power grid and the nonlinear load. It can accurately capture the instantaneous changes in harmonics on the power grid side. For typical nonlinear load scenarios such as industrial frequency converters and rectifiers, the sampling frequency is set to 10kHz, which can accurately capture the instantaneous waveforms of the 3rd, 5th, 7th, and 11th harmonics, providing high-quality data support for subsequent prediction and control. For example, in a frequency converter drive scenario in an industrial production line, Hall current sensors and voltage sensors are directly installed on the series power grid line on the input side of the frequency converter. They can continuously capture the instantaneous waveform data of three-phase current and voltage. After the analog signals are converted into digital signals by a high-speed AD sampling unit, they are transmitted to the local DSP controller, ensuring the real-time performance and accuracy of data transmission.

[0027] During data preprocessing, outlier data exceeding the device's measurement range is first removed using a limiting algorithm to avoid the impact of abnormal data on the model's prediction accuracy. Then, a moving average filter is used to remove high-frequency noise during the sampling process to ensure data cleanliness. Finally, the MinMaxScaler algorithm is used to normalize all data to the [0,1] interval, providing standardized input data for subsequent GRU model predictions. This effectively eliminates the impact of dimensional differences on the model's prediction accuracy and improves the model's generalization ability.

[0028] In one embodiment of the present invention, in step S102, the preprocessed harmonic time series data is used to construct a univariate time series according to the format of the data of the 20 historical sampling times as input and the data of the next sampling time as output. The input format is [number of samples, 20, 1]. Only the harmonic current amplitude is used as the only input feature of the model. No additional feature dimension is required, which greatly reduces the computational load of the model and further improves the inference speed of the model, adapting to the needs of embedded real-time control.

[0029] The lightweight GRU model employs a dual-hidden-layer structure. Both the first and second hidden layers have a reasonable number of nodes and contain only two gating structures: update gates and reset gates. Compared to an LSTM with the same structure, this reduces the complexity of the gating units and significantly lowers the number of parameters, greatly improving inference speed while maintaining prediction accuracy. The input layer corresponds to time-series data over 20 time steps, and the output layer corresponds to the predicted amplitude and phase values ​​of harmonic currents in the next 2ms, accurately capturing the short-term dynamic trends of harmonics. The model is pre-trained offline using massive amounts of harmonic time-series data collected from industrial sites. The optimizer uses the Adam algorithm, and the loss function uses mean squared error (MSE) to ensure high prediction accuracy. After training, the model weights and parameters are fixed and exported in ONNX format, allowing for rapid porting to DSP embedded controllers for local real-time inference, meeting the real-time requirements of power electronic control.

[0030] For example, in scenarios where the inverter load frequency changes rapidly, the GRU model can accurately predict the trend of harmonic current amplitude change in the next 2ms using harmonic data from 20 historical sampling moments. This provides the control loop with an advanced compensation reference, fundamentally offsetting the compensation asynchrony problem caused by control lag and ensuring the stability of harmonic mitigation effects.

[0031] In one embodiment of the present invention, in step S103, the harmonic prediction value output by the GRU model is used as a feedforward quantity and directly superimposed on the output of the quasi-proportional resonant controller to form a composite control structure of "predictive feedforward + feedback closed loop" to achieve complementary advantages. Specifically, the predictive feedforward channel is responsible for generating an advanced compensation quantity based on the harmonic prediction value, effectively offsetting the inherent lag of the system; the quasi-proportional resonant feedback loop is responsible for tracking the residual harmonics without steady-state error, further improving the compensation accuracy and ensuring that the harmonic mitigation effect meets national standards.

[0032] The quasi-proportional resonant controller is designed to address the 3rd, 5th, and 7th harmonics, achieving high open-loop gain at the corresponding harmonic frequencies. This enables zero steady-state error compensation for specific harmonics, effectively suppressing their impact on the power grid. Based on the grid voltage's phase-locked loop synchronization signal and harmonic prediction values, the amplitude and phase of the target compensation voltage are calculated. A composite control loop generates a time-domain instantaneous compensation voltage command, which is simultaneously limited and verified to ensure it does not exceed the IGBT device's rated capacity and DC-side voltage output capability. This prevents overcurrent and overvoltage faults, improving system reliability and safety.

[0033] In one embodiment of the present invention, in step S104, the generated compensation voltage command is input to the pulse width modulation (PWM) controller. A unipolar frequency doubling modulation method is used to generate a PWM waveform corresponding to the switching frequency. After power amplification and electrical isolation by the drive circuit, the IGBT device of the series active filter unit is driven to switch on and off at high speed, converting the DC energy of the DC-side capacitor into an AC compensation voltage consistent with the compensation command. This results in high conversion efficiency and fast response speed. Finally, the generated compensation voltage is injected into the power grid through a coupling transformer connected in series with the grid, generating a compensation voltage equal in magnitude but opposite in phase to the grid harmonic voltage. This effectively cancels out harmonic distortion in the grid. Simultaneously, the passive filter branch shunts and filters out fixed harmonics such as the 3rd and 5th orders. The two work synergistically to achieve efficient and proactive harmonic mitigation, further improving the harmonic mitigation effect.

[0034] For example, in one embodiment of the present invention, the system also sets up an anomaly protection mechanism. When the GRU model prediction error is detected to exceed a preset threshold, it automatically and seamlessly switches to the traditional quasi-proportional resonant control mode, suspends the incremental update of the model, and automatically resumes the prediction control mode after the harmonic data stabilizes. This avoids the failure of harmonic mitigation caused by abnormal model prediction, significantly improves the robustness and reliability of the system operation, and ensures that the system can operate stably for a long time.

[0035] Figure 2 A structural block diagram of a series hybrid active power filter based on lightweight GRU predictive control is provided for an embodiment of the present invention, as shown below. Figure 2 As shown, the filter mainly includes a harmonic signal acquisition module, a GRU timing prediction module, a prediction feedforward control module, a drive circuit, and a series-type hybrid active power filter main circuit. Each module is connected in sequence, with a compact structure and strong synergy. The main circuit is connected in series between the power grid and the nonlinear load through a coupling transformer, which can be directly adapted to the installation requirements of industrial sites.

[0036] Figure 3 A schematic diagram of the internal structure of a series hybrid active power filter (device) based on lightweight GRU predictive control, provided for an embodiment of the present invention, is shown below. Figure 3 As shown, the device includes at least one processor and a memory communicatively connected to the processor. The memory stores instructions executable by the processor. When the instructions are executed by the processor, the aforementioned harmonic mitigation method is implemented. The processor and the memory communicate via a bus. The device may also include input / output interfaces, network interfaces, and other components to adapt to the communication and control needs of industrial sites, facilitating remote monitoring and parameter adjustment of the device.

[0037] The various embodiments in this invention are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the device and medium embodiments are relatively simple in description because they are fundamentally similar to the method embodiments; relevant parts can be referred to the descriptions in the method embodiments.

[0038] The devices, media, and methods provided in the embodiments of the present invention are one-to-one correspondences. Therefore, the devices and media also have similar beneficial technical effects as their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the devices and media will not be repeated here.

[0039] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0040] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

Claims

1. A harmonic mitigation method for a series-type hybrid active power filter based on lightweight GRU predictive control, characterized in that... Includes the following steps: 1) By connecting the signal acquisition module between the power grid and the nonlinear load in series, the time-series data of harmonic current and voltage on the power grid side are collected in real time, and the collected time-series data are preprocessed. 2) Using the lightweight gated cyclic unit (GRU) model, based on the preprocessed harmonic timing data, predict the dynamic change trend of harmonic current amplitude and phase within the control window of the next 1-5ms, and output the corresponding harmonic prediction value. 3) Using a predictive feedforward combined with a quasi-proportional resonant control algorithm, compensation voltage commands for a series hybrid active power filter are generated based on the predicted harmonic current amplitude and phase. 4) The series-connected hybrid active power filter performs harmonic compensation operation according to the compensation voltage command. The compensation voltage command is converted into the actual compensation voltage through the pulse width modulation and drive unit, and injected into the power grid through the coupling transformer to achieve harmonic advance control.

2. The harmonic mitigation method for a series hybrid active power filter based on lightweight GRU predictive control according to claim 1, characterized in that... In step 1), a signal acquisition module connected in series between the power grid and the nonlinear load is used to collect real-time time-series data of harmonic current and voltage on the power grid side. The collected time-series data is then preprocessed, specifically including the following steps: (1) By deploying Hall current sensors and Hall voltage sensors on the series line between the power grid and the nonlinear load, instantaneous waveform data of three-phase harmonic current and three-phase power grid voltage on the power grid side are collected in real time, and the sampling frequency is set to 10kHz~20kHz. (2) Perform automated anomaly detection and missing value processing on the collected raw time series data, remove high-frequency noise and outlier data in the sampling process, and generate a clean time series dataset; (3) Perform MinMax normalization on the clean time series dataset to scale the data to the [0,1] interval and eliminate the influence of the units on the model prediction accuracy.

3. The harmonic mitigation method for a series hybrid active power filter based on lightweight GRU predictive control according to claim 1, characterized in that... In step 2), a lightweight gated recurrent unit (GRU) model is used to predict the dynamic change trend of harmonic current amplitude and phase within the next 1-5ms control window based on the preprocessed harmonic timing data, and the corresponding harmonic prediction value is output. This includes the following steps: (1) Organize the preprocessed harmonic time series data into a time series in chronological order and arrange it into an input format of [number of samples, time step, number of features], where the time step is set to 20 and the number of features is 1. The time series contains time series data used to characterize the amplitude and phase information of harmonic current, for example, it can be an orthogonal component sequence obtained after coordinate transformation. (2) Input the cleaned univariate time series into the lightweight GRU model, and learn the short-term time series dependency of harmonic time series data through the update gate and reset gate structure in the model; (3) Based on the learned time-series dependency, the forward propagation is used to calculate and output the predicted values ​​of harmonic current amplitude and phase within the control window of 1~5ms in the future. The predicted values ​​are then processed by inverse normalization to optimize the output data and output the final harmonic prediction parameters.

4. The harmonic mitigation method for a series hybrid active power filter based on lightweight GRU predictive control according to claim 1, characterized in that... In step 3), a predictive feedforward combined with a quasi-proportional resonant control algorithm is used to generate compensation voltage commands for the series-type hybrid active power filter based on the predicted harmonic current amplitude and phase. Specifically, this includes the following steps: (1) The harmonic prediction value output by the GRU model is used as the feedforward quantity and superimposed with the output of the quasi-proportional resonant controller to form a composite control combining prediction feedforward and feedback closed loop, so as to offset the control lag caused by sampling, calculation and switching action. (2) Based on the grid voltage synchronization signal, the amplitude and phase of the target compensation voltage are calculated by combining the harmonic prediction value. The quasi-proportional resonant controller is used to realize the zero steady-state error tracking of the main characteristic subharmonics such as the 3rd, 5th and 7th harmonics, and generate the instantaneous compensation voltage command signal in the time domain. (3) Verify the validity of the compensation voltage command to determine that the compensation voltage command does not exceed the DC side voltage limit of the series hybrid active power filter and the rated capacity of the IGBT device, and output the final compensation voltage command.

5. The harmonic mitigation method for a series hybrid active power filter based on lightweight GRU predictive control according to claim 1, characterized in that... In step 4), the series-connected hybrid active power filter performs harmonic compensation operation according to the compensation voltage command. The compensation voltage command is converted into an actual compensation voltage through pulse width modulation and a drive unit, and then injected into the power grid through a coupling transformer to achieve advanced harmonic mitigation. Specifically, this includes the following steps: (1) Input the compensation voltage command into the pulse width modulation controller, compare the compensation voltage command with the actual output voltage feedback signal of the series hybrid active power filter, calculate the error signal and generate a pulse width modulation waveform with a fixed switching frequency. (2) The pulse width modulation waveform is amplified and electrically isolated by the driving circuit to generate a gate drive signal, and the gate drive signal is applied to the main circuit of the series hybrid active power filter to control the turn-on and turn-off actions of the insulated gate bipolar transistor IGBT switching device, and convert the electrical energy stored in the DC side capacitor into the actual compensation voltage. (3) By connecting the coupling transformer in series with the power grid line, the generated compensation voltage is injected into the power grid to offset the harmonic voltage distortion in the power grid. In conjunction with the passive filter branch, fixed harmonics are filtered out to complete the harmonic advance treatment.

6. The harmonic mitigation method for a series hybrid active power filter based on lightweight GRU predictive control according to claim 3, characterized in that... In step 2), the lightweight GRU model adopts a dual hidden layer structure, with the number of nodes in the first hidden layer and the second hidden layer both being 32 to 64. Compared with long short-term memory network models with similar performance, the GRU model has the advantages of significantly reduced parameter quantity and fast inference speed due to the simplified gating structure. The model is trained offline using measured harmonic timing data in industrial scenarios. After training, it is exported in ONNX format and deployed on a DSPTMS320F28335 embedded controller. The single-step inference time is controlled within 0.5ms, which meets the real-time control requirements of power electronics.

7. The harmonic mitigation method for a series hybrid active power filter based on lightweight GRU predictive control according to claim 1, characterized in that... Following step 4), the following steps are also included: (1) Periodically acquire grid harmonic monitoring data and real-time operation data of series hybrid active power filter, and calculate harmonic mitigation effect index based on the acquired data. The harmonic mitigation effect index includes total grid harmonic distortion rate, single harmonic suppression rate, and voltage distortion improvement rate. (2) Based on the harmonic control effect index, the GRU model is trained incrementally once every fixed control period using the latest collected harmonic time series data, and the model weight parameters are dynamically updated. (3) Adjust the harmonic prediction output according to the updated GRU model and optimize the compensation voltage command in sync to ensure the stability of the harmonic mitigation effect under load fluctuation scenarios.

8. A series hybrid active power filter based on lightweight GRU predictive control, implementing the method of claim 1, characterized in that... include: At least one processor; And a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the above-described harmonic mitigation method for a series hybrid active power filter based on lightweight GRU predictive control.

9. A series hybrid active power filter based on lightweight GRU predictive control, implementing the method of claim 1, characterized in that... A series-type hybrid active power filter based on lightweight GRU predictive control includes a harmonic signal acquisition module, a GRU timing prediction module, a predictive feedforward control module, a drive circuit, and a series-type hybrid active power filter main circuit connected in sequence. The harmonic signal acquisition module includes a Hall current sensor, a Hall voltage sensor, and a high-speed AD sampling unit, which are connected in series between the power grid and the nonlinear load to acquire real-time time-series data of harmonic current and grid voltage on the power grid side. The GRU timing prediction module includes a data preprocessing unit and a lightweight GRU prediction model, which is used to process harmonic timing data and output the predicted values ​​of harmonic current amplitude and phase for the next 1 to 5 ms. The predictive feedforward control module employs a predictive feedforward combined with a quasi-proportional resonant control algorithm to generate compensation voltage commands based on the predicted harmonic values. The main circuit of the series-type hybrid active power filter includes a series active filter unit and a passive filter branch. The series active filter unit is connected in series to the power grid line through a coupling transformer. The passive filter branch is connected in parallel with the series active filter unit and is used to filter out fixed harmonics. The driving circuit is connected between the predictive feedforward control module and the series active filter unit, and is used to amplify the compensation voltage command to drive the IGBT device to turn on and off.

Images