A zero sequence compensation system of a three-phase four-bridge-arm power quality treatment device

Abstract

The application relates to the technical field of zero-sequence voltage compensation, and discloses a zero-sequence compensation system of a three-phase four-bridge-arm power quality treatment device, which comprises a synchronous sampling module, a first electrical parameter is monitored, and a second electrical parameter is collected after a zero-sequence compensation trigger mechanism starts a zero-sequence compensation process; a data prediction module is used for calculating future load current variation and harmonic components; a compensation calculation module is used for calculating compensation voltage and compensation current; a compensation execution module is used for outputting compensation power through a power electronic compensation unit; and a closed-loop optimization module is used for dynamically optimizing compensation parameters; through the multi-module collaborative architecture, the application realizes the organic combination of fast response, high-precision compensation, intelligent prediction and adaptive control, is especially suitable for application scenarios, such as data centers, which have extremely high requirements for power quality, and compared with the prior art, the application has significant improvement in response speed, compensation precision and intelligent degree, and provides a complete technical solution for power quality treatment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of zero sequence voltage compensation, and more particularly to a zero sequence compensation system of a three-phase four-bridge-arm power quality treatment device. BACKGROUND

[0002] With the rapid development of cloud computing, big data and artificial intelligence technologies, large data centers have become an important infrastructure of modern information society. A large number of server cabinets, storage devices, network devices and other IT loads are deployed in data centers, which have characteristics such as high power density, frequent load changes, and large transient impact when starting and stopping. Especially under the background of the wide application of virtualization and containerization technologies, the load of the data center presents a highly dynamic feature, and the load power can be switched from a low load state to a full load state within milliseconds, or quickly migrated between different server cabinets.

[0003] Such dynamic load characteristics pose a serious challenge to the power supply system of the data center. Frequent load changes can cause serious three-phase current imbalance, generate a large amount of zero sequence current, and form a large zero sequence voltage in the neutral line, affecting the normal operation of IT devices, and even causing device failure or data loss. The nonlinear characteristics of high-power IT devices can generate a large number of harmonic components, further deteriorating power quality. Data centers have very high requirements for power supply reliability, and any power quality problem can cause significant economic losses.

[0004] The Chinese patent with the granted publication number CN107026454B discloses a zero sequence voltage compensation device and a power distribution system. The zero sequence voltage compensation device is applied to the power distribution system. The power distribution system includes a three-phase power supply, a load and a neutral line, and the zero sequence voltage compensation device includes a voltage compensation circuit for outputting a compensation voltage to the power distribution system and a controller for controlling the phase and amplitude of the compensation voltage according to the current working voltage and working current of the power distribution system. The input end of the voltage compensation circuit and the first end of the load are connected with the three-phase power supply respectively, the first output end and the second output end of the voltage compensation circuit are connected with the second end of the load and the neutral line respectively, the input end of the controller is connected with the first end of the load, and the output end is connected with the control end of the voltage compensation circuit.

[0005] However, the prior art has obvious limitations in dealing with high-frequency dynamic load scenarios of data centers. Since the data center load can switch from low load to full load in milliseconds, the controller sampling frequency and voltage compensation circuit bandwidth design of the traditional compensation device are difficult to match such rapid changes, resulting in a lag in compensation response. When the load migrates quickly or a transient impact occurs, the phase and amplitude adjustment of the compensation voltage cannot follow the instantaneous fluctuations of the zero sequence voltage in real time, causing the neutral line voltage suppression effect to decline, and still causing the risk of unstable power supply to IT equipment. The lack of dynamic response capability makes it difficult for the existing technology to fully exert the compensation performance in complex load conditions under virtualization environment. SUMMARY

[0006] The purpose of the present application is to provide a zero sequence compensation system of a three-phase four-bridge-arm power quality treatment device to solve the above problems.

[0007] The present application provides a zero sequence compensation system of a three-phase four-bridge-arm power quality treatment device, comprising:

[0008] A synchronous sampling module monitors a first electrical parameter and establishes a zero sequence compensation trigger mechanism based on the first electrical parameter. After the zero sequence compensation trigger mechanism starts the zero sequence compensation process, a second electrical parameter is collected based on a preset sampling period.

[0009] A data prediction module calculates a zero sequence parameter after denoising the second electrical parameter. A load prediction model calculates a future load current change based on the zero sequence parameter, and a harmonic analysis model calculates a harmonic component based on the zero sequence parameter.

[0010] A compensation calculation module calculates a compensation voltage and a compensation current through an adaptive PID controller based on the zero sequence parameter, the load current change, and the harmonic component.

[0011] A compensation execution module converts the compensation voltage and the compensation current into control signals and outputs compensation power through a power electronic compensation unit.

[0012] A closed-loop optimization module collects parameters after compensation, compares them with preset compensation target values to obtain deviation information, and dynamically optimizes compensation parameters based on the deviation information.

[0013] Further, the zero sequence compensation trigger mechanism comprises:

[0014] When the zero sequence compensation trigger mechanism detects any abnormal state, the zero sequence compensation process is started. The abnormal state includes three-phase current imbalance abnormality, neutral line zero sequence current exceeding, zero sequence voltage offset abnormality, and harmonic distortion rate abnormality. Wherein:

[0015] Three-phase current imbalance includes: when the absolute value of the difference between phase A current and phase B current is greater than or equal to the preset three-phase current imbalance threshold, or when the absolute value of the difference between phase B current and phase C current is greater than or equal to the preset three-phase current imbalance threshold, or when the absolute value of the difference between phase C current and phase A current is greater than or equal to the preset three-phase current imbalance threshold.

[0016] The neutral zero-sequence current exceeds the limit when the neutral zero-sequence current is greater than or equal to the preset zero-sequence current threshold.

[0017] Zero-sequence voltage offset anomaly includes: when the zero-sequence voltage is greater than or equal to the preset zero-sequence voltage offset threshold, it indicates that the neutral point potential of the system has shifted;

[0018] Harmonic distortion rate anomalies include: when any harmonic amplitude value in the preset interval is greater than or equal to the preset harmonic distortion threshold.

[0019] Furthermore, the first electrical parameter includes three-phase current, neutral zero-sequence current, zero-sequence voltage, and harmonics; the second electrical parameter includes load power, instantaneous phase voltage of the three-phase bus, and instantaneous current of the three-phase load. The instantaneous phase voltage of the three-phase bus includes the instantaneous phase voltage of phase A, phase B, and phase C, and the instantaneous current of the three-phase load includes the instantaneous load current of phase A, phase B, and phase C.

[0020] Furthermore, the zero-sequence parameters include zero-sequence voltage, zero-sequence current, and load power change rate. The zero-sequence voltage is the sum of the instantaneous phase voltages of phase A, phase B, and phase C divided by 3. The zero-sequence current is the sum of the instantaneous load currents of phase A, phase B, and phase C. The load power change rate is the change in load power divided by the change in time.

[0021] The load forecasting model uses an LSTM neural network to predict load current changes based on historical zero-sequence parameters.

[0022] The harmonic analysis model uses a combined analytical method of FFT and wavelet transform to calculate the harmonic components in a preset interval.

[0023] Furthermore, calculating the harmonic components within the preset interval includes:

[0024] To calculate the frequency domain signal, the method is as follows: sum the product of the time domain signal and the complex exponential function within the range from 0 to the number of sampling points minus 1. The base of the complex exponential function is the natural constant, and the exponent of the complex exponential function is the negative imaginary unit multiplied by 2 times pi, multiplied by the frequency domain point index, multiplied by the time domain sampling point index, and then divided by the number of sampling points. The value range of the frequency domain point index is from 0 to the number of sampling points minus 1.

[0025] Amplitude calculations are performed on the frequency domain signal to obtain the amplitude spectrum at each frequency point;

[0026] The frequency domain index corresponding to the h-th harmonic is h multiplied by the preset fundamental frequency, divided by the frequency resolution, and then rounded down. The frequency resolution is the preset sampling frequency divided by the number of sampling points.

[0027] Extract the amplitude of the corresponding frequency domain points of the harmonics in the preset interval as each harmonic component.

[0028] Furthermore, the product of the preset first adaptive coefficient (with a negative compensation voltage) and the zero-sequence voltage is subtracted from the product of the preset second adaptive coefficient and the derivative of the zero-sequence voltage with respect to time. Then, the product of the preset third adaptive coefficient, the zero-sequence current, and the total impedance is subtracted. Next, the harmonic component term is subtracted. Finally, the product of the preset fourth adaptive coefficient, the predicted load current change, and the bus line impedance is added. Among these, the harmonic component term is the sum of three products obtained by multiplying the preset third harmonic compensation coefficient by the third harmonic component, the preset fifth harmonic compensation coefficient by the fifth harmonic component, and the preset seventh harmonic compensation coefficient by the seventh harmonic component. Among these, the total impedance is the bus line impedance plus the distribution transformer short-circuit impedance.

[0029] Furthermore, the product of the preset fifth adaptive coefficient (which is negative for compensation current) and the zero-sequence current is subtracted from the product of the preset sixth adaptive coefficient and the derivative of the zero-sequence current with respect to time.

[0030] Furthermore, the power electronic compensation unit includes a signal conversion layer, a power conversion layer, and an energy management layer;

[0031] The signal conversion layer converts the compensation voltage and compensation current commands into space vector pulse width modulation signals;

[0032] The power conversion layer adopts a three-phase four-bridge-arm inverter, which includes an A-phase bridge arm, a B-phase bridge arm, a C-phase bridge arm, and a neutral line bridge arm. It realizes power conversion and outputs compensation voltage and compensation current based on SVPWM signal. The neutral line bridge arm is configured with an independent neutral point regulation channel for independently injecting zero-sequence compensation voltage.

[0033] The energy management system provides energy support for compensation execution through supercapacitor energy storage modules, smoothing out transient compensation power fluctuations.

[0034] Furthermore, the compensated parameters include the compensated zero-sequence voltage, zero-sequence current, and harmonic components; the deviation information includes the zero-sequence voltage deviation, zero-sequence current deviation, and harmonic component deviation; and the preset adaptive coefficients and preset harmonic compensation coefficients in the compensation calculation module are corrected by the PID controller.

[0035] Furthermore, the zero-sequence voltage deviation is the compensated zero-sequence voltage minus the preset zero-sequence voltage target value, the zero-sequence current deviation is the compensated zero-sequence current minus the preset zero-sequence current target value, and the harmonic component deviation is the compensated harmonic component minus the preset harmonic component target value.

[0036] The beneficial effects of this invention are as follows: By employing an 18-bit synchronous ADC module and 20kHz high-precision synchronous sampling, the time synchronization and amplitude accuracy of the three-phase voltage and current signals are ensured. By using an LSTM neural network load prediction algorithm, based on historical load power change rate data, future load current changes are predicted. IT load migration commands are obtained through the BMC interface, enabling prediction of load change trends and providing forward-looking compensation capabilities, significantly improving the system's adaptability to dynamic migration and transient impacts of data center IT loads during start-up and shutdown. A multi-dimensional power quality assessment system is established, including three-phase current imbalance, neutral zero-sequence current, zero-sequence voltage offset, and 2nd-30th harmonic distortion. The joint analytical method of FFT and wavelet transform improves the signal-to-noise ratio, accurately locates the main harmonic orders, and enables rapid identification and precise management of abnormal power quality conditions.

[0037] By integrating an adaptive PID controller and a multi-parameter fusion compensation calculation model, the compensation coefficient is dynamically adjusted according to the load power change rate. When a transient impact is detected, the correlation coefficient is increased by 20% to 50% to enhance the response capability. Through closed-loop optimization control, the compensation parameters are optimized in real time, ensuring that the system maintains the optimal compensation effect under different operating conditions. By adopting a three-phase four-bridge-arm inverter topology, the neutral bridge arm has an independent zero-sequence voltage injection function and is configured with an independent neutral point adjustment channel. This allows for precise control of the neutral point potential without affecting the three-phase balanced output, achieving independent adjustment of the zero-sequence voltage and solving the technical problem that traditional three-phase three-bridge-arm structures cannot independently handle the zero-sequence component.

[0038] The supercapacitor energy storage module features a fast response time, dynamically releasing and absorbing energy during compensation execution to effectively smooth transient compensation power fluctuations. When a sudden load change is detected, it immediately provides transient power support to the inverter, ensuring the continuity and stability of the compensation output. Through a comprehensive safety protection system integrating overcurrent protection, overtemperature protection, and harmonic monitoring, the system protects response time and employs IGBT soft shutdown, ensuring timely protection while avoiding hard shutdown impacts. The remote operation and maintenance layer supports fault alarm response delays of ≤1s, enabling real-time monitoring and remote maintenance of the system status.

[0039] By employing a combined FPGA and DSP architecture, high-speed data processing and complex algorithm computation are achieved collaboratively. A remote operation and maintenance interface based on the Modbus-TCP protocol supports seamless integration with data center monitoring systems, providing remote adjustment of compensation parameters and significantly improving the system's intelligence and ease of operation and maintenance.

[0040] This invention achieves an organic combination of rapid response, high-precision compensation, intelligent prediction, and adaptive control through an innovative five-module collaborative architecture. It is particularly suitable for application scenarios with extremely high power quality requirements, such as data centers. Compared with existing technologies, it has significant improvements in response speed, compensation accuracy, and intelligence, providing a complete technical solution for power quality management. Attached Figure Description

[0041] Figure 1 This is a module example diagram of the zero-sequence compensation system of a three-phase four-arm power quality management device according to the present invention;

[0042] Figure 2 This is an example diagram of the data prediction module of the zero-sequence compensation system of a three-phase four-arm power quality management device of the present invention;

[0043] Figure 3 This is an example diagram of the compensation calculation module of the zero-sequence compensation system of a three-phase four-arm power quality management device of the present invention;

[0044] Figure 4 This is an example diagram of the compensation execution module of the zero-sequence compensation system of a three-phase four-arm power quality management device according to the present invention. Detailed Implementation

[0045] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, features described in some examples may be combined in other examples.

[0046] A zero-sequence compensation system for a three-phase four-arm power quality management device, such as Figure 1 As shown, it includes:

[0047] The synchronous sampling module 101 monitors the first electrical parameter and establishes a zero-sequence compensation triggering mechanism based on the first electrical parameter. When the zero-sequence compensation triggering mechanism starts the zero-sequence compensation process, it collects the second electrical parameter based on a preset sampling period.

[0048] This module implements data acquisition and monitoring functions for the zero-sequence compensation system based on a multi-dimensional sampling unit, providing a high-precision real-time data foundation for subsequent data prediction and compensation calculation. The unit adopts a layered architecture design, comprising three functional layers: a hardware sampling layer, a signal monitoring layer, and a data transmission layer.

[0049] The hardware sampling layer consists of an 18-bit synchronous ADC module, a Hall voltage / current sensor, and a load status acquisition module. The Hall sensor has an accuracy of 0.1% and a bandwidth of 1MHz. Working in conjunction with the synchronous ADC module, it enables high-precision synchronous sampling at a sampling frequency of 20kHz, ensuring the time synchronization and amplitude accuracy of the three-phase voltage and current signals.

[0050] The necessity of adopting a layered architecture lies in achieving high precision and real-time performance in data acquisition. The layered architecture decouples hardware sampling, signal monitoring, and data transmission functions, avoiding mutual interference between layers and ensuring each layer focuses on its core function. The advantage of using an 18-bit synchronous ADC in the hardware sampling layer is that it provides sufficient quantization accuracy. Compared to 12-bit or 14-bit ADCs, 18-bit accuracy can reduce the quantization error to below 0.0038%, meeting the sensitive detection requirements of power quality management for minute signal changes. The selection of 0.1-level accuracy and 1MHz bandwidth for the Hall sensor is based on data center load characteristic analysis. 0.1-level accuracy ensures that the current measurement error is less than 0.1%, and the 1MHz bandwidth can accurately capture the high-frequency harmonic components generated by the switching power supply. The setting of the 20kHz sampling frequency is based on the Nyquist theorem, which can accurately reconstruct harmonic components below the 30th order, meeting the requirements of power quality standards for harmonic analysis.

[0051] The signal monitoring layer establishes an intelligent zero-sequence compensation triggering mechanism based on real-time data acquired by the hardware sampling layer. This layer achieves rapid identification of abnormal power quality conditions by continuously monitoring the primary electrical parameters of the power system. The primary electrical parameters include three-phase current, neutral zero-sequence current, zero-sequence voltage, and harmonics, forming a multi-dimensional power quality assessment system.

[0052] When the zero-sequence compensation triggering mechanism detects any abnormal state, the system automatically starts the zero-sequence compensation process. Abnormal states include three-phase current imbalance, neutral line zero-sequence current exceeding the standard, zero-sequence voltage deviation, and harmonic distortion rate abnormality.

[0053] Three-phase current imbalance anomalies include: when the absolute value of the difference between phase A current and phase B current is greater than or equal to the preset three-phase current imbalance threshold, or when the absolute value of the difference between phase B current and phase C current is greater than or equal to the preset three-phase current imbalance threshold, or when the absolute value of the difference between phase C current and phase A current is greater than or equal to the preset three-phase current imbalance threshold, it indicates that the degree of three-phase load imbalance exceeds the allowable range. The default value of the preset three-phase current imbalance threshold is 0.08 times the rated line current. This threshold is determined according to the load characteristics of the data center and the three-phase imbalance limit requirements in the power quality standard. The rated line current is the standard line current value of the system design.

[0054] The neutral line zero-sequence current exceeds the standard. When the neutral line zero-sequence current is greater than or equal to the preset zero-sequence current threshold, it indicates that the neutral line is carrying an excessive unbalanced current and zero-sequence compensation is required. The default value of the preset zero-sequence current threshold is 0.05 times the rated line current. This threshold is determined based on the neutral line current limit in the standard and the characteristics of the data center IT load.

[0055] Zero-sequence voltage deviation anomaly includes: when the zero-sequence voltage is greater than or equal to the preset zero-sequence voltage deviation threshold, it indicates that the neutral point potential of the system has shifted. The default value of the preset zero-sequence voltage deviation threshold is 0.015 times the rated phase voltage. This threshold is formulated based on the voltage deviation limit in the power quality standard and the power supply quality requirements of the data center. The zero-sequence voltage is equal to the sum of the A-phase voltage, B-phase voltage, and C-phase voltage divided by 3. The rated phase voltage is the standard phase voltage value designed for the system.

[0056] Abnormal harmonic distortion rate includes: when the amplitude of any harmonic order in the preset range is greater than or equal to the preset harmonic distortion threshold, it indicates that the power system is seriously polluted by harmonics. The default value of the preset range is the 2nd to 30th harmonics, and the default value of the preset harmonic distortion threshold is 0.04 times the fundamental amplitude. This threshold is determined according to the harmonic standard and the harmonic characteristics analysis of the data center switching power supply load. The amplitude of the nth harmonic is greater than or equal to the preset harmonic distortion threshold, where n is an integer from 2 to 30, and the fundamental amplitude is the amplitude of the 1st harmonic.

[0057] After initiating the zero-sequence compensation process, the data transmission layer coordinates the work of the load status acquisition module and the multi-dimensional sampling unit. The load status acquisition module obtains IT load migration instructions through the data center BMC (Baseboard Management Controller) interface, enabling prediction of load change trends and providing forward-looking information for the compensation algorithm. Simultaneously, the multi-dimensional sampling unit synchronously acquires second electrical parameters according to a preset sampling period accuracy. These second electrical parameters include load power, instantaneous phase voltages of the three-phase bus, and instantaneous currents of the three-phase load. The instantaneous phase voltages of the three-phase bus include the instantaneous phase voltages of phases A, B, and C, and the instantaneous currents of the three-phase load include the instantaneous load currents of phases A, B, and C. The default sampling period is 50 microseconds, and a complete data acquisition cycle is completed every 50 microseconds. The acquired multi-dimensional electrical parameters are then transmitted in real-time to the intelligent control unit, providing accurate and reliable input data for the subsequent data prediction module.

[0058] The data prediction module 102 performs noise reduction processing on the second electrical parameters and then calculates the zero-sequence parameters. The load prediction model calculates future load current changes based on the zero-sequence parameters, and the harmonic analysis model calculates harmonic components based on the zero-sequence parameters. Specifically, as follows... Figure 2 As shown.

[0059] This module is implemented based on an intelligent control unit, which adopts a combined FPGA and DSP architecture. The FPGA is responsible for high-speed data processing and real-time control, while the DSP is responsible for complex algorithm calculations. This unit integrates an LSTM (Long Short-Term Memory) neural network load prediction algorithm and an adaptive PID controller, achieving accurate zero-sequence compensation prediction through a multi-level data processing flow.

[0060] After receiving the second electrical parameters transmitted by the synchronous sampling module 101, the intelligent control unit first performs wavelet transform denoising processing to obtain denoised electrical parameters to ensure the accuracy of subsequent calculations.

[0061] This embodiment uses Daubechies wavelet for three-level wavelet decomposition, effectively eliminating electromagnetic interference noise and improving the signal-to-noise ratio to over 45dB, providing a high-quality data foundation for subsequent zero-sequence parameter calculation and harmonic analysis model.

[0062] The advantage of using a combined FPGA and DSP architecture lies in fully leveraging the strengths of both processors. FPGAs possess parallel processing capabilities and deterministic latency characteristics, making them suitable for handling high-speed data streams and real-time control tasks, achieving real-time response times of up to 50 microseconds. DSPs, on the other hand, have powerful floating-point arithmetic capabilities and a rich mathematical function library, making them suitable for performing complex algorithmic operations such as FFT and LSTM network inference. Compared to a single-processor solution, this heterogeneous architecture can reduce overall processing latency by more than 60% while improving algorithm accuracy.

[0063] The Daubechies wavelet was chosen for denoising due to its excellent time-frequency localization and orthogonality, which effectively separates signal and noise components. Three-level wavelet decomposition divides the signal into eight frequency bands, precisely locating noise frequency bands. Compared to traditional low-pass filters, wavelet denoising preserves the transient characteristics of the signal, avoiding distortion of useful signals. The improved signal-to-noise ratio (SNR) of over 45dB ensures the input quality of subsequent algorithms and reduces prediction errors.

[0064] The necessity of using LSTM neural networks for load forecasting lies in the significant temporal correlation and long-term dependency characteristics of data center loads. Traditional linear forecasting methods cannot capture the nonlinear patterns of load changes, while LSTM, through its gating mechanism, can selectively remember and forget historical information, effectively handling the dependencies of long-term series. A design that predicts changes over the next 50 microseconds based on historical 100-millisecond data provides forward-looking information for the compensation algorithm, improving the compensation response time from milliseconds of passive response to microseconds of proactive prediction.

[0065] Based on the noise reduction electrical parameters, the intelligent control unit calculates the key parameters required for zero-sequence compensation. The zero-sequence parameters include zero-sequence voltage, zero-sequence current, and load power change rate. Specifically, the zero-sequence voltage is equal to the sum of the instantaneous phase voltages of phases A, B, and C, divided by 3; the zero-sequence current is equal to the sum of the instantaneous load currents of phases A, B, and C; and the load power change rate is equal to the load power change divided by the time change, where the time change is 50 microseconds.

[0066] These zero-sequence parameters provide the basic data support for subsequent load prediction models and harmonic analysis models, ensuring that the compensation algorithm can accurately identify the unbalanced state of the system.

[0067] Based on the zero-sequence parameters, the load forecasting model uses an LSTM neural network to predict load change trends using historical data. An LSTM neural network is a special type of recurrent neural network whose structure includes three gating units: an input gate, a forget gate, and an output gate, enabling it to effectively handle long-term dependencies in time-series data.

[0068] The LSTM training steps include:

[0069] Collect historical electrical parameter data under typical load scenarios in data centers, including operating data for different load rates (20%-100%) and different load types (computing, storage, and network). The data collection period is 30 consecutive days, and the sampling frequency is 20kHz.

[0070] Zero-sequence voltage, zero-sequence current, and load power change rate are extracted as input features. A sliding time window with a length of 100 milliseconds (2000 sampling points) is constructed, and the prediction target is the load current change in the next 50 microseconds.

[0071] A 3-layer LSTM network was constructed with 64, 32, and 16 neurons in the hidden layers, respectively. Dropout regularization was used to prevent overfitting, and the dropout rate was set to 0.2.

[0072] The Adam optimizer was used, with a learning rate of 0.001, a batch size of 128, and 200 training epochs. An early stopping mechanism was employed to prevent overfitting.

[0073] Validate the model performance on an independent test set to ensure that the prediction error is less than ±2%, and evaluate the model’s generalization ability through cross-validation.

[0074] The trained model parameters are embedded into the DSP to achieve real-time inference, with the inference time controlled within 10 microseconds.

[0075] The load prediction model predicts the load current change in the next 50 microseconds based on zero-sequence parameter data within the past 100 milliseconds, with a prediction error of less than or equal to ±2%. Through this forward-looking prediction, the system can respond to load changes in advance, significantly improving the real-time performance and accuracy of zero-sequence compensation.

[0076] The harmonic analysis model uses a combined analytical method of FFT (Fast Fourier Transform) and wavelet transform to calculate the 2nd to 30th harmonic components.

[0077] The advantage of using a joint analytical method combining FFT and wavelet transform lies in the complementary characteristics of the two transforms. FFT offers excellent frequency domain resolution, accurately identifying the amplitude and phase information of each harmonic, but it is sensitive to the non-stationary characteristics of the signal. Wavelet transform, on the other hand, possesses good time-frequency localization properties, effectively suppressing spectral leakage and noise interference. By using both methods together—FFT for accurate frequency domain analysis and wavelet transform for signal preprocessing and spectral leakage suppression—the joint analysis improves harmonic detection accuracy and suppresses spectral leakage compared to a single FFT method.

[0078] In the FFT harmonic analysis model, the time-domain signal refers to the electrical signal that varies with time, specifically the instantaneous current signal of the three-phase load after wavelet transform denoising. This signal is a continuous waveform data with time as the horizontal axis and current amplitude as the vertical axis. The frequency-domain signal refers to the frequency domain representation signal obtained after FFT transformation, which is discrete spectrum data with frequency as the horizontal axis and the amplitude and phase of each frequency component as the vertical axis. It is used to characterize the distribution of each frequency component in the time-domain signal.

[0079] The choice of 1024-point FFT is based on a balance between frequency resolution and computational complexity. Frequency resolution equals the sampling frequency divided by the number of FFT points. At a sampling frequency of 20kHz, the frequency resolution of a 1024-point FFT is 19.53Hz, which can clearly distinguish the harmonic components of the 50Hz fundamental frequency. Compared to a 512-point FFT, 1024 points provide higher frequency resolution and more accurately locate harmonic frequencies; compared to a 2048-point FFT, 1024 points have lower computational complexity, meeting the requirements of real-time processing. The 51.2-millisecond analysis window length includes 2.56 fundamental frequency periods, effectively suppressing the impact of fundamental frequency fluctuations on harmonic analysis.

[0080] The number of sampling points refers to the total number of points used to digitize and sample the time-domain signal within a complete FFT analysis window. This system uses 1024-point FFT analysis, meaning that the time-domain signal is sampled 1024 times at equal intervals within each FFT analysis window. The system's sampling frequency is 20kHz, corresponding to a sampling period that is the reciprocal of the sampling frequency, i.e., 50 microseconds. In 1024-point FFT analysis, the analysis window length is the number of sampling points multiplied by the sampling period, i.e., 1024 multiplied by 50 microseconds, approximately 51.2 milliseconds. The selection of the number of sampling points is based on a balance between frequency resolution requirements and computational complexity; 1024-point FFT provides sufficient frequency resolution to accurately identify the 2nd to 30th harmonic components.

[0081] The frequency domain signal is equal to the sum of the products of the time domain signal and the complex exponential function within the range from 0 to the number of sampling points minus 1. The base of the complex exponential function is the natural constant, and the exponent is the negative imaginary unit multiplied by twice pi, multiplied by the frequency domain index, multiplied by the time domain sampling point index, and then divided by the number of sampling points. The frequency domain index ranges from 0 to the number of sampling points minus 1. A specific formula is shown below:

[0082] , ;

[0083] in, For time-domain signals, For frequency domain signals, The number of sampling points. For time-domain sampling point index, For frequency domain point index, It is the imaginary unit.

[0084] Through this transformation, the system converts the current signal in the time domain into a harmonic component representation in the frequency domain, with each frequency point corresponding to the amplitude and phase information of the harmonic component at a specific frequency.

[0085] The process of obtaining harmonic components includes: first, calculating the amplitude of the frequency domain signal to obtain the amplitude spectrum of each frequency point; then, determining the frequency domain point index corresponding to each harmonic based on the preset fundamental frequency of 50Hz and the sampling frequency. The default value of the fundamental frequency is 50Hz, and the frequency domain point index corresponding to the h-th harmonic is h multiplied by the preset fundamental frequency, divided by the frequency resolution, and then rounded down. The frequency resolution is equal to the sampling frequency divided by the number of sampling points; finally, extracting the amplitude of the frequency domain points corresponding to the harmonics in the preset interval as each harmonic component for subsequent compensation calculations.

[0086] By combining FFT and wavelet transform, the system can effectively suppress spectral leakage and accurately locate the main harmonic orders, typically the 3rd, 5th, and 7th, providing accurate harmonic component data for the compensation calculation of the compensation calculation module 103.

[0087] At this point, the data prediction module has completed the entire processing flow from the original second electrical parameters to the prediction results. The output zero-sequence parameters, load prediction results, and harmonic components will serve as the core input data for the compensation calculation module 103.

[0088] The compensation calculation module 103, based on the zero-sequence parameters, load current changes, and harmonic components, calculates the compensation voltage and compensation current using an adaptive PID controller, specifically as follows: Figure 3 As shown.

[0089] The necessity of using an adaptive PID controller lies in the variable operating conditions and nonlinear characteristics faced by zero-sequence compensation systems. Traditional fixed-parameter PID controllers cannot adapt to the dynamic changes in data center loads, while adaptive PID controllers can adjust control parameters in real time according to the system's operating status, ensuring optimal compensation performance under different load conditions. Compared to traditional PID controllers, adaptive PID controllers can reduce steady-state error and improve dynamic response speed.

[0090] The advantage of the multi-parameter fusion compensation calculation model lies in its comprehensive consideration of multiple influencing factors of zero-sequence compensation. Single-parameter compensation methods are prone to local optimization problems, while the multi-parameter fusion model, by simultaneously considering five dimensions—zero-sequence voltage steady-state compensation, dynamic response, impedance coupling effect, harmonic suppression, and load prediction—can achieve a globally optimal compensation strategy. The introduction of a real-time temperature correction mechanism solves the problem of transformer impedance changing with temperature; compared to the fixed impedance model, temperature correction can improve compensation accuracy.

[0091] This module uses an adaptive PID controller based on an intelligent control unit to achieve accurate calculation of zero-sequence compensation. It receives zero-sequence parameters, load prediction results, and harmonic components from the data prediction module 102 as input data, and generates precise compensation commands through a multi-level calculation process. The zero-sequence parameters provide basic information about the system's unbalanced state; the load current change provides a forward-looking compensation basis, avoiding delayed response; and the harmonic components provide the frequency domain compensation target, achieving a wideband compensation effect.

[0092] The intelligent control unit first preprocesses and extracts parameters from the input data. Based on the zero-sequence voltage, zero-sequence current, load power change rate, and third, fifth, and seventh harmonic components obtained by the data prediction module 102, the system establishes a multi-parameter fusion compensation calculation model. This model comprehensively considers the steady-state characteristics and dynamic trends of the zero-sequence components, system impedance characteristics, harmonic pollution level, and forward-looking information on load changes to form a comprehensive compensation strategy.

[0093] During the compensation calculation process, the system first determines the key impedance parameters. The total impedance equals the bus line impedance plus the distribution transformer short-circuit impedance, where the bus line impedance is the line impedance value of the system bus, and the distribution transformer short-circuit impedance is the short-circuit impedance of the distribution transformer. Considering the influence of temperature on transformer impedance, the system adopts a real-time correction mechanism. The distribution transformer short-circuit impedance is equal to the distribution transformer short-circuit impedance at 25 degrees Celsius multiplied by (1 plus a temperature difference term). The temperature difference term is 0.004 multiplied by (the difference between the real-time transformer temperature and 25 degrees Celsius), where the real-time transformer temperature is the current temperature value of the transformer. This real-time correction ensures the accuracy of the impedance parameters, providing reliable basic data for subsequent compensation calculations.

[0094] Based on accurate impedance parameters and input data, the system employs a multi-parameter fusion model to calculate the compensation voltage and current. The calculation of the compensation voltage comprehensively considers five dimensions: steady-state compensation of zero-sequence voltage, dynamic response, impedance coupling effect, harmonic suppression, and forward-looking compensation for load changes.

[0095] The compensation voltage equals the product of the negative preset first adaptive coefficient and the zero-sequence voltage, minus the product of the preset second adaptive coefficient and the derivative of the zero-sequence voltage with respect to time, then minus the product of the preset third adaptive coefficient, the zero-sequence current and the total impedance, then minus the harmonic component term, and finally add the product of the preset fourth adaptive coefficient, the predicted load current change and the bus line impedance. The harmonic component term is the sum of three products: the preset third harmonic compensation coefficient multiplied by the third harmonic component, the preset fifth harmonic compensation coefficient multiplied by the fifth harmonic component, and the preset seventh harmonic compensation coefficient multiplied by the seventh harmonic component. The preset first adaptive coefficient has a default value of 0.85, determined based on the zero-sequence voltage steady-state compensation requirements and system stability analysis; the preset second adaptive coefficient has a default value of 0.12, set according to the zero-sequence voltage dynamic response characteristics and control system bandwidth requirements; the preset third adaptive coefficient has a default value of 0.75, formulated based on the theoretical analysis and simulation verification results of the zero-sequence current and system impedance coupling effect; the preset fourth adaptive coefficient has a default value of 0.65, determined based on the load prediction accuracy and forward-looking compensation effect optimization; the preset third harmonic compensation coefficient has a default value of 0.90, the preset fifth harmonic compensation coefficient has a default value of 0.85, and the preset seventh harmonic compensation coefficient has a default value of 0.80. These coefficients are set according to the frequency domain analysis results of the typical harmonic distribution characteristics of the data center and the degree of influence of each harmonic on zero-sequence compensation.

[0096] The first item achieves direct compensation of zero-sequence voltage, the second item provides dynamic response capability, the third item considers the coupling effect of zero-sequence current and system impedance, the fourth item accurately suppresses major harmonic components, and the fifth item achieves forward-looking compensation based on load prediction.

[0097] The calculation of the compensation current focuses on the direct suppression and dynamic tracking of the zero-sequence current:

[0098] The compensation current equals the product of the negative preset fifth adaptive coefficient and the zero-sequence current, minus the product of the preset sixth adaptive coefficient and the derivative of the zero-sequence current with respect to time. The preset fifth adaptive coefficient has a default value of 0.92, determined based on the direct compensation effect of the zero-sequence current and the system response speed requirements; the preset sixth adaptive coefficient has a default value of 0.18, set according to the tracking accuracy of the zero-sequence current change trend and the dynamic stability analysis results.

[0099] The first item enables direct compensation of zero-sequence current, while the second item provides the ability to dynamically track the changing trend of zero-sequence current.

[0100] To achieve dynamic optimization of the compensation effect, an adaptive PID controller is integrated into the system. The mathematical expression of this controller is:

[0101] The controller output equals the product of the preset proportional coefficient and the error signal, plus the product of the preset integral coefficient and the integral of the error signal from 0 to the current time t, plus the product of the preset differential coefficient and the derivative of the error signal with respect to time.

[0102] The controller output is the output signal of the PID controller. The preset proportional coefficient has a default value of 1.25, which is determined based on the system steady-state error requirements and response speed optimization. The preset integral coefficient has a default value of 0.08, which is set according to the system steady-state accuracy requirements and integral saturation protection requirements. The preset derivative coefficient has a default value of 0.15, which is formulated based on the system overshoot control and noise suppression capability balance analysis results. The error signal is the system error input signal, and the integral variable is the variable in the integral operation.

[0103] The adaptive PID controller dynamically adjusts the compensation coefficients based on the load power change rate, achieving real-time optimization of the compensation strategy. When the load power change rate increases, the system identifies a transient impact state. At this time, the preset second, fourth, and sixth adaptive coefficients are increased by preset adjustment ranges to enhance transient response capability. The default value of the preset adjustment range is 20% to 50%, and this adjustment range is determined based on transient impact response requirements and system stability boundary analysis. This adaptive adjustment mechanism ensures that the system maintains optimal compensation performance under different operating conditions.

[0104] The system uses preset first to sixth adaptive coefficients as its adaptive adjustment coefficients, and preset harmonic compensation coefficients as the compensation adjustment coefficients corresponding to each harmonic. These coefficients are dynamically adjusted by an adaptive PID controller to ensure the accuracy and real-time performance of the compensation calculation.

[0105] At this point, the compensation calculation module has completed the entire calculation process from input data to compensation instructions, and the output compensation voltage and compensation current will be transmitted to the compensation execution module 104 for precise execution.

[0106] The compensation execution module 104 converts the compensation voltage and compensation current into control signals and outputs compensation power through the power electronic compensation unit, specifically as follows: Figure 4 As shown.

[0107] This module is based on a power electronic compensation unit to achieve precise execution of zero-sequence compensation commands. The power electronic compensation unit is an integrated power conversion device specifically designed to execute zero-sequence compensation commands and output corresponding compensation voltages and currents. This unit adopts a hierarchical collaborative architecture, including three functional layers: a signal conversion layer, a power conversion layer, and an energy management layer, forming a complete execution chain from receiving compensation commands to receiving compensation outputs.

[0108] The signal conversion layer receives the compensation voltage and compensation current commands output by the compensation calculation module 103, and achieves precise conversion from digital signals to analog control signals through the intelligent control unit built into the power electronic compensation unit. The intelligent control unit first preprocesses the compensation commands, converting the compensation voltage and compensation current into space vector pulse width modulation (SVPWM) signals suitable for the control of power electronic devices. SVPWM is an advanced PWM technology. Its basic principle is to approximate a circular rotating magnetic field with the composite vector of three-phase sine waves. Compared with traditional PWM technology, it has higher voltage utilization and lower harmonic content, and is particularly suitable for the precise control requirements of zero-sequence compensation.

[0109] During SVPWM signal generation, the intelligent control unit uses a sector determination algorithm to determine the spatial position of the current voltage vector. The SVPWM sector determination formula is as follows: the first sector determination reference value equals the beta-axis voltage component; the second sector determination reference value equals (the square root of three divided by two, multiplied by the alpha-axis voltage component) plus (half multiplied by the beta-axis voltage component); the third sector determination reference value equals (the negative square root of three divided by two, multiplied by the alpha-axis voltage component) plus (half multiplied by the beta-axis voltage component). Here, the first, second, and third sector determination reference values ​​are the reference values ​​for sector determination, and the alpha-axis and beta-axis voltage components are the voltage components in the alpha-beta coordinate system. Through accurate sector determination, the intelligent control unit can generate the optimal switching sequence, ensuring the accuracy and stability of the compensation output.

[0110] The power conversion layer is the core execution part of the power electronic compensation unit, achieving precise output of compensation power based on a three-phase four-arm inverter. Compensation power refers to the sum of active and reactive power output by the power electronic compensation unit to achieve zero-sequence compensation. It is calculated as compensation power equal to compensation voltage multiplied by compensation current multiplied by the power factor, where the power factor is the cosine of the phase difference between the compensation voltage and compensation current. The magnitude of the compensation power directly determines the effectiveness of zero-sequence compensation and the system's energy consumption; its precise control is a key technical indicator for achieving efficient zero-sequence compensation. The three-phase four-arm inverter adopts a four-arm topology with phase A, phase B, phase C, and a neutral arm. Phase A, phase B, and phase C correspond to the three phases of the three-phase system, while the neutral arm is specifically used for zero-sequence compensation control. The four-arm inverter receives the SVPWM signal generated by the signal conversion layer and achieves DC-to-AC power conversion through precise switching control. The neutral arm of this inverter features an independent zero-sequence voltage injection function. This arm is equipped with an independent neutral point adjustment channel, enabling precise control of the neutral point potential and independent adjustment of the zero-sequence voltage without affecting the three-phase balanced output. When the SVPWM signal is output to the three-phase four-arm inverter, the inverter precisely injects compensation voltage through the independent neutral point adjustment channel of the neutral arm, simultaneously outputting a compensation current that matches the zero-sequence current compensation requirement, ensuring the dual effect of zero-sequence compensation.

[0111] The necessity of choosing a three-phase four-arm inverter topology lies in its unique neutral arm design. Traditional three-phase three-arm inverters cannot independently control the zero-sequence component, while the four-arm topology, by adding a neutral arm, can independently adjust the neutral point potential without affecting the three-phase balanced output, achieving precise control of the zero-sequence voltage. Compared to the three-arm scheme, this topology can improve the zero-sequence compensation accuracy.

[0112] The energy management layer is the energy support component of the power electronic compensation unit, providing stable energy support for the compensation process through supercapacitor energy storage modules. These supercapacitor energy storage modules feature a rapid response time of ≤10 microseconds, dynamically releasing and absorbing energy during compensation execution to effectively smooth transient power fluctuations. When the power electronic compensation unit detects a sudden load change or transient impact, the supercapacitor energy storage modules respond immediately, providing transient compensation power support to the three-phase four-arm inverter, ensuring the continuity and stability of the compensation output. Through the coordinated operation of these three layers, the entire power electronic compensation unit achieves a rapid response capability of ≤10 microseconds for compensation power response time, meeting the stringent requirements of dynamically changing IT loads in data centers.

[0113] The introduction of supercapacitor energy storage modules solves the problem of transient power demand during the compensation process. Sudden changes in data center load generate transient compensation power demands. Traditional DC power supplies have slow response times, while supercapacitors have a fast response time of ≤10 microseconds, providing transient power support at the moment of load change, ensuring the continuity and stability of the compensation output. Compared to pure power supply solutions, supercapacitor energy storage can shorten the compensation response time.

[0114] The closed-loop optimization module 105 collects the compensated parameters, compares them with the preset compensation target value to obtain deviation information, and dynamically optimizes the compensation parameters based on the deviation information.

[0115] The necessity of employing closed-loop optimization control lies in the external disturbances and parameter drift problems faced by zero-sequence compensation systems. Open-loop control systems cannot perceive the compensation effect and are easily affected by factors such as load changes, ambient temperature, and component aging, leading to a gradual decrease in compensation accuracy. Closed-loop control, through real-time feedback of the compensation effect, can automatically correct control parameters, ensuring that the system always maintains optimal compensation performance. Compared to open-loop control, closed-loop optimization can improve the long-term compensation accuracy retention rate.

[0116] This module implements closed-loop optimization control of the zero-sequence compensation system based on the protection and monitoring unit. This unit adopts a layered collaborative architecture, comprising five functional layers: a feedback acquisition layer, a target comparison layer, a dynamic optimization layer, a protection and monitoring layer, and a remote operation and maintenance layer. This forms a complete closed-loop control link from compensation effect evaluation to system protection. The advantage of this five-layered collaborative architecture lies in achieving functional decoupling and specialized division of labor. The feedback acquisition layer focuses on high-precision data acquisition, the target comparison layer is responsible for effect evaluation, the dynamic optimization layer performs parameter adjustment, the protection and monitoring layer ensures system safety, and the remote operation and maintenance layer provides a human-machine interface. Compared to a centralized control architecture, this layered design improves system reliability and maintainability; a failure in one layer will not affect the normal operation of other layers.

[0117] The feedback acquisition layer receives the compensation execution results from the compensation execution module 104 and collects the system status parameters after compensation in real time through a multi-dimensional sampling unit. This layer uses the same 18-bit synchronous ADC module and Hall sensor as the synchronous sampling module 101 to ensure the accuracy and real-time performance of the feedback data. The multi-dimensional sampling unit synchronously collects the compensated parameters with a sampling period of 50 microseconds. The compensated parameters include the compensated zero-sequence voltage, zero-sequence current, and harmonic components, providing an accurate feedback data basis for subsequent target comparison and dynamic optimization.

[0118] The target comparison layer receives real-time data from the feedback acquisition layer and establishes a multi-dimensional compensation effect evaluation system. This layer accurately compares the acquired compensated parameters with preset compensation target values ​​to form a systematic effect evaluation mechanism. The preset compensation target values ​​are set based on the technical requirements of the data center's low-voltage power distribution system. The preset compensation target values ​​include preset zero-sequence voltage target values, preset zero-sequence current target values, and preset harmonic component target values. The preset zero-sequence voltage target value is set to a compensated zero-sequence voltage less than or equal to 0.015 times the rated phase voltage. Its default value is based on power quality standards and data center power supply quality requirements to ensure that neutral point potential deviation is effectively suppressed. The preset zero-sequence current target value is set to a compensated zero-sequence current less than or equal to 0.03 times the rated line current. Its default value is determined based on standards and data center IT load characteristics analysis to achieve precise control of the neutral line zero-sequence current. The preset harmonic component target value is set to a compensated harmonic component less than or equal to 0.04 times the fundamental amplitude. Its default value is set according to harmonic standards and data center power quality requirements to ensure the harmonic quality of the power system.

[0119] During the target comparison process, the system generates multi-dimensional deviation information. This deviation information refers to the difference between the compensated parameters and the preset compensation target values, used to quantify the degree of deviation between the compensation effect and the expected target. The deviation information includes deviations in three dimensions: zero-sequence voltage deviation, zero-sequence current deviation, and harmonic component deviation. The zero-sequence voltage deviation equals the compensated zero-sequence voltage minus the preset zero-sequence voltage target value. A positive deviation indicates insufficient zero-sequence voltage compensation, while a negative deviation indicates over-compensation. The zero-sequence current deviation equals the compensated zero-sequence current minus the preset zero-sequence current target value. A positive deviation indicates that the zero-sequence current suppression effect has not met expectations, while a negative deviation indicates that the zero-sequence current suppression effect has exceeded expectations. The harmonic component deviation equals the compensated harmonic component minus the preset harmonic component target value. A positive deviation indicates insufficient harmonic suppression effect, while a negative deviation indicates good harmonic suppression effect. The introduction of multi-dimensional deviation information enables a comprehensive evaluation of the compensation effect. Single-dimensional evaluation can easily lead to one-sided optimization, while a comprehensive evaluation of three dimensions—zero-sequence voltage deviation, zero-sequence current deviation, and harmonic component deviation—can fully reflect the performance status of the compensation system. The application of the moving average filtering algorithm effectively suppresses random noise in the deviation information and improves the stability of the optimization algorithm.

[0120] The deviation information is collected based on a real-time comparison calculation mechanism. The target comparison layer calculates the deviation according to a preset comparison period, with a default value of 50 microseconds. This period is synchronized with the sampling period of the feedback acquisition layer to ensure the real-time nature and accuracy of the deviation information. Within each comparison period, the target comparison layer first obtains the latest compensated parameters from the feedback acquisition layer, and then compares them item by item with the corresponding preset compensation target value to generate deviation information in three dimensions. To ensure the reliability of the deviation information, the system uses a moving average filtering algorithm to preprocess the deviation information. The default moving average window length is 10 sampling points. This window length is determined based on a balance analysis between the stability requirements of the deviation information and the dynamic response speed, effectively suppressing random noise in the deviation information and improving its stability and usability. Through this multi-dimensional target comparison mechanism, the system can comprehensively evaluate the compensation effect and provide accurate deviation information for dynamic optimization.

[0121] The dynamic optimization layer, based on the deviation information provided by the target comparison layer, optimizes the compensation parameters in real time through the adaptive PID controller of the intelligent control unit. This layer adopts the same FPGA and DSP combination architecture as the compensation calculation module 103, ensuring high-speed execution of the optimization algorithm. The adaptive PID controller dynamically corrects the preset first to sixth adaptive coefficients in the compensation calculation module 103 according to the deviation between the feedback result and the target value, achieving precise adjustment of the compensation strategy. The optimization process uses a preset iteration period, with a default value of 0.1 milliseconds. This period is determined based on the overall system control cycle and real-time requirements, maintaining synchronization with the overall system control cycle to ensure the real-time performance of the closed-loop control.

[0122] When the compensation effect deviates from the target value, the adaptive PID controller immediately activates the parameter correction mechanism. The specific adjustment method is as follows:

[0123] The adjustment method for the first to sixth preset adaptive coefficients is based on the quantitative analysis of deviation information and adaptive control theory. When the zero-sequence voltage deviation increases, the system uses a proportional-integral adjustment algorithm to correct the preset first and second adaptive coefficients. The adjustment formula for the preset first adaptive coefficient is: the adjusted preset first adaptive coefficient equals the original preset first adaptive coefficient plus the product of the preset first coefficient adjustment gain and the zero-sequence voltage deviation, where the default value of the preset first coefficient adjustment gain is 0.15, which is determined based on the zero-sequence voltage steady-state compensation response characteristics and system stability boundary analysis. The adjustment formula for the preset second adaptive coefficient is: the adjusted preset second adaptive coefficient equals the original preset second adaptive coefficient plus the product of the preset second coefficient adjustment gain and the derivative of the zero-sequence voltage deviation with respect to time, where the default value of the preset second coefficient adjustment gain is 0.08, which is set according to the zero-sequence voltage dynamic response speed requirements and the control system stability requirements.

[0124] When the zero-sequence current deviation increases, the system uses an adaptive proportional-derivative adjustment algorithm to correct the preset fifth and sixth adaptive coefficients. The adjustment formula for the preset fifth adaptive coefficient is: the adjusted preset fifth adaptive coefficient equals the original preset fifth adaptive coefficient plus the product of the preset fifth coefficient adjustment gain and the zero-sequence current deviation. The default value for the preset fifth coefficient adjustment gain is 0.12, determined based on the optimization of the direct compensation effect of the zero-sequence current and the system response accuracy requirements. The adjustment formula for the preset sixth adaptive coefficient is: the adjusted preset sixth adaptive coefficient equals the original preset sixth adaptive coefficient plus the product of the preset sixth coefficient adjustment gain and the derivative of the zero-sequence current deviation with respect to time. The default value for the preset sixth coefficient adjustment gain is 0.06, set based on the balance analysis results of the zero-sequence current change trend tracking accuracy and dynamic stability.

[0125] When the system impedance characteristics change or the load prediction accuracy decreases, the system uses a feedforward compensation adjustment algorithm to correct the preset third and fourth adaptive coefficients. The adjustment formula for the preset third adaptive coefficient is: the adjusted preset third adaptive coefficient equals the original preset third adaptive coefficient plus the product of the preset third coefficient adjustment gain, zero-sequence current deviation, and total impedance change rate. The default value for the preset third coefficient adjustment gain is 0.10, determined based on theoretical analysis of the coupling effect between zero-sequence current and system impedance, and real-time correction requirements. The total impedance change rate equals the current total impedance minus the reference total impedance, divided by the reference total impedance, which is the standard total impedance value during system design. The adjustment formula for the preset fourth adaptive coefficient is: the adjusted preset fourth adaptive coefficient equals the original preset fourth adaptive coefficient plus the product of the preset fourth coefficient adjustment gain and the load prediction error. The default value for the preset fourth coefficient adjustment gain is 0.09, determined based on the requirements for optimizing load prediction accuracy and improving forward-looking compensation effects. The load prediction error equals the actual load current change minus the predicted load current change.

[0126] The adjustment methods for the preset third, fifth, and seventh harmonic compensation coefficients are based on frequency domain analysis and harmonic suppression effect evaluation. When the harmonic component deviation exceeds the allowable range, the system uses a frequency domain adaptive adjustment algorithm to correct the compensation coefficients for each harmonic. The adjustment formula for the preset third harmonic compensation coefficient is: the adjusted preset third harmonic compensation coefficient equals the original preset third harmonic compensation coefficient plus the product of the preset third harmonic coefficient adjustment gain and the third harmonic component deviation. The default value of the preset third harmonic coefficient adjustment gain is 0.18, which is determined based on the typical distribution characteristics of the third harmonic in the data center load and the optimization requirements for suppression effect. The adjustment formula for the preset fifth harmonic compensation coefficient is: the adjusted preset fifth harmonic compensation coefficient equals the original preset fifth harmonic compensation coefficient plus the product of the preset fifth harmonic coefficient adjustment gain and the fifth harmonic component deviation. The default value of the preset fifth harmonic coefficient adjustment gain is 0.16, which is set based on the frequency domain analysis results of the influence of the fifth harmonic on zero-sequence compensation and the suppression accuracy requirements. The formula for adjusting the preset seventh harmonic compensation coefficient is as follows: the adjusted preset seventh harmonic compensation coefficient is equal to the original preset seventh harmonic compensation coefficient plus the product of the preset seventh harmonic coefficient adjustment gain and the seventh harmonic component deviation. The default value of the preset seventh harmonic coefficient adjustment gain is 0.14, which is determined based on the propagation characteristics of the seventh harmonic in the power system and the difficulty of compensation control.

[0127] To ensure the stability and convergence of coefficient adjustments, the system imposes boundary constraints on all adjusted coefficients. The preset adjustment range for the first to sixth adaptive coefficients is limited to ±30% of their original default values. This limit is determined based on system stability analysis and control theory boundary conditions to prevent over-adjustment leading to system oscillations. The preset adjustment range for the third, fifth, and seventh harmonic compensation coefficients is limited to ±25% of their original default values. This limit is set based on the stability requirements of harmonic compensation control and frequency domain characteristic analysis results to ensure the stability of harmonic suppression effects. When an adjusted coefficient exceeds the boundary constraints, the system automatically limits it to the boundary value and records the boundary constraint event, providing reference information for system operation and maintenance and parameter optimization.

[0128] Through this dynamic adjustment mechanism based on deviation information quantitative analysis, the system achieves continuous improvement and adaptive adjustment of the compensation effect, ensuring optimal zero-sequence compensation performance under different operating conditions. The dynamic optimization algorithm employs an adaptive adjustment mechanism due to the time-varying characteristics of the compensation system. Fixed compensation parameters cannot adapt to load and environmental changes, while adaptive adjustment based on deviation information can track system state changes in real time, ensuring that the compensation parameters are always near their optimal values. The boundary constraint mechanism prevents system instability caused by excessive parameter adjustment, ensuring the safety of the optimization process.

[0129] Based on dynamic optimization, the protection and monitoring layer monitors the system's operating status in real time to ensure the safe and reliable operation of the zero-sequence compensation system. This layer integrates overcurrent protection, overtemperature protection, and harmonic monitoring modules, forming a comprehensive safety protection system. The protection and monitoring layer continuously monitors key operating parameters of the three-phase four-arm inverter, including inverter temperature, capacitor voltage, and compensation current amplitude. When system operating parameters exceed safe limits, the protection and monitoring layer immediately activates the corresponding protection mechanisms. The overcurrent protection mechanism is triggered when the compensation current is greater than or equal to a preset overcurrent protection threshold, with a default value of 1.2 times the rated line current. This threshold is determined based on the safe operating area and overload capacity analysis of power electronic devices to prevent overload damage. The overtemperature protection mechanism is activated when the IGBT module temperature is greater than or equal to a preset overtemperature protection threshold, with a default value of 85 degrees Celsius. This threshold is set based on the thermal characteristics and safe operating temperature range of the IGBT devices. The IGBT module temperature is the same as that of the insulated-gate bipolar transistor module, ensuring that power devices operate within a safe temperature range. The protection action uniformly adopts the IGBT soft shutdown method, and the preset protection response time is 2-10 microseconds. This response time is determined based on the balance analysis of protection timeliness and system impact minimization. It ensures the timeliness of protection and avoids the impact of hard shutdown on the system, thus ensuring the safe operation of the device.

[0130] The remote operation and maintenance layer, serving as the outermost interface of the closed-loop optimization module, enables remote monitoring and operation and maintenance management of the system status. This layer builds its communication interface based on the Modbus-TCP protocol, supporting seamless integration with data center monitoring systems. The remote operation and maintenance layer collects operational data from the first four functional layers in real time, including compensation effect evaluation results, dynamic optimization status, protection action records, and other key information, providing comprehensive system status information to operation and maintenance personnel through standardized communication protocols. When the protection monitoring layer detects an abnormal state or triggers a protection action, the remote operation and maintenance layer immediately generates a fault alarm message. The default value for the preset fault alarm response delay is less than or equal to 1 second. This response delay is determined based on data center operation and maintenance response requirements and communication protocol transmission characteristics, ensuring that operation and maintenance personnel can promptly obtain system status information. Simultaneously, the remote operation and maintenance layer supports remote adjustment of compensation parameters. Operation and maintenance personnel can remotely modify preset adaptive coefficients and preset target value settings according to load changes and compensation effects, achieving flexible system configuration and optimization adjustments, facilitating real-time monitoring and maintenance of the entire zero-sequence compensation system.

[0131] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments under the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.

Claims

1. A zero-sequence compensation system for a three-phase four-arm power quality management device, characterized in that, include: The synchronous sampling module monitors the first electrical parameter and establishes a zero-sequence compensation triggering mechanism based on the first electrical parameter. When the zero-sequence compensation triggering mechanism starts the zero-sequence compensation process, it collects the second electrical parameter based on a preset sampling period. The first electrical parameter includes three-phase current, neutral line zero-sequence current, zero-sequence voltage, and harmonics. The data prediction module performs noise reduction processing on the second electrical parameters and then calculates the zero-sequence parameters. The load prediction model calculates the future load current change based on the zero-sequence parameters, and the harmonic analysis model calculates the harmonic components based on the zero-sequence parameters. The second electrical parameters include load power, instantaneous phase voltage of the three-phase bus, and instantaneous current of the three-phase load. The instantaneous phase voltage of the three-phase bus includes the instantaneous phase voltage of phase A, phase B, and phase C. The instantaneous current of the three-phase load includes the instantaneous load current of phase A, phase B, and phase C. The zero-sequence parameters include zero-sequence voltage, zero-sequence current, and load power change rate. The zero-sequence voltage is the sum of the instantaneous phase voltages of phase A, phase B, and phase C divided by 3. The zero-sequence current is the sum of the instantaneous load currents of phase A, phase B, and phase C. The load power change rate is the load power change divided by the time change. The compensation calculation module calculates the compensation voltage and compensation current based on the zero-sequence parameters, load current changes, and harmonic components using an adaptive PID controller. The compensation execution module converts the compensation voltage and compensation current into control signals and outputs compensation power through the power electronic compensation unit. The closed-loop optimization module collects the compensated parameters, compares them with the preset compensation target value to obtain deviation information, and dynamically optimizes the compensation parameters based on the deviation information.

2. The zero-sequence compensation system of a three-phase four-arm power quality management device according to claim 1, characterized in that, Zero-order compensation triggering mechanisms include: When the zero-sequence compensation triggering mechanism detects any abnormal state, the zero-sequence compensation process is initiated. Abnormal states include three-phase current imbalance, excessive neutral zero-sequence current, abnormal zero-sequence voltage deviation, and abnormal harmonic distortion rate. Three-phase current imbalance includes: when the absolute value of the difference between phase A current and phase B current is greater than or equal to the preset three-phase current imbalance threshold, or when the absolute value of the difference between phase B current and phase C current is greater than or equal to the preset three-phase current imbalance threshold, or when the absolute value of the difference between phase C current and phase A current is greater than or equal to the preset three-phase current imbalance threshold. The neutral zero-sequence current exceeds the limit when the neutral zero-sequence current is greater than or equal to the preset zero-sequence current threshold. Zero-sequence voltage offset anomaly includes: when the zero-sequence voltage is greater than or equal to the preset zero-sequence voltage offset threshold, it indicates that the neutral point potential of the system has shifted; Harmonic distortion rate anomalies include: when any harmonic amplitude value in the preset interval is greater than or equal to the preset harmonic distortion threshold.

3. The zero-sequence compensation system of a three-phase four-arm power quality management device according to claim 2, characterized in that, The load forecasting model uses an LSTM neural network to predict load current changes based on historical zero-sequence parameters. The harmonic analysis model uses a combined analytical method of FFT and wavelet transform to calculate the harmonic components in a preset interval.

4. The zero-sequence compensation system of a three-phase four-arm power quality management device according to claim 3, characterized in that, The calculation of harmonic components in the preset interval includes: To calculate the frequency domain signal, the method is as follows: sum the product of the time domain signal and the complex exponential function within the range from 0 to the number of sampling points minus 1. The base of the complex exponential function is the natural constant, and the exponent of the complex exponential function is the negative imaginary unit multiplied by 2 times pi, multiplied by the frequency domain point index, multiplied by the time domain sampling point index, and then divided by the number of sampling points. The value range of the frequency domain point index is from 0 to the number of sampling points minus 1. Amplitude calculations are performed on the frequency domain signal to obtain the amplitude spectrum at each frequency point; The frequency domain index corresponding to the h-th harmonic is h multiplied by the preset fundamental frequency, divided by the frequency resolution, and then rounded down. The frequency resolution is the preset sampling frequency divided by the number of sampling points. Extract the amplitude of the corresponding frequency domain points of the harmonics in the preset interval as each harmonic component.

5. The zero-sequence compensation system of a three-phase four-arm power quality management device according to claim 1, characterized in that, The product of the preset first adaptive coefficient (with a negative compensation voltage) and the zero-sequence voltage is subtracted from the product of the preset second adaptive coefficient and the derivative of the zero-sequence voltage with respect to time. Then, the product of the preset third adaptive coefficient, the zero-sequence current, and the total impedance is subtracted. Next, the harmonic component term is subtracted. Finally, the product of the preset fourth adaptive coefficient, the predicted load current change, and the bus line impedance is added. Among them, the harmonic component term is the sum of three products obtained by multiplying the preset third harmonic compensation coefficient by the third harmonic component, the preset fifth harmonic compensation coefficient by the fifth harmonic component, and the preset seventh harmonic compensation coefficient by the seventh harmonic component. Among them, the total impedance is the bus line impedance plus the distribution transformer short-circuit impedance.

6. The zero-sequence compensation system of a three-phase four-arm power quality management device according to claim 5, characterized in that, The product of the negative preset fifth adaptive coefficient and the zero-sequence current is subtracted from the product of the preset sixth adaptive coefficient and the derivative of the zero-sequence current with respect to time.

7. The zero-sequence compensation system of a three-phase four-arm power quality management device according to claim 1, characterized in that, The power electronic compensation unit includes a signal conversion layer, a power conversion layer, and an energy management layer; The signal conversion layer converts the compensation voltage and compensation current commands into space vector pulse width modulation signals; The power conversion layer adopts a three-phase four-bridge-arm inverter, which includes an A-phase bridge arm, a B-phase bridge arm, a C-phase bridge arm, and a neutral line bridge arm. It realizes power conversion and outputs compensation voltage and compensation current based on SVPWM signal; the neutral line bridge arm is configured with an independent neutral point adjustment channel. The energy management system provides energy support for compensation execution through supercapacitor energy storage modules.

8. The zero-sequence compensation system of a three-phase four-arm power quality management device according to claim 1, characterized in that, The compensated parameters include the compensated zero-sequence voltage, zero-sequence current, and harmonic components; the deviation information includes the zero-sequence voltage deviation, zero-sequence current deviation, and harmonic component deviation; the preset adaptive coefficient and preset harmonic compensation coefficient in the compensation calculation module are corrected by the PID controller.

9. The zero-sequence compensation system of a three-phase four-arm power quality management device according to claim 8, characterized in that, The zero-sequence voltage deviation is the zero-sequence voltage after compensation minus the preset zero-sequence voltage target value; the zero-sequence current deviation is the zero-sequence current after compensation minus the preset zero-sequence current target value; and the harmonic component deviation is the harmonic component after compensation minus the preset harmonic component target value.

Images