Circulating current suppression and current sharing control method of UPS parallel system

Abstract

According to the circulating current suppression and current sharing control method of the UPS parallel system, a self-adaptive state observer and a self-adaptive virtual impedance compensator are embedded in a digital controller of an inverter, and the inverter is controlled according to the following steps that a, output phase current and output voltage signals of the inverter are collected; b, the adaptive state observer identifies a real-time impedance estimation value of an inverter output loop on line; c, assigning the real-time impedance estimation value to the virtual impedance, and calculating the theoretical voltage drop generated by the output current on the virtual impedance; d, generating a voltage reference instruction; e, the voltage reference instruction is sent to a voltage and current double closed-loop PID controller at the rear stage of the inverter, and closed-loop control of the parallel system is completed. According to the invention, accurate perception of line impedance is realized through the adaptive state observer, and a compensated voltage reference instruction is generated through the adaptive virtual impedance compensator, so that the operation efficiency and reliability of the UPS parallel system are greatly improved, and the service life of the UPS parallel system is greatly prolonged.

Description

Technical Field

[0002] This invention relates to a method for parallel connection of UPS to suppress circulating current or share current, belonging to the field of energy storage technology. Background Technology

[0004] In the power supply systems of critical facilities such as data centers and communication base stations, uninterruptible power supplies (UPS) are core devices that ensure power continuity and prevent data loss and equipment damage. To meet redundancy and expansion requirements, multiple UPS inverter modules are often connected in parallel in practical applications. Their equivalent circuit is as follows: Figure 1 As shown, this architecture can improve the reliability and capacity scalability of the power supply system through the collaborative output of multiple modules, and has become the mainstream configuration method for critical power supply scenarios.

[0005] A utility model patent with application number CN201621075287.5 in the Chinese Patent Database discloses a UPS parallel circuit, including at least two UPS units. Each UPS includes an input terminal, a frequency converter module, a transformer module, and an output terminal connected in sequence. The frequency converter module includes a rectifier circuit, an inverter circuit, a signal acquisition circuit, and a control circuit. The control circuit includes a controller, a processor, and a PLL (phase-locked loop) circuit. The signal acquisition circuit acquires the voltage and current signals output by the inverter circuit and transmits them to the processor. The signal acquisition circuit also acquires the voltage signal output by the inverter circuit and transmits it to the PLL circuit. The output signals of the processor and the PLL circuit are connected in parallel and transmitted to the controller. The controller outputs a control signal to the inverter module. This utility model eliminates the need for connecting lines between the parallel UPS units, enabling wire-free parallel UPS connections. It does not limit the distance between the parallel UPS units, introduces no interference, and improves system reliability. Furthermore, the UPS units are electrically isolated from each other, facilitating installation and maintenance; expansion is also simpler and faster.

[0006] In existing parallel UPS inverter systems, the most commonly used control strategy is droop control. This strategy simulates the power frequency and excitation characteristics of a synchronous generator, and can achieve automatic power distribution of the load without interconnection communication lines. It is widely used in the industrial field due to its simple structure and low cost. However, its core control logic still follows the basic principle of droop control, relying on the line impedance characteristics to achieve power distribution.

[0007] However, the aforementioned traditional control technologies and related improvement schemes have significant limitations in practical engineering applications. The core problem stems from the difficulty in achieving perfect impedance matching in the output circuits of each parallel inverter unit. During actual wiring and long-term operation, factors such as differences in line length, contact resistance fluctuations, temperature drift, and component aging inevitably lead to inconsistencies in the line impedances of each parallel branch. This impedance mismatch can trigger a series of serious problems: on the one hand, it can cause fundamental current circulation between parallel inverters (such as...). Figure 1 Parallel circulation in This not only increases system power loss and thermal stress on power electronic devices, but in severe cases it can also trigger overcurrent protection, causing the entire power supply system to fail. On the other hand, it can cause steady-state power distribution errors, causing some inverters to operate under unbalanced load for a long time, further aggravating device losses and shortening the system's lifespan.

[0008] To address the problems caused by impedance mismatch, the industry has proposed a variety of improvement solutions, which can be mainly divided into two categories: one is an online compensation solution based on parameter identification, and the other is a static compensation solution based on fixed virtual impedance or manual parameter adjustment. Among them, online identification techniques such as frequency domain injection and recursive least squares (RLS) have been applied in academic research and some industrial scenarios. Methods such as extended Kalman filter (EKF) and unscented Kalman filter (UKF) have also been tried for line impedance parameter estimation. However, these identification methods all have obvious defects: RLS and Kalman filter (KF) have problems such as slow convergence speed and oscillation of estimated values ​​in high noise environment or near steady state of system. Moreover, for nonlinear models of inverter parallel systems, the identification accuracy is extremely low, which is difficult to support the control requirements of complex scenarios. EKF linearizes the nonlinear function by first-order Taylor expansion and ignores higher-order terms. When the nonlinearity of the system is high, it will introduce significant errors, resulting in insufficient estimation accuracy. Although UKF avoids function approximation through Sigma point propagation, it is easily affected by computer or CPU rounding errors, which can lead to non-positive definite noise covariance matrix, causing covariance degradation and ultimately causing algorithm collapse. Another type of fixed virtual impedance or manual parameter adjustment scheme, although simpler to implement in engineering, belongs to the open-loop or static compensation method based on a fixed model. It lacks the ability to adapt to sudden changes in line impedance or long-term time-varying conditions. When the actual impedance deviates from the model preset value, the control performance will deteriorate sharply and cannot maintain the best compensation effect for a long time.

[0009] In addition to the core defects in impedance identification and compensation mentioned above, existing control technologies also have the following shortcomings: First, they have weak dynamic response capabilities. Some schemes, due to a lack of virtual inertia or insufficient inertia, will generate large current surges when the load changes abruptly, affecting system stability. Second, they lack accuracy and robustness. Under complex switching noise and nonlinear load disturbances, existing online impedance identification methods are difficult to achieve high accuracy and fast convergence, and cannot effectively resist external interference. Third, they lack adaptive capabilities. The system cannot track the slow time-varying characteristics of line impedance caused by factors such as temperature rise and aging. After long-term operation, the initial compensation strategy will gradually fail, and the circulating current problem and uneven power distribution problem will reappear.

[0010] In summary, existing UPS inverter parallel control technologies lack the ability to online and accurately sense and dynamically compensate for line impedance, thus failing to fundamentally solve a series of problems caused by impedance mismatch. This seriously affects the operating efficiency, reliability, and service life of parallel systems. Therefore, there is an urgent need for a new control scheme that can overcome the limitations of existing technologies to meet the high-performance requirements of UPS parallel systems in critical power supply scenarios. Summary of the Invention

[0012] The purpose of this invention is to address the shortcomings of existing technologies by providing a circulating current suppression and current sharing control method for UPS parallel systems, thereby solving a series of problems caused by impedance mismatch in UPS parallel systems and improving the system's operating efficiency, reliability, and service life.

[0013] The problem described in this invention is solved by the following technical solution:

[0014] A circulating current suppression and current sharing control method for a parallel UPS system is disclosed. The method embeds an adaptive state observer and an adaptive virtual impedance compensator into the digital controller of each parallel UPS inverter. The adaptive state observer is functionally used to identify the equivalent line impedance of the inverter's output circuit with high accuracy online. Its core components include a parameter tracking algorithm based on an unscented Kalman filter framework. The adaptive virtual impedance compensator is functionally used to calculate a feedforward compensation amount to correct the voltage command based on the impedance identification result and the real-time output current. Its core components include a dynamic virtual impedance assignment unit and a vector subtractor. Then, each UPS inverter is controlled according to the following steps:

[0015] a. Real-time acquisition of the output phase current signal of each inverter using current Hall sensors and voltage Hall sensors. oabc and output voltage signal u oabc ;

[0016] b. The adaptive state observer bases its data on the output phase current signal i. oabc and output voltage signal uoabc The signal is used to identify the real-time impedance estimate Z of the inverter output circuit online. est =[R est ,L est ], where R est This is the real-time resistance estimate, L est It is a real-time inductance estimate;

[0017] c. The adaptive virtual impedance compensator dynamically adjusts the real-time impedance estimate Z. est =[R est ,L est The virtual impedance Z assigned to this UPS inverter virtual And calculate the output current I in real time. o In virtual impedance Z virtual The theoretical pressure drop ΔV=I generated above o ·Z virtual ;

[0018] d. The adaptive virtual impedance compensator references the ideal voltage output from the pre-amplifier controller of the UPS inverter to the command E. ref The voltage reference command U is generated by vector subtraction of the theoretical voltage drop ΔV. ref ;

[0019] e. Set the voltage reference command U ref The voltage and current dual closed-loop PID controller fed into the UPS inverter generates a PWM signal to drive the inverter bridge, thus completing the closed-loop control of the entire UPS parallel system.

[0020] The circulating current suppression and current sharing control method for the above-mentioned UPS parallel system uses an adaptive state observer that integrates fading factor and square root filtering techniques as its adaptive state observer.

[0021] The circulating current suppression and current sharing control method for the above-mentioned UPS parallel system, specifically the steps for the adaptive state observer to identify the real-time impedance estimate of the inverter output circuit are as follows:

[0022] S1: Initialization, setting the initial value of the state vector X0 and the initial value of the error covariance matrix P0, where the state vector includes the filter inductor current and the line resistance and inductance to be identified;

[0023] S2: Sigma point sampling, based on the current state estimate X. k and its covariance P k According to the unscented transformation rule, 2n+1 sampling points are deterministically calculated, where n is the dimension of the state vector. These sampling points are used to accurately capture the probability distribution of the nonlinear characteristics of the system.

[0024] S3: Time prediction. The Sigma points are propagated through the nonlinear state equations characterizing the system dynamics. The propagated point set is weighted and summed to calculate the one-step prediction value of the state. With the prediction error covariance matrix ;

[0025] S4: Introduce a fading factor to reduce the prediction error covariance matrix. With a fading factor greater than or equal to 1 Multiplication is an operation designed to increase prediction uncertainty and prevent the filter gain from converging prematurely, thereby giving the observer the ability to track parameters that are slowly changing over time.

[0026] S5: Covariance square root update, adopts numerically stable square root filtering methods such as QR decomposition and Cholesky factor update to process the predicted covariance after the fading factor adjustment, so as to ensure that it always maintains positive semidefiniteness in the iterative calculation and fundamentally avoids numerical divergence.

[0027] S6: State Update and Output. Combining the new system measurements, the predicted state is optimally corrected using Kalman gain to obtain the current state estimate. With covariance Finally, the line resistance is extracted from the updated state vector. With inductance Real-time estimated value;

[0028] S7: Iterative recursion, let The steps S2 to S6 are repeated to achieve online, continuous, and adaptive identification of the line impedance.

[0029] In the above-mentioned circulating current suppression and current sharing control method for UPS parallel systems, the UPS inverter front-end controller is a virtual synchronous machine controller or a droop controller.

[0030] Beneficial effects

[0031] The present invention has the following beneficial effects:

[0032] 1) High precision and high reliability:

[0033] This invention achieves online and accurate sensing of line impedance through an "adaptive state observer" and actively compensates for the voltage drop caused by impedance mismatch at the control command level through "feedforward compensation." This is fundamentally different from traditional methods that rely on error feedback for passive adjustment.

[0034] Based on circuit theory, when the output voltage vector difference between two inverters at the common point is eliminated, the circulating current can theoretically be reduced to zero. Therefore, this invention can significantly reduce the circulating current in a parallel system and greatly improve the power sharing accuracy. Based on the theoretical analysis and simulation verification of this control method, it is expected that the effective value of the circulating current can be suppressed to within 2% of the rated output current, and the power unevenness can be controlled to within 3%, far exceeding the traditional droop control scheme.

[0035] 2) Excellent dynamic stability and robustness

[0036] This invention employs the "square root of fading factor unscented Kalman filter" framework. Its numerical stability ensures that the observer will not diverge during long-term operation in embedded systems. The introduction of the "fading factor" endows the system with the ability to track slowly changing impedance. This allows the system to maintain excellent performance even when faced with scenarios involving slow impedance changes such as line aging, contact resistance variations, and ambient temperature fluctuations. Simultaneously, the UKF's ability to handle nonlinearity ensures that the system maintains identification accuracy and control stability even under nonlinear load disturbances (such as rectifiers).

[0037] 3) Strong engineering applicability and versatility

[0038] The core algorithm of this invention is embedded in the control chip as a software upgrade, requiring no additional hardware costs. Furthermore, the "adaptive virtual impedance compensator," as an independent module, can work in conjunction with various front-end control strategies such as virtual synchronous machines and droop control. This means that the technology can be seamlessly integrated into the company's existing and future UPS product platforms, significantly reducing the complexity and cost of technology upgrades. It provides a universal and efficient solution for improving the parallel performance of the entire product line.

[0039] 4) A prototype of predictive maintenance has been achieved.

[0040] The high-precision line impedance value continuously output by this invention is itself a highly valuable system status monitoring indicator. Maintenance personnel can monitor the changing trend of this impedance value to detect potential faults such as loose line connections and aging contact points in advance, thereby upgrading system maintenance from "reactive repair" to "predictive maintenance," further ensuring the reliability of the power supply system.

[0041] In summary, this invention achieves accurate online sensing of line impedance through an adaptive state observer and generates compensated voltage reference commands through an adaptive virtual impedance compensator, effectively solving the impedance mismatch problem in UPS parallel systems and greatly improving the operating efficiency, reliability, and service life of UPS parallel systems. Attached Figure Description

[0043] The invention will now be described in further detail with reference to the accompanying drawings.

[0044] Figure 1 This is the equivalent circuit diagram of inverters in parallel;

[0045] Figure 2 This is a diagram showing the system configuration and signal flow of the present invention (only one UPS branch is shown in the diagram).

[0046] Figure 3 This is the line impedance identification process of the present invention;

[0047] Figure 4 This is a flowchart illustrating the technical implementation of the present invention.

[0048] The symbols in the image and text are: It is the first UPS inverter. It is the output current of the first UPS inverter. It is the active power of the first UPS inverter. It is the reactive power of the first UPS inverter. It is the inductive reactance of the first UPS inverter branch. It is the resistor in the first UPS inverter branch. It is the second UPS inverter. It is the output current of the second UPS inverter. It is the active power of the second UPS inverter. It is the reactive power of the second UPS inverter. It is a parallel circulation. It is the resistor in the second UPS inverter branch. It is the inductive reactance of the second UPS inverter branch. It is the system resistance. It is systemic resistance. It is the DC input voltage of the UPS inverter. It is the DC current of the UPS inverter. It is the branch resistance. It is a branch resistance. It is the current flowing through the filter inductor in the ABC stationary coordinate system. It is the output phase current signal. It is the output voltage signal. It is the output current in the dq coordinate system. It is the output voltage in the dq synchronous rotating coordinate system. It is the output current in the αβ coordinate system, R est This is the real-time resistance estimate, L est It is a real-time inductance estimate. and These are the virtual resistance and virtual inductance values ​​obtained after applying the virtual impedance algorithm. is the inductor current in the dq coordinate system, is the output voltage in the dq coordinate system, is the angular frequency, is the reference ideal output electromotive force, Z est is the real-time impedance estimation value, E ref is the ideal voltage reference command, Z virtual is the virtual impedance, I o is the output current, ΔV is the theoretical voltage drop, U ref is the voltage reference command, Z line is the actual line impedance, λ is the fading factor, P is the active power of the branch, Q is the reactive power of the branch, is the initial value of the estimated value, is the initial covariance, is the prior estimated value, is the covariance corresponding to the prior estimated value. Detailed implementation manner

[0050] Aiming at the deficiencies of the existing UPS parallel system, the present invention provides a method for suppressing circulating current and current sharing that can accurately identify the line impedance online and perform dynamic compensation in real time. This method constructs an intelligent closed-loop of "sensing - decision - execution" to actively offset the influence of the mismatch of the physical line impedance, thereby fundamentally suppressing the parallel circulating current and achieving accurate power sharing. The "sensing - decision - execution" intelligent closed-loop constructed by the present invention is not an abstract concept, but dynamically operates within the control cycle of each parallel UPS inverter through the following specific and realizable steps:

[0051] 1. Sensing link: High-precision state observation

[0052] This link corresponds to the operation of the adaptive state observer (the core is the A-SRUKF algorithm), aiming to solve the problem of "what is the current state of the system".

[0053] The specific steps are as follows:

[0054] Data acquisition: Real-time acquisition of the voltage u abc and current i oabc signals at the output end of the inverter.

[0055] State Prediction and Update: Based on the augmented state-space model, the A-SRUKF algorithm performs a series of recursive calculations, including Sigma point sampling, nonlinear propagation, fading factor adjustment, and square root update. This algorithm acts like an "intelligent scout," penetrating strong switching noise to accurately perceive the critical state of the physical line impedance [R,L], which cannot be directly measured. The fading factor endows it with "memory decay" characteristics, enabling it not only to perceive the current state but also to keenly detect the slow time-varying trend of the impedance, achieving continuous tracking.

[0056] 2. Decision-making process: Dynamic optimization settings

[0057] This step corresponds to the key instruction of "assigning the identification results to the virtual impedance", which aims to solve the problem of "how to adjust".

[0058] The specific steps are as follows:

[0059] Dynamic mapping: This involves mapping the impedance estimate Z output in real time from the sensing element. est The virtual impedance Z within the control algorithm is directly and instantly assigned. virtual That is, let Z virtual =Z est This is not a simple signal transmission, but an autonomous decision-making process based on real-time intelligence. The core of the system's decision-making is: "To make my virtual output characteristics completely equivalent to the real physical circuit characteristics that I perceive." This decision enables the equivalent impedance exhibited by the control system to automatically match the ever-changing physical impedance, providing the correct "action target" for the precise execution of the next step.

[0060] 3. Execution phase: Feedforward precision compensation

[0061] This step corresponds to the work of the "adaptive virtual impedance compensator," which aims to solve the problem of "how to achieve adjustment."

[0062] The specific steps are as follows:

[0063] Pressure drop calculation: based on the Z-value set in the decision-making process. virtual and the real-time measured output current I o Based on the vector form of Ohm's law, the theoretical pressure drop ΔV=I is calculated precisely. o ·Z virtual .

[0064] Instruction correction: The ideal voltage instruction E generated by the upper-level controller (such as VSG / droop control) is modified accordingly. ref The vector subtraction of this vector subtraction with the calculated ΔV yields the final executed voltage command U. ref =E ref-ΔV. This achieves "feedforward compensation." The system calculates and cancels the effect of line voltage drop the instant the voltage command is issued. It actively eliminates the potential difference caused by circulating current at its source, rather than passively adjusting after the circulating current is generated, achieving an intelligent leap from "feedback correction" to "feedforward prevention."

[0065] The above three steps cycle repeatedly within each control cycle (typically tens to hundreds of microseconds):

[0066] 1) After executing the new instruction, output the new current and voltage.

[0067] 2) New currents and voltages are captured by the sensing element and used to update the understanding of impedance and environmental conditions.

[0068] 3) The updated understanding triggers new decisions, which in turn optimizes the execution of the next cycle.

[0069] This closed loop enables the system to become an intelligent agent with the ability to "self-perceive, self-decision-make, and self-optimize," dynamically adapting to complex operating environments. Ultimately, it theoretically and proactively counteracts the effects of physical impedance mismatch, achieving fundamental suppression of parallel circulating currents and precise power distribution.

[0070] The innovation of this invention lies in the control algorithm level. Its main feature is the embedding of two functional modules—an adaptive state observer and an adaptive virtual impedance compensator—in the digital controller of each parallel UPS inverter. This is achieved by executing a specific algorithm program. The system configuration and signal flow are as follows: Figure 2 As shown in the attached diagram above, and further details are provided in the accompanying description. Figure 2 An explanation was provided.

[0071] Its core algorithm follows a closed-loop logic of "observation-decision-compensation," specifically manifested in the following steps:

[0072] Observation: Inverter output voltage u obtained from sampling oabc With output current i oabc In the αβ stationary coordinate system, an unscented Kalman filter algorithm with a fading factor is used to perform online, high-precision, and robust estimation of the augmented state vector, including line resistance and inductance. The key steps of the square root unscented Kalman filter recursion with a fading factor include:

[0073] 1. Sigma point sampling: In the formula , where refers to the covariance matrix of the current state.

[0074] 2. Gradual disappearance: .

[0075] 3. Square root update: QR decomposition and Cholesky decomposition are used to update the covariance square root.

[0076] 4. Measurement update, output optimal state estimate , which includes R est With L est .

[0077] 5. Decision: The line impedance value [R] estimated in real time in step 1 is used as the basis for the decision. est ,L est The virtual impedance of the inverter is dynamically assigned, i.e., Z is set to... virtual =R est +jωL est .

[0078] 6. Compensation: Based on the virtual impedance Z virtual With real-time output current I o Calculate the theoretical pressure drop And compare it with the ideal voltage reference command E generated by the upstream controller (such as droop control or VSG). ref Perform vector subtraction to obtain the compensated voltage command U. ref =E ref -ΔV is used to drive subsequent PWM modulation.

[0079] 7. Signal Acquisition Terminal: The input terminal of the adaptive state observer is connected to the current Hall sensor and voltage Hall sensor on the inverter output side via the ADC (Analog-to-Digital Converter) of the DSP main control chip, for real-time acquisition of the output phase current signal i, which characterizes the actual operating state of the system. oabc With output voltage signal u oabc .

[0080] Core processing and collaboration relationships:

[0081] Adaptive state observer: Receives the above output phase current signal i oabc With output voltage signal u oabc Its function is to calculate and output the real-time impedance estimate Z of the current physical circuit with high accuracy and robustness under strong PWM switching noise and load disturbance. est =[R est ,L est Based on the unscented Kalman filter, to ensure the system can still track the slow time-varying line impedance parameters caused by component aging and temperature rise near steady state, a square root filter is introduced to prevent the prediction covariance matrix from approximating as non-positive definite due to rounding errors by the computer or CPU, which could cause the UKF algorithm to crash. The overall line impedance identification process is as follows: Figure 3 As shown. This part is Figure 2 The functional description of the SRUKF square root unscented Kalman filter module is like... Figure 2 The signal flow direction in the inverter is used to control the output voltage and current. After coordinate transformation, and As input to the square root unscented Kalman filter algorithm with a fading factor, and then according to Figure 3 The calculation process yields the final result. The "nonlinear model" is a differential equation describing the LC filter and circuitry of the UPS inverter, elevating the unknown parameters (R, L) to state variables, related to the circuit current (i...). α i β Estimate together. Estimate the Sigma points (a set of [i] points). α i β Substituting the guessed values ​​of [R,L] into this equation allows each guess to evolve independently and precisely according to physical laws, taking a small step (one control cycle) and observing its evolution. By observing the results of this large set of guesses, UKF can more accurately deduce what the true impedance value [R,L] should be. This is called propagation (deduction) through a nonlinear model.

[0082] Adaptive virtual impedance compensator: It receives commands from two directions, one being the ideal voltage reference command E from the upstream controller (traditional droop controller). ref This instruction determines the system's base output voltage and frequency. Secondly, it provides the real-time impedance estimate [R] from the adaptive state observer. est ,L est ].

[0083] The working mechanism of the compensator: It dynamically adjusts [R] est ,L est Assign a value to the virtual impedance Z virtual And calculate the output current I in real time. o The theoretical voltage drop ΔV=I across this virtual impedance o ·Z virtual Subsequently, it will use the ideal instruction E generated by the previous level controller. ref The calculated voltage drop ΔV is subtracted from the vector value to generate the final, compensated voltage reference command U. ref .

[0084] Command output terminal: The output terminal of the adaptive virtual impedance compensator will generate U ref The command is sent to the voltage and current dual closed-loop PID controller in the subsequent stage, which finally generates the PWM signal to drive the inverter bridge, thus completing the closed-loop control of the entire system.

[0085] The specific implementation process is as follows: Figure 4 As shown.

[0086] Compared with existing technologies (such as fixed-parameter virtual impedance or schemes based on traditional identification algorithms), the essential difference of this invention lies in constructing a cooperative control architecture with "self-sensing" and "dynamic compensation" capabilities, specifically reflected in:

[0087] 1) From "static preset" to "dynamic adaptive":

[0088] Existing technology: Virtual impedance Z virtual It is usually calculated offline and uses a pre-set fixed value. It cannot sense the actual line impedance Z. line The differences and time-varying nature of these factors mean that the compensation effect is limited and will deteriorate over time.

[0089] This invention: Virtual impedance Z virtual Z is a dynamic variable driven online and in real time by an observer. virtual =Z est ≈Z line This enables the system to automatically adapt to impedance mismatches caused by various reasons, achieving a leap from "open-loop compensation" to "closed-loop self-adaptation".

[0090] 2) From "rough estimation" to "precise perception":

[0091] Existing technologies may employ simple recursive least squares methods, but in complex nonlinear and noisy environments like UPS systems, it is often difficult to simultaneously achieve high accuracy, convergence speed, and robustness.

[0092] This invention employs a specially designed square-root unscented Kalman filter with a fading factor as the core of the observer. By applying the fading factor λ (λ≥1) to the prediction covariance matrix, the uncertainty of the prior estimate is artificially increased. This effectively prevents the filter from "gain shrinking" in steady state, forcing it to maintain sensitivity to newly generated measurement data, thus endowing the algorithm with the crucial ability to track slowly time-varying line impedances (such as those caused by temperature rise or aging). This design simultaneously solves three major challenges: nonlinear accuracy, numerical stability, and parameter tracking, providing sensing accuracy and reliability far exceeding traditional methods.

[0093] 3) From "passive flow equalization" to "active error reduction":

[0094] Existing technologies such as droop control belong to "passive flow equalization," which means that the system passively adapts to the existence of circulating flow by adjusting its own output in order to achieve balance. This is a type of regulation with static error.

[0095] This invention: via a feedforward U ref =E ref-ΔV compensation proactively and preemptively counteracts the root cause of circulating current—line voltage drop—at the control command level. This is an "elimination" rather than a "regulation" at the source, and theoretically, it can achieve zero steady-state error control.

[0096] Explanation of relevant professional terms

[0097] Circulating current: refers to the current that flows between parallel inverters but does not flow through the load; it is ineffective current.

[0098] The fading factor is a scalar coefficient greater than or equal to 1 used to amplify the covariance matrix in the prediction step of the Kalman filter to maintain the filter's sensitivity to new data and prevent it from "falling asleep".

[0099] Square root unscented Kalman filter: an algorithm that achieves filtering by directly maintaining and updating the square root of the covariance matrix (such as the Cholesky factor), which can fundamentally solve the problem of non-positive definite covariance matrix caused by rounding errors and ensure numerical stability.

[0100] Virtual impedance: A virtual impedance introduced in the control algorithm. Its voltage drop is calculated by software and the output voltage command is corrected to change the equivalent output impedance.

[0101] Figure 2 This is a diagram showing the system configuration and signal flow of the present invention (only one UPS branch is shown in the diagram).

[0102] Figure 2 The DC power source provides DC power to the three-phase full-bridge inverter. The DC power is converted into three-phase AC power by the three-phase full-bridge inverter. The three-phase power output from the inverter passes through an LC filter to remove switching frequency harmonics, resulting in a high-quality sine wave. This sine wave voltage passes through the output line impedance of each unit and finally reaches the common AC bus, jointly powering the downstream AC loads.

[0103] The adaptive state observer (i.e., the square root unscented Kalman filter module) uses the output voltage signal u collected by the voltage sensor and current sensor on the output side of the inverter. abc and output current signal i oabc As input, the observer, in the αβ stationary coordinate system, uses the established augmented state-space model and the square root unscented Kalman filter algorithm to optimally estimate the state vector containing line resistance and inductance, and finally outputs the current line impedance [R]. est ,L est The real-time identified value of the line impedance [R]. est ,L est Assign virtual impedance Z virtual And based on Ohm's law in vector form, according to the output current Io Real-time calculation of its Z virtual The theoretical voltage drop ΔV generated is calculated in either the αβ stationary coordinate system or the dq synchronous rotating coordinate system, and the result is a voltage vector containing amplitude and phase information.

[0104] Based on the measured output active power P and reactive power Q, and according to the inherent droop characteristics or rotor motion equations, the ideal voltage reference command E required by the system is calculated. ref ∠δ. The goal of this instruction is to adjust the frequency and voltage amplitude to achieve macroscopic power distribution.

[0105] Then issue the ideal command E for droop control. ref Subtracting the line voltage drop ΔV predicted by the impedance identification module, we obtain the final voltage command U sent to the underlying pulse width modulation driver. ref .

[0106] The final voltage command is then applied to the voltage-current dual closed-loop model to generate SPWM modulation, which is fed back to the three-phase full-bridge inverter for switching regulation, forming a closed loop.

Claims

1. A method for circulating current suppression and current sharing control in a parallel UPS system, characterized in that, The method embeds an adaptive state observer and an adaptive virtual impedance compensator in the digital controller of each parallel UPS inverter, and then controls each UPS inverter according to the following steps. a. Real-time acquisition of the output phase current signal of each inverter using current Hall sensors and voltage Hall sensors. oabc and output voltage signal u oabc； b. The adaptive state observer bases its data on the output phase current signal i. oabc and output voltage signal u oabc The signal is used to identify the real-time impedance estimate Z of the inverter output circuit online. est =[R est ,L est ], where R est This is the real-time resistance estimate, L est It is a real-time inductance estimate; c. The adaptive virtual impedance compensator dynamically adjusts the real-time impedance estimate Z. est =[R est ,L est The virtual impedance Z assigned to this UPS inverter virtual And calculate the output current I in real time. o In virtual impedance Z virtual The theoretical pressure drop ΔV=I generated above o ·Z virtual ; d. The adaptive virtual impedance compensator references the ideal voltage output from the pre-amplifier controller of the UPS inverter to the command E. ref The voltage reference command U is generated by vector subtraction of the theoretical voltage drop ΔV. ref ; e. Set the voltage reference command U ref The voltage and current dual closed-loop PID controller fed into the UPS inverter generates a PWM signal to drive the inverter bridge, thus completing the closed-loop control of the entire UPS parallel system.

2. The circulating current suppression and current sharing control method for a UPS parallel system according to claim 1, characterized in that, The adaptive state observer is an unscented Kalman filter observer that integrates fading factor and square root filtering techniques.

3. The circulating current suppression and current sharing control method for a UPS parallel system according to claim 2, characterized in that, The specific steps for the adaptive state observer to identify the real-time impedance estimate of the inverter output circuit are as follows: S1: Initialization, setting the initial value of the state vector X0 and the initial value of the error covariance matrix P0, where the state vector includes the filter inductor current and the line resistance and inductance to be identified; S2: Sigma point sampling, based on the current state estimate X. k and its covariance P k According to the unscented transformation rule, 2n+1 sampling points are deterministically calculated, where n is the dimension of the state vector. These sampling points are used to accurately capture the probability distribution of the nonlinear characteristics of the system. S3: Time prediction. The Sigma points are propagated through the nonlinear state equations characterizing the system dynamics. The propagated point set is weighted and summed to calculate the one-step prediction value of the state. With the prediction error covariance matrix ; S4: Introduce a fading factor to reduce the prediction error covariance matrix. With a fading factor greater than or equal to 1 Multiplication is an operation designed to increase prediction uncertainty and prevent the filter gain from converging prematurely, thereby giving the observer the ability to track parameters that are slowly changing over time. S5: Covariance square root update, adopts numerically stable square root filtering methods such as QR decomposition and Cholesky factor update to process the predicted covariance after the fading factor adjustment, so as to ensure that it always maintains positive semidefiniteness in the iterative calculation and fundamentally avoids numerical divergence. S6: State Update and Output. Combining the new system measurements, the predicted state is optimally corrected using Kalman gain to obtain the current state estimate. With covariance Finally, the line resistance is extracted from the updated state vector. With inductance Real-time estimated value; S7: Iterative recursion, let The steps S2 to S6 are repeated to achieve online, continuous, and adaptive identification of the line impedance.

4. The circulating current suppression and current sharing control method for a UPS parallel system according to claim 1, characterized in that, The UPS inverter front-end controller is a virtual synchronous machine controller or a droop controller.

Images