A bypass backfeed detection method for an ups

Abstract

The present application relates to the bypass back-feeding detection technical field of UPS, and particularly relates to a bypass back-feeding detection method of UPS, comprising the following steps: inverter switch transient state, capturing the high-frequency transient response of the inverter output end and the bypass input end in the microsecond level window; calculating the transient transfer entropy from the inverter end to the bypass end as the causal coupling parameter; the inverter and the bypass are not electrically connected directly, which is the first causal state; there is an electrical direct connection and forms a metal path, which is the second causal state; monitoring whether a step mutation occurs; if the step mutation occurs and meets the preset stability criterion, it is determined that bypass back-feeding occurs. The present application uses the inverter switch transient state as a natural excitation source, realizes passive bypass back-feeding detection, detects the essential change from parasitic coupling to metal direct connection through direct detection of the circuit topology, and does not rely on the surface characteristics such as signal amplitude and frequency which are easily disturbed, fundamentally eliminating the misjudgment caused by external factors such as power grid fluctuation and load mutation.

Description

Technical Field

[0001] This invention relates to the field of bypass backflow detection technology for UPS, and more particularly to a bypass backflow detection method for UPS. Background Technology

[0002] A UPS is a device that continuously provides stable power to a load through an inverter or bypass when the power grid fails or the power quality deteriorates. The bypass is a backup power supply path in a UPS that directly connects the power grid input terminal to the load terminal. During normal operation, it is powered by the inverter. When the inverter is overloaded or fails, it switches to bypass power supply. However, the bypass switch or static switch may accidentally conduct under abnormal conditions, causing the power output of the inverter to flow back into the bypass input terminal. This phenomenon is called bypass backflow. Bypass backflow not only causes energy loss, but may also cause grid-side protection malfunctions, equipment damage, or even personal safety accidents. Therefore, timely and accurate detection of bypass backflow is of great engineering significance.

[0003] Chinese Patent Publication No. CN118169602A discloses a bypass backflow detection method, controller, UPS, and storage medium for a UPS. The method includes: upon receiving a target detection signal, controlling the target parameter of the UPS inverter output to change according to a preset trend; the target detection signal is used to indicate bypass backflow detection of the UPS; detecting whether the changing trend of the target parameter of the UPS bypass input is the same as the preset trend; if they are the same, determining that bypass backflow has occurred in the UPS.

[0004] It is evident that the aforementioned existing technologies avoid external stimuli, but mainly rely on superficial features such as signal amplitude and frequency for threshold comparison, which is easily affected by external factors such as power grid voltage fluctuations and load changes, resulting in a high misjudgment rate. Summary of the Invention

[0005] To address this issue, the present invention provides a bypass backflow detection method for UPS, which overcomes the problem in the prior art that relies on superficial features such as signal amplitude and frequency for threshold comparison, which is easily affected by external factors such as grid voltage fluctuations and load changes, resulting in a high false positive rate.

[0006] To achieve the above objectives, the present invention provides a bypass backflow detection method for a UPS, comprising the following steps: Using the inverter switching transient as a natural excitation, the high-frequency transient response of the inverter output and bypass input is synchronously captured within a microsecond time window under the triggering of the switching edge. Based on the dual-ended signal within the microsecond-level time window, the transient transfer entropy or conditional Granger causality strength from the inverter end to the bypass end is calculated as a causal coupling parameter for quantifying the driving strength of the switching event to the bypass response, and the causal coupling parameter is output as a topological coupling strength measurement value characterizing the tightness of the electrical connection between the inverter end and the bypass end. There is no direct electrical connection between the inverter and the bypass. The signals at both ends are only weakly causally correlated by parasitic stray coupling, which is defined as the first causal state. There is an electrical direct connection between the inverter and the bypass, forming a metal path. The transient signal of the inverter can generate a stable strong causal drive on the bypass, which is defined as the second causal state. Based on the causal coupling parameters, monitor whether a step transition occurs from the first causal state to the second causal state; If the step change is detected and the preset stability criterion is met, it is determined that bypass backfeeding has occurred.

[0007] Furthermore, the baseline of the first causal state is established based on the null hypothesis model. Under the full operating condition of confirming no bypass backflow, the empirical zero distribution of the causal coupling parameters is statistically analyzed, and a rejection region is constructed based on a preset significance level as a statistical criterion to distinguish between weak causal pseudo-associations and strong causal drives.

[0008] Furthermore, before calculating the causal coupling parameters, the dual-end signals within the microsecond-level time window are subjected to adaptive power frequency notch filtering and steady-state harmonic removal, and are approximated as discrete sequences through symbolization or segmented aggregation to eliminate the interference of random noise on transient causal calculation.

[0009] Furthermore, based on the reverse causal strength of the bypass end to the inverter end calculated synchronously, when the forward causal strength undergoes a statistically significant step change while the reverse strength remains unchanged, and the ratio of the forward to reverse strength exceeds a preset asymmetry threshold, it is confirmed as a reverse-feeding exclusive unidirectional causal topology of the inverter actively driving the bypass.

[0010] Furthermore, the determination process of the preset stability criterion is as follows: Over multiple consecutive switching cycles, the causal coupling parameter consistently rejects the null hypothesis, and the fluctuation of the parameter value between adjacent cycles is lower than a preset steady-state threshold.

[0011] Furthermore, based on the real-time collected inverter DC bus voltage, load current, and junction temperature parameters, the causal parameters in the first causal state are compensated through a pre-built operating condition-parasitic coupling mapping model to remove the influence of gradual operating condition changes on parasitic coupling, retaining only the causal step component caused by direct electrical connection.

[0012] Furthermore, when multiple inverters are connected in parallel, the causal intensity of the bypass response of each excitation source is calculated based on the independent switching transient of each inverter as different excitation sources, and the inverter unit that has experienced backfeeding is located based on the excitation source where the intensity undergoes a step change.

[0013] Furthermore, after determining that bypass backflow has occurred, the bypass switch is tripped based on the protection command, and the original waveforms, causal parameter sequences, and null hypothesis rejection domain boundaries before and after the abrupt change are frozen to generate an unalterable fault snapshot.

[0014] Furthermore, during the operation period when the system confirms no bypass backflow, the null hypothesis baseline model corresponding to the first causal state is smoothly updated based on the continuously collected parasitic coupling transient data, in order to adapt to the slow drift of parasitic coupling caused by device aging and environmental changes.

[0015] Furthermore, the calculation of the causal coupling parameters, state differentiation, and abrupt change determination are all performed online during the normal power supply operation of the inverter, and the selection of the microsecond-level time window is adaptively matched with the attenuation duration of the switching transient oscillation.

[0016] Compared with the prior art, the beneficial effects of the present invention are that by utilizing the inverter's own switching transients as a natural excitation source, the present invention achieves completely passive bypass backflow detection without any external detection excitation equipment. Compared with the existing active disturbance detection methods, the detection process has zero interference to the normal power supply of the UPS, no perception at the load end, and the power supply quality indicators such as output voltage, frequency, and harmonic content are not affected in any way. At the same time, by directly detecting the essential change in circuit topology from parasitic coupling to direct metal connection—that is, the step change from weak causal pseudo-correlation to strong causal drive—rather than relying on easily interfered surface features such as signal amplitude and frequency, the present invention fundamentally eliminates misjudgments caused by external factors such as grid fluctuations and load changes.

[0017] Furthermore, by constructing a bidirectional causal asymmetry verification mechanism, this invention can reliably distinguish between two fundamentally different physical scenarios: inverter reverse-feeding bypass and synchronous coupling of grid disturbances. In terms of circuit physics, reverse-feeding bypass is a unidirectional causal topology where the inverter actively drives the bypass through a metal direct-connection circuit, while grid disturbances are a bidirectional coupling relationship that acts simultaneously on both ends. By synchronously calculating the positive and negative causal strengths and verifying the asymmetry ratio, this invention effectively solves the problem that transient disturbances such as grid voltage drops and flicker may cause synchronous fluctuations in the signals at both ends, which may be misjudged as reverse-feeding. This significantly improves the detection accuracy under complex grid operating conditions.

[0018] Furthermore, this invention establishes a statistical judgment framework based on a null hypothesis model and rejection region, replacing empirical fixed thresholds with quantifiable confidence levels, thereby achieving the scientific and standardized nature of the judgment criteria. At the same time, through a life-cycle unsupervised adaptive update mechanism, the benchmark model can automatically follow the slow drift of parasitic parameters caused by device aging and environmental changes, maintaining detection sensitivity without decay over the long term. This avoids the problem of the misjudgment rate increasing year by year due to equipment aging in traditional fixed threshold schemes, ensuring the consistency of detection accuracy throughout the entire life cycle of the equipment.

[0019] Furthermore, in a multi-inverter parallel system, this invention utilizes the inherent carrier phase-shifting characteristics of each module to achieve time-division multiplexing. This allows for the differentiation of different excitation sources without the need for additional hardware, enabling precise unit-level location of backfeeding faults. Maintenance personnel can directly replace or repair faulty modules, significantly shortening fault recovery time. Simultaneously, it can distinguish between single-unit backfeeding and multi-unit synchronous backfeeding conditions, providing accurate evidence for root cause analysis and significantly improving the maintainability and availability of the parallel redundant system. Attached Figure Description

[0020] Figure 1 This is a general flowchart of an embodiment of the present invention; Figure 2 This is a flowchart illustrating the causal baseline self-learning process of an embodiment of the present invention. Figure 3 This is a flowchart of the bidirectional causality verification process according to an embodiment of the present invention; Figure 4 This is a flowchart of the multi-machine parallel traceability process according to an embodiment of the present invention. Detailed Implementation

[0021] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0022] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0023] Please see Figure 1 The diagram shown is a general flowchart of an embodiment of the present invention. This embodiment proposes a bypass backflow detection method for a UPS, comprising the following steps: Step 1: Using the inverter switching transient as a natural excitation, the high-frequency transient response of the inverter output and bypass input is synchronously captured within a microsecond time window under the triggering of the switching edge. Step 2: Based on the dual-ended signal within the microsecond-level time window, calculate the transient transfer entropy or conditional Granger causality strength from the inverter end to the bypass end, as a causal coupling parameter for quantifying the driving strength of the switching event to the bypass response, and output the causal coupling parameter as a topological coupling strength measurement value characterizing the tightness of the electrical connection between the inverter end and the bypass end. Step 3: There is no direct electrical connection between the inverter and the bypass, and the signals at both ends are only formed by parasitic stray coupling to form a weak causal pseudo-correlation, which is defined as the first causal state; There is a direct electrical connection between the inverter and the bypass to form a metal path, and the transient signal of the inverter can generate a stable strong causal drive for the bypass, which is defined as the second causal state. Step 4: Based on the causal coupling parameters, monitor whether a step transition occurs from the first causal state to the second causal state; Step 5: If the step change is detected and the preset stability criterion is met, it is determined that bypass backfeeding has occurred.

[0024] In the embodiment, the inverter power device is an IGBT or a MOSFET. The duration of the rising or falling edge of each switching action is on the order of tens to hundreds of nanoseconds. During this period, due to the interaction between parasitic inductance and parasitic capacitance in the circuit, a high-frequency damped oscillation with a frequency of hundreds of kilohertz to tens of megahertz is excited at the inverter output. The duration of this oscillation is on the order of microseconds.

[0025] In this embodiment, the starting point of the microsecond-level time window is the trigger moment of the switching edge of the PWM drive signal. The length of the microsecond-level time window is determined according to the actual decay time of the inverter's parasitic oscillation. The selection criterion is the time required for the oscillation amplitude to decay to a preset proportion of the initial peak value, multiplied by a preset expansion coefficient. The expansion coefficient is a value slightly greater than 1, which is used to completely capture the energy at the tail of the oscillation decay, while avoiding the introduction of invalid noise due to an excessively long time window. The capture method is to use a sampling rate of not less than twice the highest oscillation frequency to synchronously convert the voltage or current signals at the inverter output terminal and the bypass input terminal through a high-bandwidth voltage sensor or a high-frequency current probe to obtain two sets of discrete sampling sequences.

[0026] In this embodiment, the calculation process for transient propagation entropy is as follows: First, the two sets of sampling sequences within the microsecond-level time window are symbolized to obtain the inverter-side symbol sequence and the bypass-side symbol sequence. Then, based on the definition of transfer entropy, the amount of information transferred from the inverter-side to the bypass-side is calculated within the microsecond-level time window.

[0027] In this embodiment, the method for calculating the amount of information transmitted is as follows: The difference between the conditional probability distribution of the bypass symbol under known inverter symbol history and the conditional probability distribution of the bypass symbol under known only its own history is analyzed. The greater the difference, the stronger the causal drive of the inverter to the bypass.

[0028] The causal coupling parameter corresponding to the first causal state has a lower value range, while the value corresponding to the second causal state has a significantly higher value range. There is a clear step difference between the two, constituting a step change.

[0029] Specifically, this invention requires no external excitation equipment and utilizes the inverter's own switching transients as a detection probe, resulting in zero interference with the normal power supply process of the UPS. By directly detecting the essential change in circuit topology from parasitic coupling to direct metal connection, it fundamentally eliminates misjudgments caused by grid fluctuations and load abrupt changes. The causal coupling parameters themselves are directional, naturally distinguishing between inverter reverse bypass and grid disturbance reverse coupling. Each switching cycle can be tested once, achieving true real-time online monitoring.

[0030] In this embodiment, the controller adopts a DSP and FPGA collaborative architecture. The FPGA is responsible for high-speed synchronous acquisition, while the DSP is responsible for causal calculation and fault determination. The FPGA continuously monitors the rising and falling edges of the inverter's PWM drive signal, triggering one acquisition at each edge. Two analog-to-digital converters are started simultaneously at a preset high sampling rate, which are connected to the inverter's output voltage sensor and the bypass input voltage sensor, respectively. Data within a preset microsecond time window is continuously acquired to obtain two sets of discrete sampling sequences.

[0031] After the DSP acquires these two sets of sequences from the FPGA, it extracts the data within a microsecond-level time window starting from the edge trigger with a short delay and performs preprocessing: the power frequency component is filtered out by an adaptive notch filter, the low-frequency steady-state harmonics are filtered out by a high-pass filter, and the high-frequency transient oscillation component is retained. The two sets of preprocessed sequences are then symbolized: according to the signal amplitude range, the sampling points are mapped to a preset number of symbols to obtain two symbol sequences. The selection of the number of symbols needs to balance information retention and computational complexity. Based on the symbol sequences, the transfer entropy from the inverter end to the bypass end is calculated to obtain a causal coupling parameter value.

[0032] This causal coupling parameter serves as both a measurement value and a criterion: on one hand, it directly outputs a topological coupling strength measurement value to quantify the tightness of the electrical connection between the inverter and bypass terminals. A lower value indicates weaker electrical coupling at both ends, closer to a parasitic coupling state; a higher value indicates stronger electrical coupling at both ends, closer to a direct metal connection state. On the other hand, based on the step abrupt change of this parameter between the weak and strong coupling ranges, it determines whether a bypass backflow fault has occurred. This value is compared with a pre-stored first causal state benchmark threshold. If the current value is less than the benchmark threshold, it is determined to be in the first causal state, and no alarm is triggered.

[0033] The above process is repeated at each PWM switch edge. When the calculated causal coupling parameter value suddenly jumps to the high value range corresponding to the second causal state, it is marked as a suspected mutation event. The subsequent multiple switch transients are continuously tracked. If the causal coupling parameter value is stable in the second causal state range for a preset number of consecutive times, and the fluctuation between two adjacent values ​​is lower than the preset steady-state threshold, the mutation is confirmed to be stable, a bypass backflow fault is determined to have occurred, and alarm and protection signals are output.

[0034] Please see Figure 2 As shown, it is a flowchart of the causal benchmark self-learning process of an embodiment of the present invention. The benchmark of the first causal state is established based on the null hypothesis model. By statistically analyzing the empirical zero distribution of the causal coupling parameters under the full working condition of confirming no bypass backflow, and constructing a rejection region based on a preset significance level, it serves as a statistical criterion for distinguishing between weak causal pseudo-associations and strong causal drives.

[0035] In this embodiment, the process of establishing the null hypothesis baseline model is as follows: During the UPS factory testing or on-site installation and commissioning phase, confirm that the bypass switch is in a normal open state and there is no bypass backflow fault. Sequentially set the UPS to operate under various different load conditions from light load to full load, covering the main load ranges that may occur in the actual operation of the equipment. Run the UPS for a sufficient amount of time under each condition, collect the two-terminal waveform of the inverter switch transient for each time, and calculate the causal coupling parameter value for each time. Collect at least the minimum number of samples required to meet the statistical significance test for each condition.

[0036] All sample values ​​collected under all operating conditions are aggregated to form a sample set. A distribution fit test is performed to determine whether it conforms to a normal distribution. If it does, the sample mean and sample standard deviation are calculated to describe the zero distribution. If it does not, nonparametric methods such as kernel density estimation are used to obtain the probability density function.

[0037] The preset significance level is determined based on the following: during normal, fault-free operation, the probability that the causal coupling parameter accidentally exceeds the threshold due to statistical randomness must be controlled within an acceptablely low range. If the sample follows a normal distribution, the upper limit threshold of the rejection region is obtained by adding the quantile of the significance level corresponding to the standard normal distribution to the sample mean and multiplying it by the sample standard deviation. This threshold is the statistical criterion for distinguishing between the first causal state and the second causal state.

[0038] Specifically, a rigorous statistical method is used to establish a judgment benchmark, which makes the judgment criteria have a quantifiable confidence level; the benchmark model covers the full load conditions, ensuring that there is a unified judgment logic under different operating conditions; and a rejection region mechanism based on the significance level clearly distinguishes normal statistical fluctuations from real anomalies in a mathematical sense.

[0039] In this embodiment, after starting the benchmark modeling program: the electronic load is set to a low percentage of the rated power, the UPS inverter operates normally, and the system waits to reach thermal steady state. The FPGA triggers acquisition at each PWM switching edge, and the DSP calculates the causal coupling parameter value each time. To balance the amount of data and statistical representativeness, an equally spaced sampling method is used to collect a sample size that meets the statistical requirements. The electronic load is adjusted to multiple load rate points in sequence to cover the complete operating range from light load to full load. Multiple sets of data are repeatedly collected to obtain multiple sets of datasets. The multiple sets of data are merged to form a full-condition sample set. A frequency distribution histogram is plotted, and the sample mean and sample standard deviation are calculated. The normality test method is used to verify the distribution type. The rejection region threshold is calculated based on the sample mean, sample standard deviation and preset significance level. The threshold and related parameters are written into the controller's non-volatile memory as benchmark judgment parameters for the entire life cycle of the equipment.

[0040] Please see Figure 3 As shown, it is a flowchart of the bidirectional causality verification in an embodiment of the present invention. Before calculating the causal coupling parameters, the dual-end signals in the microsecond-level time window are subjected to adaptive power frequency notch filtering and steady-state harmonic removal, and are approximated as discrete sequences by symbolization or segmented aggregation to eliminate the interference of random noise on transient causal calculation.

[0041] In this embodiment, the adaptive power frequency notch filter adopts an IIR notch filter structure. The center frequency is dynamically adjusted in real time by detecting the power grid frequency through a phase-locked loop to ensure effective filtering of power frequency components within the normal fluctuation range of the power grid frequency. The selection of the notch bandwidth parameter needs to take into account both the filtering effect and the fidelity of the transient signal. Steady-state harmonic elimination is achieved through a high-pass filter, whose cutoff frequency is set in the frequency band between the lowest effective frequency of the switching transient oscillation and the highest frequency of the steady-state harmonic. Since there is a clear interval between the two in the frequency domain, it can not only completely retain the transient oscillation signal, but also effectively filter out the integer harmonics of the power frequency.

[0042] In this embodiment, the symbolization process uses the interval division method: the amplitude range of the signal within the microsecond-level time window is calculated and evenly divided into a preset number of equal-length sub-intervals. Each sub-interval corresponds to a symbol. The selection of the number of symbols is based on the following criteria: too few symbols will result in too much information loss, while too many symbols will amplify the noise effect and increase the computational complexity. The amplitude of each sampling point is mapped to the symbol corresponding to its sub-interval to obtain a symbol sequence.

[0043] Specifically, adaptive power frequency notch filtering ensures power frequency interference suppression capability under power grid frequency fluctuation environment; high-pass filter allows causal analysis to focus only on the switching transient itself; symbolization processing greatly reduces the complexity of transfer entropy calculation, while naturally suppressing the amplitude interference of random noise, because small noise fluctuations will not change the symbol assignment in most cases.

[0044] In this embodiment, the signal preprocessing steps include: The DSP reads the current grid frequency value output by the phase-locked loop, updates the notch filter coefficients to align its center frequency with the real-time grid frequency, filters out the power frequency component from the sampling points within the microsecond time window using the notch filter, and then passes the notch-filtered sequence through a high-order high-pass filter to remove low-frequency steady-state harmonics, resulting in a clean high-frequency transient oscillation waveform sequence. The amplitude of each sampling point is iterated to find the maximum and minimum values ​​to determine the amplitude range, which is then evenly divided into a preset number of sub-intervals. Each sampling point is mapped to a corresponding symbol to obtain a symbol sequence. The symbol sequences from the inverter end and the bypass end are sent to the transfer entropy calculation module. The design of the sub-interval width must ensure that random noise of normal amplitude does not cause frequent symbol jumps.

[0045] Please continue reading. Figure 3 As shown, based on the synchronous calculation of the reverse causal strength from the bypass end to the inverter end, when the forward causal strength undergoes a statistically significant step change while the reverse strength remains unchanged, and the ratio of the forward to reverse strength exceeds a preset asymmetry threshold, it is confirmed as a reverse-feeding dedicated unidirectional causal topology of the inverter actively driving the bypass.

[0046] In the embodiment, the calculation method of reverse causality strength is completely symmetrical with that of forward causality, only the input and output roles are swapped. Both are calculated synchronously within the same microsecond time window, using the same symbol sequence and calculation parameters. The statistical significance judgment criteria are consistent with the rejection region mechanism: when the forward causality strength exceeds the null hypothesis rejection region threshold, the forward mutation is determined to be significant, and the reverse is the same.

[0047] The determination of the preset asymmetry threshold is based on the analysis of circuit physical characteristics: Under actual bypass reverse feed conditions, the inverter end directly drives the bypass end through a low-impedance metal path, resulting in extremely high forward causal strength; while the reverse influence of the bypass end on the inverter end is isolated by the inverter output impedance, and the reverse causal strength remains weak. There is a difference of at least one order of magnitude between the two. Under the condition that grid disturbances simultaneously couple both ends, the forward and reverse causal strengths increase synchronously, and the ratio is close to one. The value of the asymmetry threshold is set between the high ratio under reverse feed conditions and the low ratio under grid disturbance conditions, which can be determined by measuring the distribution range of the asymmetry ratio under the two known conditions.

[0048] Specifically, bidirectional causal asymmetry verification is a reliable criterion for distinguishing between the inverter's reverse bypass and the synchronous impact of grid disturbances, effectively solving the problem of misjudgment that may be caused by transient disturbances such as grid voltage drops and flicker.

[0049] In the embodiments, Scenario 1 is a real bypass backfeed: the forward value is in the high value range and much greater than the rejection domain threshold, the reverse value is in the low value range and less than the rejection domain threshold, and the asymmetry ratio is much greater than the preset asymmetry threshold. When all three conditions are met, it is confirmed as a unidirectional causal topology of inverter actively driving bypass. Scenario 2 is a grid voltage sag disturbance: both the forward and reverse values ​​are greater than the rejection domain threshold, the asymmetry ratio is close to one and much less than the preset asymmetry threshold, it is determined to be an external grid disturbance and no backfeed alarm is triggered.

[0050] Please continue reading. Figure 1 As shown, the determination process of the preset stability criterion is as follows: Over multiple consecutive switching cycles, the causal coupling parameter consistently rejects the null hypothesis, and the fluctuation of the parameter value between adjacent cycles is lower than a preset steady-state threshold.

[0051] In this embodiment, the number of consecutive switching cycles is determined by a trade-off between the switching frequency and the detection response time: too few cycles result in insufficient anti-shake effect, while too many cycles increase the detection delay. The selection of the number of cycles should ensure that the total confirmation time can effectively filter out occasional interference without significantly delaying the protection action. The specific condition for rejecting the null hypothesis is that the causal coupling parameter value of each switching cycle is greater than the null hypothesis rejection threshold. The fluctuation between adjacent cycles is calculated by finding the maximum and minimum values ​​of the causal coupling parameter in multiple consecutive cycles and calculating the difference.

[0052] The determination of the preset steady-state threshold is based on the following: Under the actual bypass reverse irrigation operation, the fluctuations between adjacent cycles are only caused by measurement noise and calculation errors, and the fluctuation amplitude is very small; while the false jumps caused by random disturbances often manifest as drastic parameter fluctuations. The steady-state threshold is set as a proportion of the rejection region threshold, which should be able to accommodate normal small fluctuations and exclude large oscillations.

[0053] Specifically, the multi-cycle joint confirmation mechanism effectively filters out false jumps caused by single random noise; fluctuation constraints ensure the consistency of the judgment state; and the configurable number of consecutive cycles and fluctuation thresholds enable the criterion to be adapted to UPS equipment with different switching frequencies and power levels.

[0054] In this embodiment, the detection program maintains a circular buffer with a length equal to a preset number of consecutive cycles. When a calculated value is greater than the rejection region threshold, it is stored in the buffer. Subsequent values ​​are filled in sequentially until the buffer is full. Check condition one: compare each value in the buffer with the rejection region threshold one by one. If all values ​​are satisfied, condition one passes. If any value is not satisfied, the buffer is cleared and the count is restarted. Check condition two: find the maximum and minimum values ​​in the buffer and calculate the difference. If the difference is less than the steady-state threshold, condition two is satisfied. When both conditions are satisfied, the fault is confirmed. The steady-state threshold is based on the statistical value of the fluctuation of adjacent cycles measured under actual reverse irrigation conditions. This statistical value is usually much smaller than the rejection region threshold itself. The steady-state threshold is set as the upper limit of the distribution of this statistical value.

[0055] Please continue reading. Figure 1 As shown, based on the real-time collected inverter DC bus voltage, load current and junction temperature parameters, the causal parameters in the first causal state are compensated through a pre-built operating condition-parasitic coupling mapping model, so as to remove the influence of gradual operating condition changes on parasitic coupling and retain only the causal step component caused by electrical direct connection.

[0056] In this embodiment, the pre-built operating condition-parasitic coupling mapping model is a nonlinear mapping model that uses cubic polynomial regression or radial basis function neural network. The input variables are DC bus voltage, load current and power device junction temperature, and the output variable is the causal parameter offset caused by parasitic coupling under the current operating condition. This model is established by collecting sample data through various different operating condition combinations during the UPS factory test phase to confirm that there is no bypass backflow. The statistical mean of the causal coupling parameters under each operating condition is used as the benchmark value, and the mapping relationship is established by regression or fitting methods.

[0057] The operating point selection method is as follows: the DC bus voltage is divided into several equal points within the preset deviation range of the rated value, the load current is divided into several equal points within the range from light load to full load, and the junction temperature is divided into several equal points within the range from room temperature to the rated maximum operating temperature, forming a grid covering the entire operating space.

[0058] During online compensation, the current operating condition parameters are collected in real time and input into the pre-built mapping model to obtain the corresponding parasitic coupling offset. The offset is then subtracted from the originally calculated causal coupling parameters to obtain the compensated value. The compensation is only meaningful when the parameters are in the first causal state interval. When the parameters have transitioned to the second causal state interval, the compensation amount is negligible relative to the step amplitude and does not affect the fault determination.

[0059] Specifically, the operating condition compensation model eliminates the slight influence of factors such as load changes, bus voltage fluctuations, and temperature drift on parasitic coupling, so that parameter changes under normal operating conditions will not be misjudged as faults; the compensation only targets small, gradual changes in the first causal state and does not affect large step changes in the second causal state.

[0060] In this embodiment, before the load surge, the pre-built mapping model outputs a compensation value based on the current operating conditions. After compensation, the original causal coupling parameters are in the first causal state range. After the load surge, the DC bus voltage fluctuates briefly, the load current rises, and the junction temperature rises slowly. The mapping model outputs a changing compensation value. The original parameters rise slightly due to the change in operating conditions, and the compensated values ​​remain in the first causal state range.

[0061] When a real bypass backflow occurs, the original parameters abruptly change from the first causal state interval to the second causal state interval. At this time, the compensation value remains at a small level because the operating conditions have not changed drastically. After compensation, it is still in the second causal state interval and will not cover up the real fault characteristics. The training data of the pre-built mapping model comes from the factory test stage and is stored in the controller's non-volatile memory. The compensation value is obtained by directly looking up the table or interpolating during online operation.

[0062] Please see Figure 4 As shown, it is a flowchart of the multi-machine parallel tracing process of an embodiment of the present invention. When multiple inverters are connected in parallel, the independent switching transient of each inverter is used as a different excitation source. The causal intensity of the bypass response of each excitation source is calculated respectively, and the inverter unit that has reverse feed is located based on the excitation source where the intensity changes by a step change.

[0063] In this embodiment, the multi-inverter parallel system adopts an interleaved phase-shifting strategy, in which the carrier phases of adjacent modules are staggered by a fixed angle, which is equal to one complete cycle divided by the number of parallel modules. This characteristic allows the switching edges of each module to be naturally separated on the time axis, and there is a distinguishable time interval between the switching edges of any two modules. The controller assigns an independent causal state monitoring channel to each inverter, and each channel maintains the PWM edge trigger timestamp, the positive causal coupling parameter value, and the continuous anomaly counter.

[0064] The positioning logic is as follows: when the causal coupling parameter of a certain channel undergoes a step change and meets the stability criterion while other channels remain normal, it is determined that the module corresponding to that channel has experienced bypass backfeeding; if multiple channels experience step changes at the same time, it is determined that multiple inverters have experienced bypass backfeeding synchronously, and the judgment basis is the difference in the time when the abnormal counter of each channel reaches the preset number of confirmations.

[0065] Specifically, by utilizing the inherent carrier phase-shifting characteristics of the parallel system for time-division multiplexing, different excitation sources can be distinguished without adding additional hardware, thus achieving precise positioning at the unit level.

[0066] In this embodiment, multiple inverter modules of the same model operate in parallel. The PWM edges of each inverter are alternately arranged on the time axis, with sufficient time intervals between adjacent edges. The FPGA simultaneously monitors the PWM drive signal edges of all inverters, records the timestamps and corresponding numbers, and sequentially triggers acquisition and calculation at the switching edges of each inverter to update the corresponding channel status. After a period of operation, if the abnormal counter of a certain channel reaches the preset number of consecutive confirmations and meets the stability criterion first, while other channels remain normal, it is determined that the inverter corresponding to that channel has experienced bypass backfeed. If multiple channels reach the number of confirmations almost simultaneously, it is determined that multi-machine synchronous backfeed.

[0067] Please continue reading. Figure 1 As shown, after determining that a bypass backflow has occurred, the bypass switch is tripped based on the protection command, and the original waveforms, causal parameter sequences and null hypothesis rejection domain boundaries before and after the abrupt change are frozen to generate an unalterable fault snapshot.

[0068] In this embodiment, the bypass switch tripping command is a level signal sent by the controller through the digital output interface. The delay from the determination to the issuance of the tripping command is controlled within several control cycles. The bypass switch is selected from a circuit breaker with an excitation trip unit or a shunt trip accessory.

[0069] The data freeze range includes data for each preset duration before and after the mutation moment. The frozen data types include: the original high-frequency voltage waveform sampling sequence of the inverter output and bypass input; the causal coupling parameter time series corresponding to the mutation moment and several switching transients before and after it; the null hypothesis rejection region threshold and its corresponding mean, standard deviation, and significance level parameters; and the operating condition parameters at the judgment moment.

[0070] Immutability is achieved through digital signatures: all frozen data is concatenated into binary data blocks according to a predetermined format, a hash algorithm is used to calculate the digest value, and then the digest is signed using the private key stored inside the controller. The data blocks and signatures are stored together in non-volatile memory, and any modification will result in the recalculated digest value being inconsistent with the decrypted signature digest value.

[0071] Specifically, the protection action and data freeze are executed simultaneously to ensure that the circuit state at the moment of the fault is completely preserved; the tamper-proof fault snapshot provides a reliable data basis for fault root cause analysis and responsibility determination.

[0072] In this embodiment, after the stability criterion is met, the fault flag is set, the level of the controller output pin flips, driving the bypass switch trip coil to energize. The bypass switch completes mechanical disconnection to cut off the bypass path, and at the same time sends a fault message to the host computer. The DSP extracts data of a preset duration before and after the sudden change moment from the circular buffer, assembles it into a binary data block according to a predetermined format of file header, original waveform segment, causal parameter segment, reference parameter segment, and operating condition data segment, calculates the digest value using a hash algorithm, signs it with the asymmetric encryption private key in the secure storage area, and writes the data block and signature value into non-volatile memory. When viewed later, the signature is verified with the public key to confirm that the data is intact and has not been tampered with.

[0073] Please continue reading. Figure 2 As shown, during the operation period when the system confirms no bypass backflow, the null hypothesis baseline model corresponding to the first causal state is smoothly updated based on the continuously collected parasitic coupling transient data, in order to adapt to the slow drift of parasitic coupling caused by device aging and environmental changes.

[0074] In this embodiment, the smooth update uses a weighted moving average algorithm. The currently stored baseline mean and standard deviation are historical values, and the mean and standard deviation of the normal operating condition samples collected in each update cycle are new observations. The updated value is a weighted combination of historical values ​​and new observations. The selection of weight coefficients determines the model's tracking rate of recent changes, and a balance needs to be struck between tracking speed and stability.

[0075] The update cycle is set based on the following: the drift of parasitic parameters caused by device aging and environmental changes is an extremely slow process with a time scale of months. An update cycle of days is sufficient to track the drift trend and avoid interference from short-term operating condition fluctuations. The safety constraints are: only data within the period when there is no bypass backflow alarm is included in the update sample; if a backflow alarm is triggered in a certain update cycle, all data in that cycle is discarded to ensure that abnormal data under fault conditions does not pollute the baseline model. The updated baseline parameters are immediately used to recalculate the rejection domain threshold.

[0076] Specifically, the smooth update mechanism enables the judgment benchmark to automatically follow the slow characteristic drift throughout the entire life cycle of the device; the weighted moving average algorithm suppresses short-term fluctuations while tracking slow changes; and the safety constraints ensure the purity and reliability of the benchmark.

[0077] In this embodiment, the device automatically performs a baseline update according to a preset update cycle during operation: it checks the alarm records in the previous update cycle. If there is no backflow alarm, the data for that cycle is valid. It statistically analyzes all causal coupling parameter samples collected under normal operating conditions in that cycle, removes extreme sample points that exceed the preset deviation range, calculates the mean and standard deviation of the remaining samples as new observations, and updates the mean and standard deviation using a weighted moving average algorithm. Since the weight coefficient is small, the single update amplitude is extremely small, and it requires long-term accumulation to reflect the slow drift of the baseline. This drift exactly matches the natural rate of device aging, so that the judgment threshold is always kept at a reasonable statistical significance level. If a bypass backflow fault occurs in a certain update cycle, all data for that cycle is discarded, and the baseline remains unchanged.

[0078] Please continue reading. Figure 1 As shown, the calculation of the causal coupling parameters, state differentiation, and mutation determination are all performed online during the normal power supply operation of the inverter, and the selection of the microsecond-level time window is adaptively matched with the attenuation duration of the switching transient oscillation.

[0079] In this embodiment, online execution means that the entire detection process is completed as a background task during the normal operation of the inverter, without interruption or impact on the main control program's real-time closed-loop control of the inverter's output voltage and current. Specifically, the high-frequency waveform acquisition is completed in parallel by the FPGA hardware, without occupying DSP computing time; the causal calculation and judgment logic is executed in a low-priority background task of the DSP, and is completed during the idle period between two switching transients.

[0080] The adaptive matching process of the microsecond-level time window is completed during the device's first power-on self-test: First, a complete switching transient oscillation waveform is captured with a relatively long preset microsecond-level time window; the envelope of the oscillation waveform is extracted, and the time required for the amplitude to decay from the switch edge to a preset proportion of the initial peak value is calculated as the decay duration; the detection microsecond-level time window length is set as the decay duration multiplied by a preset expansion coefficient. The selection of the expansion coefficient must fully cover the effective range of the transient oscillation, while avoiding the introduction of invalid noise due to an excessively long microsecond-level time window. If parasitic parameters change due to subsequent device replacement or maintenance, the system will remeasure and update them during the next power-on self-test.

[0081] Background task timing arrangement: After the DSP completes the voltage and current closed-loop control calculation in each control cycle, it uses the remaining idle time to execute the detection algorithm. If the DSP's computing power is insufficient to complete all calculations within the specified idle time, a pipeline approach can be adopted, introducing a fixed delay of one switching cycle. The impact of this delay on the real-time performance of fault detection is negligible.

[0082] Specifically, the online execution feature ensures that the testing process has no impact on the UPS power quality indicators; the microsecond-level time window adaptive matching enables the same method to be adapted to inverters with different power levels and device parameters without the need for manual parameter adjustment.

[0083] In this embodiment, the microsecond-level adaptive matching process for time windows is as follows: Upon initial power-up, the inverter undergoes a soft start, triggering a long-time-window high-frequency acquisition and waveform data analysis. The oscillation envelope is calculated point by point from the switch edge. The envelope extraction method involves taking the absolute value of the waveform and then obtaining a smooth envelope line through a low-pass filter. After finding the peak point of the envelope, the system searches backward for sampling points where the amplitude drops to a preset percentage of the peak value. The time difference is recorded as the attenuation duration. The detection window length is equal to the attenuation duration multiplied by a preset expansion coefficient. After writing the data to non-volatile memory, a self-test is completed.

[0084] Online execution sequence: When the inverter is running normally, each PWM switching edge triggers FPGA hardware acquisition, and the time taken is equal to the time window length of the adaptive matching. The FPGA transmits the data to the DSP memory through direct memory access. After the DSP main control program completes the closed-loop control calculation, it sequentially performs signal preprocessing, symbolization processing, causal coupling parameter calculation, threshold comparison and state determination during the idle time. When the next switching edge arrives, the determination result is ready. The whole process does not preempt or delay the main control program.

[0085] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A bypass backflow detection method for a UPS, characterized in that, Includes the following steps: Using the inverter switching transient as a natural excitation, the high-frequency transient response of the inverter output and bypass input is synchronously captured within a microsecond time window under the triggering of the switching edge. Based on the dual-ended signal within the microsecond-level time window, the transient transfer entropy or conditional Granger causality strength from the inverter end to the bypass end is calculated as a causal coupling parameter for quantifying the driving strength of the switching event to the bypass response, and the causal coupling parameter is output as a topological coupling strength measurement value characterizing the tightness of the electrical connection between the inverter end and the bypass end. There is no direct electrical connection between the inverter and the bypass. The signals at both ends are only weakly causally correlated by parasitic stray coupling, which is defined as the first causal state. There is an electrical direct connection between the inverter and the bypass, forming a metal path. The transient signal of the inverter can generate a stable strong causal drive on the bypass, which is defined as the second causal state. Based on the causal coupling parameters, monitor whether a step transition occurs from the first causal state to the second causal state; If the step change is detected and the preset stability criterion is met, it is determined that bypass backfeeding has occurred.

2. The bypass backflow detection method for UPS according to claim 1, characterized in that, The baseline of the first causal state is established based on the null hypothesis model. Under the full operating condition of confirming no bypass backflow, the empirical zero distribution of the causal coupling parameters is statistically analyzed, and a rejection region is constructed based on a preset significance level as a statistical criterion to distinguish between weak causal pseudo-associations and strong causal drives.

3. The bypass backflow detection method for UPS according to claim 2, characterized in that, Before calculating the causal coupling parameters, the dual-end signals within the microsecond-level time window are subjected to adaptive power frequency notch filtering and steady-state harmonic removal, and are approximated as discrete sequences through symbolization or segmented aggregation to eliminate the interference of random noise on transient causal calculation.

4. The bypass backflow detection method for UPS according to claim 3, characterized in that, Based on the reverse causal strength of the bypass end to the inverter end based on synchronous calculation, when the forward causal strength undergoes a statistically significant step change while the reverse strength remains unchanged, and the ratio of the forward to reverse strength exceeds a preset asymmetry threshold, it is confirmed as a reverse-feed dedicated unidirectional causal topology of the inverter actively driving the bypass.

5. The bypass backflow detection method for a UPS according to claim 4, characterized in that, The determination process for the preset stability criterion is as follows: Over multiple consecutive switching cycles, the causal coupling parameter consistently rejects the null hypothesis, and the fluctuation of the parameter value between adjacent cycles is lower than a preset steady-state threshold.

6. The bypass backflow detection method for a UPS according to claim 5, characterized in that, Based on real-time collected inverter DC bus voltage, load current and junction temperature parameters, the causal parameters in the first causal state are compensated through a pre-built operating condition-parasitic coupling mapping model to remove the influence of gradual operating condition changes on parasitic coupling and retain only the causal step component caused by electrical direct connection.

7. The bypass backflow detection method for a UPS according to claim 6, characterized in that, When multiple inverters are connected in parallel, the independent switching transients of each inverter are used as different excitation sources. The causal strength of the bypass response of each excitation source is calculated separately, and the inverter unit that has reverse feed is located based on the excitation source where the strength changes abruptly.

8. The bypass backflow detection method for a UPS according to claim 7, characterized in that, After determining that a bypass backflow has occurred, the bypass switch is tripped based on the protection command, and the original waveforms, causal parameter sequences, and null hypothesis rejection domain boundaries before and after the abrupt change are frozen to generate an unalterable fault snapshot.

9. The bypass backflow detection method for a UPS according to claim 8, characterized in that, During the operation period when the system confirms no bypass backflow, the null hypothesis baseline model corresponding to the first causal state is smoothly updated based on the continuously collected parasitic coupling transient data to adapt to the slow drift of parasitic coupling caused by device aging and environmental changes.

10. The bypass backflow detection method for a UPS according to claim 9, characterized in that, The calculation of the causal coupling parameters, state differentiation, and abrupt change determination are all performed online during the normal power supply operation of the inverter. The selection of the microsecond-level time window is adaptively matched with the attenuation time of the switching transient oscillation.

Images