Multi-phase UPS battery parameter cooperative monitoring and fault prediction system

Abstract

The invention discloses a multi-phase UPS (Uninterrupted Power Supply) battery parameter cooperative monitoring and fault prediction system, which relates to the technical field of batteries and comprises a dynamic synchronous acquisition subsystem and a coupling correction subsystem. The method comprises the following steps: determining a transient pulling characteristic corresponding to each phase after disturbance occurs, obtaining a coupling degree between phases on corresponding parameters in the transient pulling characteristic through correlation coefficient analysis, obtaining a disturbance residual error by utilizing a CP decomposition mode, constructing a multi-stage abnormal measurement sequence based on the disturbance residual error, and obtaining a multi-stage abnormal measurement result. Based on normal coupling brought by UPS multiphase natural energy migration, a multi-stage abnormal measurement sequence is dynamically corrected, whether a backtracking instruction is triggered or not is judged, and if the backtracking instruction is triggered, the sampling period of the parameter of the corresponding phase is dynamically adjusted.

Description

Technical Field

[0001] The present invention relates to the technical field of batteries, and in particular to a multi-phase UPS battery parameter collaborative monitoring and fault prediction system. Background Art

[0002] In existing multi-phase UPS systems, a single-point voltage or current sampling method combined with a fixed threshold judgment is commonly used to detect the health status of the battery pack. Some systems introduce simple statistical models based on the remaining capacity (SOC) and the state of health (SOH) in order to achieve online estimation of the battery status when the UPS is running under load. However, such methods often only perform threshold judgments on a single electrical parameter, making it difficult to fully capture the transient pull characteristics of the UPS multi-phase inverter output on the battery under different load conversions. For example, under normal three-phase balanced load conditions of the UPS, the voltage and current at the battery end may fluctuate significantly simultaneously for a short period of time, but this is often due to the natural energy rebalancing of the UPS, not an abnormality caused by battery degradation. However, due to the lack of in-depth analysis of the multi-phase coupling behavior of the UPS, existing methods are prone to false alarms in this scenario, reducing the reliability of monitoring. Summary of the Invention

[0003] In view of the above-mentioned problems existing in the prior art, the present application provides a multi-phase UPS battery parameter collaborative monitoring and fault prediction system.

[0004] The present disclosure provides a multi-phase UPS battery parameter coordinated monitoring and fault prediction system, including:

[0005] The dynamic synchronous acquisition subsystem installs multiple sensor measurement points on the DC bus branch between the battery pack and the inverter to determine the transient pull characteristics corresponding to each phase after the disturbance occurs. Through correlation coefficient analysis, the coupling degree between each phase on the corresponding parameters within the transient pull characteristics is obtained, and the disturbance residual is obtained using the CP decomposition method.

[0006] The coupling correction subsystem constructs a multi-stage abnormal measurement sequence based on the disturbance residual. Based on the normal coupling brought about by the natural energy migration of UPS multi-phase, it dynamically corrects the multi-stage abnormal measurement sequence and determines whether to trigger the backtracking instruction. If the backtracking instruction is triggered, the sampling period of the parameters of the corresponding phase is dynamically adjusted.

[0007] Optionally, the dynamic synchronous acquisition subsystem includes an acquisition module, an offset module, and a decomposition and extraction module;

[0008] The acquisition module, when the multi-phase UPS system is under normal load conditions, connects a programmable electronic load in parallel to the multi-phase output end of the UPS system and installs multiple sensor measurement points on the DC bus branch between the battery pack and the inverter to record the transient characteristics corresponding to each phase after the disturbance occurs, including the battery output voltage, battery output current and complex impedance corresponding to each phase after the disturbance occurs.

[0009] Optionally, the offset module determines, based on the transient characteristics, the transient pull characteristics corresponding to each phase after the disturbance occurs, including the transient drop amplitude of the battery voltage corresponding to each phase, the current peak change, and the impedance trajectory, and performs tensorization processing on the transient pull characteristics corresponding to each phase to form a third-order tensor. Based on the third-order tensor, correlation coefficient analysis is performed to obtain the coupling degree between each phase on the corresponding parameters in the transient pull characteristics. The coupling degree is used to capture the cross-disturbance of the transient pull characteristics between different phases, and the coupling degree between each phase on the corresponding parameters in the transient pull characteristics is used as a supplementary attribute of the third-order tensor.

[0010] Optionally, a decomposition and extraction module uses CP decomposition to decompose the third-order tensor into a low-rank basis state and a perturbation residual. The low-rank basis state represents the cooperative behavior mode of the multi-phase UPS under normal load conversion, and the perturbation residual is used to explicitly characterize signs of abnormal single-phase or cross-phase degradation.

[0011] Optionally, the coupling correction subsystem includes an anomaly identification module, a correction module, a backtracking module, and a correction module;

[0012] The anomaly identification module expands the disturbance residual in the time dimension to obtain the disturbance residual vector. Based on the disturbance residual vector, the disturbance residual at different time points is analyzed, and the anomaly intensity at each time point is obtained by calculating the Frobenius norm of the disturbance residual vector. In order to capture the anomaly accumulation trend in different time periods, a sliding window is set, and the average anomaly intensity is calculated on each sliding window to form a multi-stage anomaly measurement sequence.

[0013] Optionally, in order to further reduce the risk of normal coupling caused by the natural energy migration of UPS multi-phase being misjudged as abnormal, the correction module calculates the coupling degree between the UPS phases when facing the same parameters in the same sliding window and takes the average value to form a phase linkage index; based on the numerical value of the phase linkage index, the average abnormal intensity on the corresponding sliding window is corrected in stages to obtain the corrected average abnormal intensity, and the multi-stage abnormality measurement sequence is updated.

[0014] Optionally, the backtracking module pre-sets an anomaly threshold. If the corrected average anomaly intensity of multiple consecutive sliding windows in the updated multi-stage anomaly measurement sequence exceeds the anomaly threshold, the backtracking instruction is triggered to trace back the fluctuation of the updated multi-stage anomaly measurement sequence to the decomposition and extraction module, and at each backtracking, the decomposition rank value of the original CP decomposition is adaptively increased by 1, so that the decomposed low-rank basis state is closer to the current actual operating state of the UPS system. Until the backtracking instruction is not triggered, the backtracking action is stopped to obtain the final low-rank basis state and disturbance residual.

[0015] Optionally, the correction module, based on the final disturbance residual, takes the average absolute residual of each parameter of each phase in the time dimension to form a sensitivity matrix. Each element in the sensitivity matrix refers to each parameter of the corresponding phase, and the average absolute residual is taken in the time dimension. Each element in the sensitivity matrix is ​​used to reflect the average abnormal concentration of the corresponding parameter of the corresponding phase. If the corresponding element in the sensitivity matrix exceeds the preset sensitivity threshold, the sampling period is dynamically adjusted, otherwise no dynamic adjustment is performed.

[0016] Optionally, the sampling period can be adjusted dynamically as follows: ,in, For the Phase The sampling period of the parameters, is the base sampling period, is the adjustment coefficient, For the Phase The parameters are taken as the mean absolute residual in the time dimension.

[0017] Beneficial effects of the present invention:

[0018] This invention achieves dynamic, synchronous acquisition of key battery parameters in a multiphase UPS by deploying multiple sensor points along the DC bus branch between the battery pack and the UPS inverter. By introducing controllable perturbations during UPS operation, the system accurately extracts the transient pull-through characteristics corresponding to each phase under multiphase load disturbance conditions. By employing correlation coefficient analysis within the transient pull-through characteristics to capture the parameter linkage between phases under the naturally coupled state of the UPS multiphase, and combining this with CP decomposition to obtain multidimensional perturbation residuals, the system enhances the dynamic characterization of the coupled behavior of the UPS multiphase operation. Furthermore, through a coupled correction subsystem, a multi-stage anomaly metric sequence is constructed based on the obtained perturbation residuals. This sequence is dynamically corrected using the coupling patterns of the UPS during normal multiphase energy migration, effectively filtering out false anomaly signals caused by normal UPS load migration and further reducing the probability of false alarms. Furthermore, by monitoring whether the corrected anomaly metric sequence consistently exceeds a set threshold and triggering a backtracking mechanism, the sampling period of the corresponding phases and parameters is automatically and dynamically adjusted, enabling system sampling resources to be focused on sensitive or potentially degraded phases and parameters, achieving adaptive optimization of the acquisition strategy and enhanced monitoring flexibility. Therefore, compared with existing single-point voltage threshold monitoring or single SOC estimation methods, the present invention can further improve the health monitoring accuracy and predictive diagnostic capabilities of multi-phase batteries in UPS multi-phase natural energy migration scenarios, reduce misjudgments and resource waste caused by unidentified multi-phase coupling, and improve the operational safety of UPS power supply systems and the scientific nature of maintenance decisions.

[0019] By sequentially placing an anomaly identification module, a correction module, a backtracking module, and a correction module within the coupling correction subsystem, the present invention enables phased, multi-angle dynamic identification of multi-dimensional anomaly signs exhibited by UPS multiphase batteries after transient disturbances. First, the anomaly identification module expands the disturbance residual along the time dimension, extracts the disturbance residual vector, and uses the Frobenius norm calculation method to obtain the anomaly intensity at each time point, thereby forming a multi-stage anomaly metric sequence within a sliding window, accurately reflecting the anomaly accumulation of UPS multiphase batteries in different time periods. Furthermore, the correction module statistically calculates the coupling degree of each UPS phase under the same parameters within the same sliding window and calculates its mean to form a phase linkage index. Based on this index, the anomaly metric sequence is then phase-corrected, achieving dynamic adaptive filtering of normal coupling behavior caused by natural energy migration in the UPS multiphase, reducing the probability of normal cooperative behavior being misjudged as degradation anomalies. This not only enhances the targeted anomaly detection, but also effectively avoids multiple false alarms caused by normal UPS load migration coupling, improving the stability of the UPS monitoring system and the reliability of battery health status identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the present application or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application.

[0021] Figure 1 This is a module diagram of a multi-phase UPS battery parameter collaborative monitoring and fault prediction system according to the present invention. DETAILED DESCRIPTION

[0022] To make the objectives, technical solutions, and advantages of this application more clear, the technical solutions of this application will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments of this application, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of this application.

[0023] like Figure 1 As shown, the present invention proposes a multi-phase UPS battery parameter collaborative monitoring and fault prediction system, including:

[0024] The dynamic synchronous acquisition subsystem installs multiple sensor measurement points on the DC bus branch between the battery pack and the inverter to determine the transient pull characteristics corresponding to each phase after the disturbance occurs. Through correlation coefficient analysis, the coupling degree between each phase on the corresponding parameters within the transient pull characteristics is obtained, and the disturbance residual is obtained using the CP decomposition method.

[0025] The coupling correction subsystem constructs a multi-stage abnormal measurement sequence based on the disturbance residual. Based on the normal coupling brought about by the natural energy migration of UPS multi-phase, it dynamically corrects the multi-stage abnormal measurement sequence and determines whether to trigger the backtracking instruction. If the backtracking instruction is triggered, the sampling period of the parameters of the corresponding phase is dynamically adjusted.

[0026] UPS refers to the entire uninterruptible power supply system, which usually includes:

[0027] Rectifier (rectifies AC input to DC);

[0028] DC bus (connecting the battery and the rectifier / inverter);

[0029] Battery packs (energy storage);

[0030] Inverter (converts DC into AC);

[0031] Output filter, bypass switch, monitoring module, etc.;

[0032] In an embodiment of the present invention, a dynamic synchronous acquisition subsystem can capture the transient pull behavior of a multi-phase UPS during a disturbance. Specifically, the dynamic synchronous acquisition subsystem deploys multiple voltage, current, and impedance sensor measurement points on the DC bus branch between the battery pack and the UPS inverter to record in real time the transient response of the battery output of each phase after a disturbance event, such as a step load switching or inverter inter-phase power balance adjustment. For example, when the UPS is under normal load conditions and a rapid load disturbance is achieved by connecting a programmable electronic load in parallel at the output end, it can be observed that the battery terminal voltage of phase A drops by 0.3V in a short period of time, while phases B and C experience transient drops of 0.25V and 0.27V almost simultaneously. At this time, based on the synchronous sampling data, the system can intuitively identify the transient pull characteristics during the normal energy balance process through phase comparison, avoiding misjudging the voltage drop of phase A alone as single-phase degradation.

[0033] In this process, the transient pull characteristics are used to characterize the rapid offset behavior of the voltage, current, and impedance of each phase battery output in a micro-time segment under short-cycle load disturbances of the UPS three-phase or multi-phase.

[0034] Through coupling calculation and CP decomposition, the normal coupling mode in the multi-phase UPS system is effectively extracted and the abnormal disturbance residual is explicitly distinguished. Specifically, in the dynamic synchronous acquisition subsystem, the correlation coefficient analysis of the transient pull characteristics is used to obtain the coupling between different phases under the same parameters, thereby quantifying the normal multi-phase energy migration phenomenon. For example, after an inverter automatic balancing event, the Pearson correlation coefficient of the battery phase offset voltage waveforms of phase A and phase B was calculated to obtain a coupling degree of 0.93, indicating that the process was mainly the natural load power migration within the UPS. These inter-phase coupling degrees are then used as tensor supplementary attributes, and the CP decomposition method is used to decompose the third-order tensor into a low-rank ground state and disturbance residuals, ensuring that the normal UPS multi-phase cooperative behavior is completely classified into the ground state mode, and only the part beyond the normal coupling offset range is retained in the disturbance residual.

[0035] Here, CP decomposition is used to automatically extract the main cooperative behaviors (low-rank ground states) of the UPS system from the multidimensional time series tensor, while the perturbation residuals explicitly identify uncoupled behavioral features that may be related to degradation, loss, and localized failures.

[0036] The coupled correction subsystem implements dynamic anomaly correction and sampling adaptive adjustment based on the natural multiphase coupling of the UPS, further reducing false positives. Specifically, in the coupled correction subsystem, a multi-stage anomaly measurement sequence is first constructed based on the perturbation residual. The phase linkage index is then formed using the mean coupling degree of different UPS phases under the same parameters to dynamically correct outliers. For example, if the phase linkage index remains high (e.g., 0.89) when the anomaly intensity is detected to be maintained at a high level in a continuous sliding window, it indicates that the UPS multiphase coupling is still robust. The system suppresses the anomaly through correction to avoid false positives. If the multi-stage anomaly still exceeds the threshold after correction, a backtracking instruction is triggered. By dynamically increasing the CP decomposition rank (e.g., from 3 to 4), the tensor ground state is further adapted to the new normal UPS coupling mode, thereby reducing the perturbation residual and lowering the false positive rate.

[0037] Finally, based on the ultimately stabilized disturbance residuals, the system calculates the average absolute residuals of each parameter for each phase in the time dimension to form a sensitivity matrix, which is used to guide real-time sampling period adjustments. For example, if the matrix shows that the average residual of the phase offset voltage of phase A is 0.15, while the phase offset voltage of phase C is only 0.03, the system automatically adjusts the sampling period of the phase offset voltage of phase A to 60% of the reference period, that is, sampling at a higher frequency to improve monitoring resolution.

[0038] Coupling correction is used to ensure that the normal multi-phase load self-balancing of the UPS will not be mistakenly detected as abnormal;

[0039] The backtracking instruction represents the automatic adjustment process of the tensor rank value based on abnormal feedback, forming a self-optimizing closed loop.

[0040] Therefore, the present invention gradually realizes a complete closed-loop process from abnormal feature detection to tensor model update and then to sampling resource focused optimization by collecting multi-dimensional and multi-phase data between battery output and UPS inverter, combining correlation coefficient analysis, CP decomposition, anomaly correction and dynamic sampling adjustment. This effectively overcomes the problems of false alarms, omissions and lack of adaptability to natural coupling behavior in existing UPS multi-phase battery monitoring, and greatly improves the health prediction accuracy and operational safety of multi-phase UPS batteries.

[0041] Furthermore, the dynamic synchronous acquisition subsystem includes an acquisition module, an offset module, and a decomposition and extraction module;

[0042] The acquisition module, when the multi-phase UPS system is under normal load conditions, connects a programmable electronic load in parallel to the multi-phase output end of the UPS system (close to the inverter output side) and installs multiple sensor measurement points on the DC bus branch between the battery pack and the inverter to record the transient characteristics corresponding to each phase after the disturbance occurs, including the battery output voltage, battery output current and complex impedance corresponding to each phase after the disturbance occurs.

[0043] It should be noted that in many UPS system designs, all phase inverters are connected to the same battery bus. However, even when the same battery bus is used, the transient absorption current of inverters in different phases will be slightly different. This will cause slight voltage and chemical potential differences in the battery output when the load of three or more phases changes rapidly. When accumulated over a long period of time, these slight differences will cause some batteries to age faster and the load fluctuations of a single phase to be amplified, affecting the life of the entire machine. Therefore, phase-by-phase data collection is required in UPS monitoring.

[0044] The multi-phase output end of the UPS system refers to the AC output end of the entire UPS system after inversion, usually a multi-phase AC output;

[0045] Multiple sensor measurement points are used to capture subtle changes in voltage, current, and impedance over time in real time;

[0046] The purpose of connecting a programmable electronic load in parallel to the multi-phase output end of the UPS system (near the inverter output side) is to increase or decrease the load power at the normal output end of the UPS for a short time, thereby stimulating the dynamic response of the UPS and battery.

[0047] The offset module determines the transient pull characteristics corresponding to each phase after the disturbance occurs based on the transient characteristics, including the transient drop amplitude of the battery voltage, the current peak change, and the impedance trajectory corresponding to each phase;

[0048] The transient drop of battery voltage refers to the instantaneous drop of battery output voltage due to the rapid increase of current after a sudden load change (such as a step increase) occurs at the UPS output end. What is recorded is the voltage drop from steady state to transient state.

[0049] The peak change of battery current refers to the instantaneous jump of battery output current from the original steady-state current when the same load step event occurs, and the difference between the instantaneous maximum value and the original value is recorded;

[0050] The impedance trajectory refers to the complex impedance changes of the battery at multiple frequencies;

[0051] In a multi-phase UPS (such as a three-phase UPS), each phase has a corresponding inverter bridge arm and filter unit. When the external load undergoes a step change (for example, from 0% to 50%) or a harmonic disturbance, the current drawn from each phase of the inverter to the battery (the transient charge and discharge peak value) and the transient voltage drop (the magnitude of the voltage drop) will vary. This is known as the transient pull characteristic.

[0052] The transient pull characteristics corresponding to each phase are tensorized to form a third-order tensor. Based on the third-order tensor, the correlation coefficient is analyzed to obtain the coupling degree between each phase on the corresponding parameters in the transient pull characteristics. The coupling degree is used to capture the cross-disturbance of the transient pull characteristics between different phases, and the coupling degree between each phase on the corresponding parameters in the transient pull characteristics is used as a supplementary attribute of the third-order tensor. Writing the coupling degree into the supplementary attribute of the tensor here does not mean directly adding the coupling degree to the tensor value. Instead, in the CP decomposition, these coupling degrees are used as one of the optimization goals, for example, keeping the highly coupled phase decomposed into a similar ground state.

[0053] In multi-phase UPS monitoring, a third-order tensor refers to the sampling data having three dimensions: phase, parameter, and time;

[0054] The specific method for obtaining the coupling degree between phases on the corresponding parameters in the transient pull characteristic is as follows: ,in, For the transient pull characteristics Neidi The coupling degree between phase A and phase B in terms of parameters, is the correlation coefficient, which is used to capture the linkage between two phase time series. and The transient pull characteristics are The inner phase A and phase B correspond to the parameter values, All are phase numbers;

[0055] In an embodiment of the present invention, when the UPS is in normal operation, the acquisition module connects a programmable electronic load in parallel to the multi-phase output end of the UPS system (i.e., the multi-phase AC output after the inverter output), for example, setting it to perform a 10% load step change in milliseconds. At the same time, multiple sensor measurement points are arranged on the DC bus branch between the battery pack and the inverter to collect the tiny fluctuations of the battery output voltage, current and complex impedance over time in real time, thereby forming a high-time-resolution data matrix of multi-phase and multi-electrical parameters.

[0056] Here, the multi-phase output terminals of the UPS are used to artificially introduce small disturbances to reveal the battery response, while the sensor measurement points on the DC bus branches are used to capture the projection effects of these disturbances on the voltage, current, and impedance of the battery terminals.

[0057] After acquiring the above disturbance data, the migration module extracts the transient pull characteristics corresponding to each phase within a short time window (for example, within 5ms) after the disturbance occurs based on transient characteristic analysis.

[0058] The transient pull characteristic refers to the dynamic response deviation of batteries in different phases after the UPS load disturbance.

[0059] The offset module tensors these multi-phase, multi-parameter transient pull characteristics into a three-order tensor, where the three dimensions correspond to phase, parameter, and time series, respectively. This is used to comprehensively characterize the coupled behavior of multi-phase UPS batteries under disturbances. Based on this, the offset module performs correlation coefficient analysis on each parameter slice of the transient tensor. For example, if the impedance correlation coefficient between phases A and B within the same transient pull window reaches 0.92, it indicates strong coupling between the two phases. If the transient voltage correlation coefficient between phases A and C is only 0.48, it indicates weak linkage between the two phases under this parameter. This coupling is used to capture the cross-response between different phases when facing the same disturbance, and can identify the linkage pattern caused by normal three-phase energy rebalancing in the UPS.

[0060] Preferably, the present invention uses these coupling indicators as supplementary attributes of the third-order tensor, that is, the transient coupling relationship of different phases under various parameters is stored in the tensor data structure, so that the subsequent CP decomposition can more effectively decompose the ground state and residual while retaining the UPS coupling behavior characteristics. Therefore, by acquiring high-precision and real-time disturbance response data in the acquisition module, extracting multi-parameter transient pull characteristics and coupling analysis in the offset module, and explicitly extracting the UPS three-phase coordinated behavior ground state and abnormal disturbance residual using CP decomposition in the decomposition and extraction module, the present invention can not only significantly improve the accuracy of characterizing the battery transient response caused by UPS multi-phase load conversion, but also reduce the misjudgment rate caused by UPS natural energy migration in subsequent anomaly measurement and retrospective correction. For example, the traditional alarm threshold method based on single-point voltage drop often produces false alarms when the UPS normal A and B phase loads migrate. However, the present invention, based on the joint analysis of coupling and tensor processing, can correctly identify such natural coupling behavior and reduce false alarms.

[0061] Furthermore, the decomposition and extraction module uses CP decomposition to decompose the third-order tensor into a low-rank ground state and a perturbation residual. The low-rank ground state represents the cooperative behavior mode of the multi-phase UPS under normal load conversion, and the perturbation residual is used to explicitly characterize the signs of abnormal single-phase or cross-phase degradation. This decomposition structure not only forms the normal coupling mode fingerprint of the current multi-phase UPS, but also provides a preliminary reference for subsequent dynamic prediction.

[0062] CP decomposition is to express the tensor as: ,in, is a third-order tensor, E is the decomposition rank value of CP decomposition, r is the number of the decomposition rank value, is the rth group of mode vectors on the first dimension of the tensor (e.g., the phase of UPS), is the mode vector of the rth group on the second dimension of the tensor (e.g. UPS parameters: voltage, current, impedance), is the pattern vector of the rth group in the third dimension of the tensor (e.g. time), is the outer product of the vectors, that is: Generates a rank 1 tensor (3D matrix) whose elements are , in other words, this is the outer product of three vectors, forming a "block".

[0063] This formula actually means that the third-order tensor is approximately represented as the superposition of E rank-1 tensors (that is, three vector outer products). Each rank-1 tensor consists of a set of characteristic pattern vectors in the phase dimension, parameter dimension, and time dimension. By accumulating the outer products of all these patterns, the original tensor data can be well reconstructed.

[0064] The CP decomposition method is used to decompose the third-order tensor into a low-rank ground state and a perturbation residual. The specific steps are:

[0065] (1) Initialization: First set the decomposition rank value, for example, E = 3 (indicates that you want to use 3 sets of main modes to describe the tensor), and randomly initialize the three mode matrices corresponding to the three directions of phase, parameter, and time;

[0066] (2) Alternating Least Squares (ALS) loop update:

[0067] The mode matrix of the UPS acquisition parameter direction (voltage, current, impedance) and the mode matrix of the time direction are treated as known quantities, and then the least squares method is used to find the optimal mode matrix of the UPS phase direction so that the overall tensor reconstruction error is minimized. Then, the next round of the UPS phase direction mode matrix and the mode matrix of the UPS acquisition parameter direction (voltage, current, impedance) are treated as known quantities, and the least squares method is used to find the optimal mode matrix of the time direction. The iteration is repeated until convergence;

[0068] (3) After decomposition, the three groups of vectors are: phase main mode, parameter main mode and time main mode. The tensor reconstructed by them together is the low-rank ground state;

[0069] (4) Calculate the perturbation residual: Obtain the perturbation residual by subtracting the third-order tensor from the low-rank basis state.

[0070] Tensor decomposition can extract the main cooperative modes (low-rank ground states) from a large amount of coupled data, and reduce the complex data to: normal coupling behavior + abnormal perturbation residuals;

[0071] In this embodiment of the present invention, the decomposition and extraction module further decomposes the third-order tensor into a low-rank ground state and a perturbation residual through CP decomposition. The low-rank ground state characterizes the coordinated behavior of the UPS multiphase system under normal load migration or balanced load distribution. For example, the natural energy rebalancing characteristics of the UPS's internal three-phase inverter are fully absorbed and represented by the low-rank ground state. The perturbation residual explicitly reveals abnormal single-phase or cross-phase degradation signs that exceed the normal UPS energy migration range. For example, if the internal resistance of the C-phase battery suddenly changes, causing the C-phase current to increase abnormally in a short period of time, this anomaly will be primarily reflected in the residual tensor. Through this approach, the present invention effectively decouples the normal coordinated energy migration pattern of the UPS multiphase system from potential abnormal degradation signs, thereby improving the resolution of UPS multiphase battery health status monitoring. Compared with traditional judgment methods that rely solely on single-channel voltage and current fluctuations, this method can effectively avoid false alarms during normal UPS three-phase migration and accurately identify and predict UPS single-phase degradation in the early stages, helping to extend battery life, reduce maintenance costs, and enhance the overall safety of the power supply system.

[0072] The third-order tensor is used to uniformly encode the multi-phase and multi-parameter responses of UPS over time in a multi-dimensional space, capturing the complex spatiotemporal coupling patterns.

[0073] CP decomposition is used to separate the tensor into a normal ground state and an abnormal residual, that is, to decompose the complex behavior into two parts: normal cooperation and over-limit anomaly.

[0074] The low-rank ground state is used to express the natural coupling behavior of the UPS's normal load transfer to avoid misjudgment. The perturbed residual is used to highlight potential degradation points that truly exceed the normal UPS's self-balancing capability.

[0075] Furthermore, the coupling correction subsystem includes an anomaly identification module, a correction module, a backtracking module, and a correction module;

[0076] The anomaly identification module expands the disturbance residual in the time dimension to obtain the disturbance residual vector. Based on the disturbance residual vector, the disturbance residual at different time points is analyzed, and the anomaly intensity at each time point is obtained by calculating the Frobenius norm of the disturbance residual vector. In order to capture the anomaly accumulation trend in different time periods, a sliding window is set, and the average anomaly intensity is calculated on each sliding window to form a multi-stage anomaly measurement sequence.

[0077] The multi-stage anomaly metric sequence is used to describe the temporal trend of anomaly evolution;

[0078] To further reduce the risk of normal coupling caused by the natural energy migration of UPS multi-phases being misjudged as abnormal, the correction module calculates the coupling between the UPS phases in the same sliding window when facing the same parameters and takes the average to form a phase linkage index. For example, if it is a three-phase UPS (phase A, phase B, phase C), the coupling between phase A and phase B, the coupling between phase A and phase C, and the coupling between phase B and phase C when facing the same parameters in the same sliding window are obtained respectively, and the average is taken to obtain the phase linkage index. Based on the value of the phase linkage index, the average abnormal intensity on the corresponding sliding window is periodically corrected to obtain the corrected average abnormal intensity, and the multi-stage abnormality measurement sequence is updated to further avoid normal coupling behavior being misjudged as degradation.

[0079] The phase linkage index is used to dynamically quantify the tightness of multiphase coupling, thereby avoiding misjudging normal coupling behavior as abnormal.

[0080] The corrected average anomaly intensity is obtained by the following formula: ,in, For the The average anomaly intensity after correction over the sliding window, For the The average anomaly intensity over the sliding window, To adjust the coefficient of the amplification strength, it is used to amplify or reduce the influence of the coupling correction. It takes a positive value (such as 0.5~2). It is a phase linkage index. If the phase linkage index tends to 1, it means that the multiphase coupling is very tight, indicating that the residual is likely to come from normal migration and will be corrected and weakened. Otherwise, the abnormal measurement will be amplified.

[0081] In this embodiment of the present invention, a coupled correction subsystem is further employed to perform in-depth, multi-stage dynamic anomaly correction analysis on the disturbance residuals extracted during UPS multi-phase battery monitoring. This coupled correction subsystem includes an anomaly identification module, a correction module, a backtracking module, and a correction module. These modules collaborate to form a closed-loop optimization logic chain.

[0082] Specifically, the anomaly identification module first expands the perturbation residual tensor along the time dimension to obtain a perturbation residual vector that evolves over time. For example, when the UPS system collects multiple time-series slices of data within a 30-second monitoring cycle, the module converts the tensor residual slice at each moment into a vector and calculates its Frobenius norm, which measures the magnitude of the anomaly deviation for all phases and parameters combined at that point in time. The Frobenius norm serves as a unified scalar measure of multidimensional residuals, concisely reflecting the overall anomaly strength of the UPS multiphase battery at that point in time.

[0083] To capture anomaly accumulation trends over different time periods, the anomaly identification module further uses a sliding window to perform local statistics on these anomaly intensity values. For example, a sliding window of length 5 can be selected, and the average anomaly intensity can be calculated window by window, thus forming a multi-stage anomaly measurement sequence to help identify potential medium- and long-term deviation patterns in the UPS. However, to avoid misidentifying short-term coordinated fluctuations caused by normal energy migration from the UPS's multi-phase batteries as degradation signals, the correction module performs pairwise correlation coefficient analysis on the fluctuation sequences of the same parameters (such as voltage or impedance) of the UPS's multi-phases (such as phases A, B, and C) within the same sliding window. For example, the coupling between phases A and B, phases A and C, and phases B and C is obtained, and the average is calculated to form the so-called phase linkage index. This phase linkage index quantifies the degree of coordination between the UPS's multi-phases within this window segment. The closer the value is to 1, the more consistent the fluctuations of the three UPS phases on this parameter, and the more likely it is normal energy migration behavior.

[0084] Based on this, the correction module dynamically modifies the average anomaly intensity within the sliding window according to the phase linkage index. For example, when the linkage index approaches 1, the correction formula reduces the magnitude of the outlier value in that window. Conversely, if the linkage index is low, the outlier value is retained or even amplified, resulting in a more reasonable corrected multi-stage anomaly metric sequence. This mechanism can effectively prevent false anomalies caused by normal UPS multiphase energy balance adjustments being misjudged as degradation. Furthermore, the backtracking module monitors the updated multi-stage anomaly metric sequence. When, for example, the corrected anomaly intensity of three consecutive sliding windows exceeds a preset threshold, the backtracking mechanism is triggered, automatically returning to the decomposition and extraction module. By adaptively increasing the decomposition rank of the original CP decomposition by 1, the tensor ground state's representation capability is enhanced, allowing the ground state to more fully absorb the current natural multiphase coupling characteristics of the UPS, further reducing the residual amplitude and significantly lowering the risk of false alarms.

[0085] Therefore, through the above-mentioned anomaly identification, correction based on phase linkage indicators, and threshold-driven backtracking decomposition optimization process, the present invention can dynamically and accurately distinguish between the natural coupling behavior of UPS multi-phase and potential degradation anomalies. It not only avoids the frequent false alarms caused by traditional monitoring methods that do not consider three-phase coordination, but also can promptly capture the abnormal evolution trends that really need to be focused on, greatly improving the predictive maintenance value of UPS batteries and the power supply security assurance capabilities.

[0086] The perturbation residual vector is used to quantify the overall abnormal excursion of UPS multi-phase and multi-parameters at each time point.

[0087] The Frobenius norm compresses the multidimensional disturbance vector into a single scalar, which is convenient for analysis on time series.

[0088] The sliding window is used to extract the multi-stage anomaly accumulation trend and form an anomaly measurement sequence.

[0089] The phase linkage index measures multi-phase synchronization by averaging the coupling degrees of each UPS phase facing the same parameter.

[0090] The correction module dynamically adjusts the abnormal sequence based on the phase linkage indicator to prevent false alarms.

[0091] The backtracking module automatically increases the tensor decomposition rank value when the outlier value is continuously high, so as to dynamically optimize the UPS coupling ground state characterization.

[0092] If the long-term sliding window outliers are increasing, it means that the current tensor decomposition base state may no longer be suitable for the normal operation mode of the latest UPS (for example, the UPS has changed its working strategy or the load structure). Therefore, the fluctuations of the multi-stage abnormal sequence need to be traced back to the tensor decomposition process;

[0093] Furthermore, the backtracking module pre-sets an anomaly threshold. If the corrected average anomaly intensity of multiple consecutive sliding windows in the updated multi-stage anomaly measurement sequence exceeds the anomaly threshold, the backtracking instruction is triggered to trace the fluctuation of the updated multi-stage anomaly measurement sequence back to the decomposition and extraction module, and at each backtracking, the decomposition rank value of the original CP decomposition is adaptively increased by 1, so that the decomposed low-rank basis state is closer to the current actual operating state of the UPS system, the residual amplitude is reduced and the false alarm rate is reduced. When the backtracking instruction is not triggered, the backtracking action is stopped to obtain the final low-rank basis state and disturbance residual.

[0094] By adaptively increasing the decomposition rank value, the decomposition ground state part is made closer to the actual normal coupling mode of the UPS, gradually reducing the potential sources of misjudgment. This method actively optimizes the coupling model through abnormal feedback and realizes dynamic self-learning.

[0095] In this multi-phase UPS battery monitoring system, each phase (e.g., A, B, and C) corresponds to a different load transient, resulting in a different transient response from the battery providing energy to each phase. Furthermore, coupled tensor decomposition may detect varying degrees of abnormal disturbances in each phase. Therefore, the system determines which phases are more worthy of monitoring (e.g., where the tensor residual is greatest), and assigns a higher sampling period to those phases. This ensures higher-resolution monitoring of high-risk phases while maintaining global monitoring, continuously converging the false positive rate and missed detection rate for the entire system.

[0096] The correction module takes the average absolute residual of each parameter of each phase in the time dimension based on the final disturbance residual to form a sensitivity matrix. Each element in the sensitivity matrix refers to each parameter of the corresponding phase. The average absolute residual is taken in the time dimension. Each element in the sensitivity matrix is ​​used to reflect the average abnormal concentration of the corresponding parameter of the corresponding phase. If the corresponding element in the sensitivity matrix exceeds the preset sensitivity threshold, the sampling period is dynamically adjusted. Otherwise, no dynamic adjustment is performed.

[0097] For each parameter of each phase, the mean absolute residual is taken in the time dimension. The specific method of obtaining it is: ,in, For the Phase Parameters, take the mean absolute residual in the time dimension, N is the monitoring period, Number the time points, For the At the time point, Phase The final perturbed residual of the parameters (i.e., the abnormal residual value).

[0098] The sampling period can be adjusted dynamically as follows: ,in, For the Phase The sampling period of the parameters, is the base sampling period, is the adjustment coefficient, which is used to control the influence weight of parameter sensitivity on the sampling period (the value is generally around 0.5~2). For the Phase The parameters are taken as the mean absolute residual in the time dimension.

[0099] In an embodiment of the present invention, the present invention further provides a backtracking module and a correction module in the coupled correction subsystem, so that the monitoring system of UPS multi-phase battery parameters has the ability of dynamic self-optimization and adaptive focusing of sampling resources, which significantly improves the predictability and robustness of collaborative health monitoring of multi-phase UPS batteries.

[0100] Specifically, the backtracking module first pre-sets an anomaly threshold to determine the maximum acceptable anomaly risk level for the UPS operating state. It then gradually checks the updated multi-stage anomaly metric sequence (i.e., the corrected average anomaly intensity sequence over the sliding window). If the average anomaly intensity exceeds the set anomaly threshold in multiple consecutive sliding windows (e.g., three consecutive sliding windows), a backtracking instruction is triggered.

[0101] Multiple consecutive sliding windows are used to avoid frequent backtracking triggered by a single short-term fluctuation, and can more reliably identify persistent abnormal deviations during load migration of multiple UPS phases. For example, after the UPS switches to energy-saving mode at night, the abnormal metric may exceed the threshold in three consecutive windows, triggering a backtracking action.

[0102] After the backtracking is triggered, the backtracking module feeds this multi-stage abnormal fluctuation information back to the decomposition and extraction module, adaptively increasing the decomposition rank by 1 based on the original tensor CP decomposition. Increasing the decomposition rank introduces more primary coupling modes, bringing the decomposed low-rank ground state closer to the UPS's new multiphase energy migration pattern, thereby reducing the residual amplitude and lowering the false alarm rate caused by normal operating state transitions. This process continues until the subsequent multi-stage anomaly metric sequence does not exceed the threshold again in multiple consecutive sliding windows, indicating that the decomposition rank is sufficient. The backtracking action automatically stops, and the low-rank ground state and abnormal residual that best match the UPS's current actual operation are obtained.

[0103] Furthermore, based on the final perturbation residual tensor, the correction module takes the average absolute residuals of each monitored parameter (such as battery output voltage, current, and impedance) in the time dimension for each phase to generate a sensitivity matrix. For example, for the output voltage parameter of phase A, the absolute value of its residual sequence is taken point by point over the monitoring period and then averaged to obtain a value that reflects the concentration of anomalies for that parameter in that phase. Each element of the sensitivity matrix indicates the average anomaly intensity for the corresponding phase and parameter. If this value exceeds a preset sensitivity threshold, a dynamic sampling period adjustment mechanism is triggered, automatically shortening the sampling period for that parameter in that phase. For example, if a significant increase in the sensitivity of phase B's impedance is detected, the sampling period is dynamically adjusted from 5ms to 2ms to enable more intensive data collection and timely capture of the evolution of potential anomalies. If the threshold is not exceeded, the baseline sampling period is maintained without dynamic adjustment.

[0104] In the above process, the abnormal threshold is used to constrain the multi-stage abnormal risk trigger points;

[0105] The decomposition rank value is used to adjust the degree of fitting of the tensor low-rank ground state to the UPS multiphase cooperative behavior;

[0106] The sensitivity matrix is ​​used to quantitatively reflect the abnormal concentration of each phase on each parameter;

[0107] Through the progressive linkage of the backtracking module and the correction module, the present invention can promptly absorb slight natural migrations in the normal operating mode of the UPS multiphase system through adaptive tensor decomposition rank adjustment. When true anomalies are concentrated, it can also achieve higher-precision early anomaly capture through dynamic focusing of sampling resources, thereby significantly improving the health prediction accuracy of the UPS multiphase battery and reducing misjudgments and waste of sampling resources. The backtracking module determines whether to trigger tensor rank adaptation through a multi-stage abnormal sliding sequence, reducing false positives under natural migration. The correction module calculates the sensitivity matrix through the final residual, dynamically adjusts the sampling period, and achieves self-focusing of monitoring resources.

[0108] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A multi-phase UPS battery parameter coordinated monitoring and fault prediction system, characterized by: include, The dynamic synchronous acquisition subsystem installs multiple sensor measurement points on the DC bus branch between the battery pack and the inverter to determine the transient pull characteristics corresponding to each phase after the disturbance occurs. Through correlation coefficient analysis, the coupling degree between each phase on the corresponding parameters within the transient pull characteristics is obtained, and the disturbance residual is obtained using the CP decomposition method. The coupling correction subsystem constructs a multi-stage abnormal measurement sequence based on the disturbance residual. Based on the normal coupling brought about by the natural energy migration of UPS multi-phase, it dynamically corrects the multi-stage abnormal measurement sequence and determines whether to trigger the backtracking instruction. If the backtracking instruction is triggered, the sampling period of the parameters of the corresponding phase is dynamically adjusted.

2. The multi-phase UPS battery parameter coordinated monitoring and fault prediction system according to claim 1, characterized in that: The dynamic synchronous acquisition subsystem includes an acquisition module, an offset module, and a decomposition and extraction module; The acquisition module, when the multi-phase UPS system is under normal load conditions, connects a programmable electronic load in parallel to the multi-phase output end of the UPS system and installs multiple sensor measurement points on the DC bus branch between the battery pack and the inverter to record the transient characteristics corresponding to each phase after the disturbance occurs, including the battery output voltage, battery output current and complex impedance corresponding to each phase after the disturbance occurs.

3. The multi-phase UPS battery parameter coordinated monitoring and fault prediction system according to claim 2, characterized in that: The offset module determines the transient pull characteristics corresponding to each phase after the disturbance occurs based on the transient characteristics, including the transient drop amplitude of the battery voltage, the change in current peak value, and the impedance trajectory corresponding to each phase. It also performs tensorization on the transient pull characteristics corresponding to each phase to form a third-order tensor. Based on the third-order tensor, correlation coefficient analysis is performed to obtain the coupling degree between each phase on the corresponding parameters within the transient pull characteristics. The coupling degree is used to capture the cross-disturbance of the transient pull characteristics between different phases. The coupling degree between each phase on the corresponding parameters within the transient pull characteristics is used as a supplementary attribute of the third-order tensor.

4. The multi-phase UPS battery parameter coordinated monitoring and fault prediction system according to claim 3, characterized in that: The decomposition and extraction module uses CP decomposition to decompose the third-order tensor into a low-rank basis state and a perturbation residual. The low-rank basis state represents the cooperative behavior mode of the multi-phase UPS under normal load conversion, and the perturbation residual is used to explicitly characterize signs of abnormal single-phase or cross-phase degradation.

5. The multi-phase UPS battery parameter coordinated monitoring and fault prediction system according to claim 4, characterized in that: The coupling correction subsystem includes an anomaly identification module, a correction module, a backtracking module and a correction module; The anomaly identification module expands the disturbance residual in the time dimension to obtain the disturbance residual vector. Based on the disturbance residual vector, the disturbance residual at different time points is analyzed, and the anomaly intensity at each time point is obtained by calculating the Frobenius norm of the disturbance residual vector. In order to capture the anomaly accumulation trend in different time periods, a sliding window is set, and the average anomaly intensity is calculated on each sliding window to form a multi-stage anomaly measurement sequence.

6. The multi-phase UPS battery parameter coordinated monitoring and fault prediction system according to claim 5, characterized in that: To further reduce the risk of normal coupling caused by the natural energy migration of UPS multiphases being misjudged as abnormal, the correction module calculates the coupling degree between the UPS phases when facing the same parameters within the same sliding window and takes the average to form a phase linkage index. Based on the value of the phase linkage index, the average anomaly intensity on the corresponding sliding window is periodically corrected to obtain the corrected average anomaly intensity and update the multi-stage anomaly measurement sequence.

7. The multi-phase UPS battery parameter coordinated monitoring and fault prediction system according to claim 6, characterized in that: The backtracking module pre-sets an anomaly threshold. If the corrected average anomaly intensity of multiple consecutive sliding windows in the updated multi-stage anomaly measurement sequence exceeds the anomaly threshold, the backtracking instruction is triggered to trace the fluctuation of the updated multi-stage anomaly measurement sequence back to the decomposition and extraction module. During each backtracking, the decomposition rank value of the original CP decomposition is adaptively increased by 1, so that the decomposed low-rank basis state is closer to the current actual operating state of the UPS system. When the backtracking instruction is not triggered, the backtracking action is stopped to obtain the final low-rank basis state and disturbance residual.

8. The multi-phase UPS battery parameter coordinated monitoring and fault prediction system according to claim 7, characterized in that: The correction module, based on the final disturbance residual, takes the average absolute residual of each parameter of each phase in the time dimension to form a sensitivity matrix. Each element in the sensitivity matrix refers to each parameter of the corresponding phase. The average absolute residual is taken in the time dimension. Each element in the sensitivity matrix is ​​used to reflect the average abnormal concentration of the corresponding parameter of the corresponding phase. If the corresponding element in the sensitivity matrix exceeds the preset sensitivity threshold, the sampling period is dynamically adjusted, otherwise no dynamic adjustment is performed.

9. The multi-phase UPS battery parameter coordinated monitoring and fault prediction system according to claim 8, characterized in that: The sampling period can be adjusted dynamically as follows: ,in, For the Phase The sampling period of the parameters, is the base sampling period, is the adjustment coefficient, For the Phase The parameters are taken as the mean absolute residual in the time dimension.

Images