Power supply switching control optimization method and system

Abstract

The invention relates to a power supply switching control optimization method and system, and the method comprises the steps: constructing a multi-dimensional data fusion monitoring system, integrating a real-time voltage ripple coefficient detection module of a main power supply, a load transient response analysis module and an environment temperature and humidity sensor array, and generating a dynamic health degree index; the method comprises the following steps: establishing a fault prediction model based on transfer learning, pre-training through multi-scene failure data in a historical database, receiving a health degree index in real time, outputting a reliability attenuation curve, and activating a dual verification mechanism when the slope of the attenuation curve exceeds a preset threshold; extracting characteristic energy distribution of a frequency band of 0.1-10 kHz; executing a transient load impact test, and measuring dynamic response parameters of voltage recovery time and overshoot amplitude; and generating a switching confidence score based on cross processing of the dual verification result and the attenuation curve, and triggering a hierarchical switching instruction. The accuracy of the switching time of the main power supply and the standby power supply can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power supply switching, and in particular to an optimized method and system for power supply switching control. Background Art

[0002] In the field of power supply switching control of a power supply system, seamless switching between a main power supply and a standby power supply is a key technology to ensure power supply continuity. The traditional switching control methods mainly have the following technical defects: Limitations of single-parameter monitoring: Most existing technologies use a single parameter such as voltage or current as the switching judgment basis, which cannot comprehensively reflect the true health status of the power supply system; when the power supply shows early performance degradation, the output voltage may still remain stable, but the ripple coefficient or dynamic response characteristics have already shown abnormalities, resulting in the system missing the best switching opportunity.

[0003] Insufficient adaptability of static thresholds: Most systems use fixed thresholds to trigger switching, and cannot adapt to different load conditions (such as differences between capacitive / inductive loads) and environmental changes (parameter drift caused by temperature fluctuations).

[0004] Disconnection problem between prediction and verification: With the development of artificial intelligence, although existing technologies have introduced prediction algorithms, there is a lack of a real-time verification mechanism for prediction results. When the prediction model is interfered by power grid harmonics, etc., false alarms may occur. In the above-mentioned several existing technologies, the switching opportunity is not good, and even mis-switching may occur. Summary of the Invention

[0005] Therefore, the technical problem to be solved by the present invention is to overcome the technical problems in the prior art that the switching opportunity between the main and standby power supplies is judged inaccurately, and mis-switching or delayed switching is likely to occur. The present invention provides an optimized method and system for power supply switching control. Through the technical chain of "comprehensive monitoring-intelligent prediction-dual verification-cross processing-hierarchical execution", the power supply switching decision is upgraded from experience-driven to data-driven, which can detect the best switching opportunity, avoid waste of resources caused by premature switching, prevent power supply interruption caused by delayed switching, and also solve the problem of mis-switching.

[0006] To solve the above technical problem, the present invention provides an optimized method for power supply switching control, which is used to improve the accuracy of the switching opportunity between the main power supply and the standby power supply in a power supply system, and includes: Construct a multi-dimensional data fusion monitoring system, integrate a real-time voltage ripple coefficient detection module of the main power supply, a load transient response analysis module, and an environmental temperature and humidity sensor array, and generate a dynamic health index; Build a fault prediction model based on transfer learning, pre-train it with multi-scenario failure data in the historical database, and receive the health metrics in real time and output a reliability decay curve, where: When the slope of the decay curve exceeds a preset threshold, activate a dual-verification mechanism: Use wavelet packet decomposition to strip high-frequency noise from the power supply output waveform and extract the characteristic energy distribution in the frequency band of 0.1 - 10 kHz; Perform a transient load impact test and measure the dynamic response parameters of the voltage recovery time and overshoot amplitude; Generate a switching confidence score based on the cross-processing of the dual-verification results and the decay curve. When the score is lower than the safety threshold, trigger a hierarchical switching instruction.

[0007] In an embodiment of the present invention, the construction of the real-time voltage ripple coefficient detection module includes the following steps: Use high-frequency sampling to collect the full-cycle waveform at the main power supply output; Filter out the 50Hz fundamental voltage component and separate the ripple voltage component; Calculate the peak difference of the ripple voltage and compare it with the difference between the preset maximum allowable ripple value and the real-time measured value. Generate a ripple health percentage index based on the comparison result to reflect the health of the current ripple state.

[0008] In an embodiment of the present invention, the implementation of the load transient response analysis module includes the following steps: Apply a stepped load change to the main power supply, with the load step amplitude being 20% - 80% of the rated value and the step rise time not exceeding 10 μs; Record the voltage transient response waveform at a sampling rate of 1 GS / s; Extract the recovery time required for the voltage to return to the steady-state value range and the maximum deviation amplitude exceeding the steady-state voltage during the transient process from the waveform; Calculate the scores of the difference between the measured recovery time and the factory-calibrated reference value, and the difference between the maximum deviation amplitude exceeding the steady-state voltage during the transient process and the reference value according to the weights, and add the two to obtain the load response health score.

[0009] In an embodiment of the present invention, the deployment of the environmental temperature and humidity sensor array includes: Arrange multiple temperature sensors and humidity sensors to cover the air inlet, power component heat dissipation area, air outlet, main circuit board, terminal block, and grounding busbar; Collect the data of each sensor in real time and calculate the temperature impact factor and humidity impact factor respectively: Generate an environmental stress index by combining the temperature and humidity factors to reflect the accelerated impact of the current environment on the equipment life.

[0010] In one embodiment of the present invention, the method for generating the dynamic health index includes: Weighted sum of the ripple health, load response health, and environmental stress index according to the initial weights; When the health of any sub-index drops by more than 20% within three consecutive sampling periods, start the dynamic weight adjustment mechanism: If the ripple health is abnormal or the load response health is abnormal, increase the weight of the abnormal item to 50%, and compress the weights of the remaining indicators proportionally; If the environmental stress is abnormal, trigger an additional calibration process to recalculate the temperature and humidity compensation parameters; Perform a benchmark test every fixed time, verify the measurement accuracy of each sub-index through a standard load, and correct the weight distribution ratio accordingly.

[0011] In one embodiment of the present invention, the process of outputting the reliability decay curve according to the health index includes: Normalize the sub-indicators of the voltage ripple, load response, and environmental parameter health collected in real time, and unify the dimension to the range of 0-100%; Calculate the moving average and trend slope of each sub-index by sliding the window at fixed time intervals to generate a fused feature vector; Input the fused feature vector into a pre-trained LSTM neural network to output a sequence of health prediction values for a future period of time; Connect multiple consecutive predicted value data points to form an initial decay curve, and generate the upper and lower boundaries of the curve based on the historical prediction error range.

[0012] In one embodiment of the present invention, the process of the wavelet packet decomposition for stripping high-frequency noise from the power supply output waveform includes: Perform six-layer decomposition on the power supply waveform using wavelet basis functions to obtain 64 sub-signals in different frequency bands; Calculate the energy distribution entropy value of each sub-frequency band. The higher the entropy value, the more complex the signal in that frequency band; Calculate the total energy ratio of the high-frequency band. When the ratio exceeds 15%, it is determined that there is abnormal noise interference; After filtering out the high-frequency components of the abnormal noise interference, retain the effective feature components in the range of 0.1-10 kHz for subsequent analysis.

[0013] In one embodiment of the present invention, the implementation steps of the transient load impact test include: Apply load impacts at least in three levels, where: the first level applies 20% of the rated load and maintains it for 10 milliseconds; the second level quickly increases to 50% of the load and maintains it for 5 milliseconds; the third level jumps to 80% of the load and maintains it for 2 milliseconds; Record the voltage response curve under each level of load impact; Extract the damping coefficient and oscillation frequency parameters through the least - squares fitting algorithm as additional indicators for evaluating the dynamic stability of the power supply.

[0014] In one embodiment of the present invention, the process of generating the switching confidence score based on the cross - processing of the dual verification result and the decay curve includes: Conduct frequency - domain energy analysis on the high - frequency noise separation result, calculate the proportion of abnormal energy in the frequency band of 0.1 - 10 kHz, and generate a noise interference index; Perform weighted scoring on the voltage recovery time and overshoot amplitude of the transient load impact test to generate a dynamic response index; Intercept the reliability decay curve of the most recent 24 hours and calculate its average slope; Normalize the noise interference index, dynamic response index, and average slope as deduction items to generate a switching confidence score; Set a dynamic safety threshold. When the switching confidence score is lower than the threshold but the gap ≤ 10 points: only switch non - core loads and keep the critical equipment powered; when the switching confidence score is lower than the threshold and the gap > 10 points: immediately initiate a full - system switch; When insulation abnormality is detected: bypass the scoring judgment, directly cut off the faulty line and switch to the standby power supply.

[0015] To solve the above - mentioned technical problems, the present invention also provides a power supply switching control system for improving the accuracy of the switching timing between the main power supply and the standby power supply, including: A multi - dimensional data fusion monitoring module for real - time monitoring of the main power supply operating status, including: a voltage ripple coefficient detection unit configured to collect and analyze the ripple characteristics of the main power supply output voltage in real time; a load transient response analysis unit configured to perform load step tests and record voltage dynamic response parameters; an environmental parameter collection unit including a distributed temperature and humidity sensor array for monitoring the power supply operating environment; A dynamic health assessment module connected to the multi - dimensional data fusion monitoring module, configured to: perform weighted fusion calculation on the collected voltage ripple, load response, and environmental parameters; generate a dynamic health index characterizing the power supply health status; A fault prediction and reliability analysis module, including: A transfer learning model unit pre - trained on a historical fault database for receiving the health index and outputting a reliability decay curve; a dual - verification trigger unit configured to initiate a verification process when the slope of the decay curve exceeds a preset threshold; Dual verification execution module, including: high-frequency noise analysis unit, which processes the power output waveform using wavelet packet decomposition technology to extract the characteristic energy in the frequency band of 0.1 - 10 kHz; dynamic load test unit, configured to perform transient load impact and measure the voltage recovery time and overshoot amplitude; Intelligent decision-making module, configured to: cross-process the dual verification results and the attenuation curve to generate a switching confidence score; when the score is lower than the dynamic safety threshold, output a hierarchical switching control signal; Hierarchical switching execution module, in response to the control signal of the intelligent decision-making module, including: primary switching unit for rapid transfer of non-critical loads; secondary switching unit for seamless switching between main and backup power supplies; emergency protection unit for millisecond-level power-off protection in case of faults.

[0016] The above technical solution of the present invention has the following advantages compared with the prior art: The power supply switching control optimization method described in the present invention mainly solves the technical problems in the prior art that the judgment of the switching timing of the main and backup power supplies is inaccurate, and false switching or delayed switching is likely to occur; By integrating the data of three types of sensors, namely voltage ripple, load transient response, and ambient temperature and humidity, the method of the present invention can comprehensively capture the operating state of the power supply system. This comprehensive monitoring method can detect potential fault signs earlier (such as abnormal ripple prior to voltage drop) compared with single-parameter monitoring, significantly improving the fault prediction ability; Fusing multi-source data into a unified health index solves the decision-making conflict problem caused by independent judgment of multiple parameters in traditional methods, and uses historical failure data for pre-training to enable the system to have the ability to quickly learn from limited new device data.

[0017] When the prediction model shows that the power supply health deteriorates rapidly, dual verification is performed through high-frequency noise analysis and transient load impact test, which can effectively filter out false alarms caused by instantaneous interference (such as power grid fluctuations), ensuring that only real performance degradation will trigger the subsequent process; Based on the cross-processing of the dual verification results and the attenuation curve, the contradiction points that are easily overlooked by traditional methods can be discovered, and misjudgments caused by system failures or algorithm execution errors can be further excluded. Finally, it is quantified by generating a switching confidence score, replacing the traditional yes / no judgment, achieving precise control of the switching timing and performing hierarchical switching. Brief Description of the Drawings

[0018] In order to make the content of the present invention easier to be clearly understood, the following further details the present invention according to the specific embodiments of the present invention and in combination with the drawings, where: Figure 1 is the step flow chart of the power supply switching control optimization method of the present invention; Figure 2 It is the structural framework diagram of the power supply switching control optimization system of the present invention. Detailed implementation manners

[0019] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments, so that those skilled in the art can better understand the present invention and be able to implement it, but the embodiments cited do not limit the present invention. Embodiment 1

[0020] Referring to Figure 1 As shown, the present invention discloses a power supply switching control optimization method for improving the accuracy of the switching timing between the main power supply and the standby power supply in a power supply system. Through a combination of multi-dimensional data fusion, intelligent prediction, and double verification, accurate judgment of the switching timing between the main and standby power supplies is achieved. The specific steps are as follows: S10. Construct a multi-dimensional data fusion monitoring system. According to the actual switching judgment method of the main power supply, deploy three types of key monitoring modules in the monitoring system, namely: a real-time voltage ripple coefficient detection module, a load transient response analysis module, and an environmental temperature and humidity sensor array. Through the real-time fusion of the three types of data, the dynamic health index generated by the system can comprehensively reflect the real-time state of the power supply; Among them: the voltage ripple coefficient detection module can detect early faults such as the aging of the power supply filter capacitor. For example, when the voltage ripple coefficient suddenly increases by 30%, it means that the capacitance performance of the power supply has decreased. The load transient response analysis module can expose defects in the power supply feedback control loop. For example, when the recovery time exceeds twice the nominal value, it indicates that there is an abnormality in the control chip. The environmental temperature and humidity sensor array is used to correct the measured values of other parameters. For example, in a high-temperature environment, the safety threshold of the ripple coefficient is automatically relaxed by 5%, so as to eliminate the influence of environmental factors on the voltage ripple coefficient.

[0021] Specifically, the construction of the real-time voltage ripple coefficient detection module includes the following steps: Use a high-speed differential probe with a sampling rate not less than 1 MHz to capture the full-cycle voltage waveform at the output end of the main power supply. Among them: the probe bandwidth needs to cover the range of 0 - 20 MHz to ensure that the high-frequency ripple unique to the switching power supply (usually 100 kHz - 2 MHz) can be captured; and, to achieve distortion-free sampling, configure an anti-aliasing filter with a cut-off frequency set to 40% of the sampling rate (i.e., 400 kHz) to effectively suppress high-frequency noise interference; compared with traditional power frequency sampling (50 Hz - 1 kHz), high-frequency sampling can reveal potential problems such as the aging of the power supply filter capacitor and abnormal switching tube drive; for example, when the ripple amplitude of the switching frequency of a certain power supply increases from the nominal 50 mVpp to 80 mVpp, the abnormal MOSFET drive circuit can be immediately identified through high-frequency sampling.

[0022] Eliminate the 50Hz sine fundamental component through a digital notch filter. The filter design adopts an adaptive algorithm: it tracks the power grid frequency fluctuation in real time (49.8Hz - 50.2Hz), dynamically adjusts the notch center frequency, sets a stopband attenuation of -40dB, ensures that the residual fundamental component < 1mV, controls the passband ripple within ±0.1dB, and retains the ripple components from 0.1Hz to 500kHz; the filtered signal only contains ripple and noise components, providing a pure data source for subsequent analysis, and can solve the problem that the power frequency component masks the high-frequency ripple.

[0023] Conduct a joint time-domain and frequency-domain analysis on the filtered signal: calculate the peak difference of the ripple voltage with a 1ms window, that is, the difference between the maximum peak and the minimum peak, and display the fluctuation amplitude in real time. Extract the characteristic frequency components through fast Fourier transform to identify abnormal switching frequencies and their harmonics; compare the measured peak difference with the difference between the preset maximum allowable ripple value and the real-time measurement value to generate a ripple health percentage index, reflecting the health degree of the current ripple state.

[0024] Specifically, the implementation of the load transient response analysis module includes the following steps: Generate an accurate stepped load change through a programmable electronic load. The step amplitude is set to 20% - 80% of the rated value (for example, apply steps of 20A → 80A → 20A to a 100A rated power supply), and the step rise time is strictly controlled within 10μs; to achieve fast switching, gallium nitride (GaN) power devices can be used to construct the load circuit, and its switching speed can reach the 2ns level; the potential problems of the power supply control loop can be effectively excited through a wide-range and fast-speed load impact.

[0025] Use a high-speed ADC (analog-to-digital converter) with a sampling rate of 1GS / s to record the voltage transient response. Its resolution is not less than 12bit to ensure that microsecond-level detailed changes can be captured. Set a synchronous trigger mechanism to ensure the precise alignment of the load step and the acquisition clock, with a time synchronization error < 1ns. After the acquired data is processed by digital filtering, it is stored as a waveform file containing timestamps for subsequent analysis.

[0026] Extract two key parameters from the acquired waveform: Recovery time required for the voltage to return to the steady-state value range: defined as the time from when the voltage drops out of the ±1% range of the steady-state value until it re-enters this range and lasts for 1ms. Calculate it in real time using a sliding window algorithm, and the window width is adaptively adjusted; Maximum deviation amplitude beyond the steady-state voltage during the transient process: measure the maximum deviation amount of the voltage beyond the steady-state value during the transient process, which is realized through the combination of a peak hold circuit and a digital comparator.

[0027] Calculate the scores of the differences between the measured recovery time and the factory-calibrated reference value, and the differences between the maximum deviation amplitude exceeding the steady-state voltage and the reference value during the transient process according to the weights, and add the two to obtain the load response health score.

[0028] Specifically, the deployment of the environmental temperature and humidity sensor array includes: Arrange multiple temperature sensors and humidity sensors, and use multiple NTC thermistors to cover three core areas in a honeycomb topology: Inlet and outlet: Monitor the external environmental temperature impact brought by air flow; Power component heat dissipation area: Directly detect the temperature rise of heating components; Main circuit board: Capture local overheating of the PCB; Use multiple capacitive humidity sensors and deploy them at: Terminal block: Detect the risk of insulation performance degradation caused by condensation; Ground busbar: Monitor the corrosion tendency of the grounding system; Inner wall of the power module housing: Evaluate the moisture intrusion caused by seal failure.

[0029] Collect sensor data in real time and calculate the temperature impact factor and humidity impact factor respectively: After collecting data through sensors, it is also necessary to preprocess the data first, including: temperature data calibration and humidity data correction, which are used to eliminate the errors of the sensors themselves and environmental interference to ensure the accuracy of subsequent analysis; then calculate the temperature impact factor and humidity impact factor respectively through the weighted fusion method, that is, for different sensor positions, different weights are assigned according to the actual situation.

[0030] Generate an environmental stress index by integrating the temperature and humidity factors, and use the linear weighted method to integrate the temperature and humidity factors to reflect the accelerated impact of the current environment on the equipment life.

[0031] Specifically, the method for generating the dynamic health index includes: Sum the ripple health, load response health, and environmental stress index by weighting according to the initial weights. In this embodiment, the initial weight allocation is based on the correlation analysis of historical failure data: Ripple anomalies usually indicate hardware aging, so the highest weight is given; The load response reflects the control performance, and the environmental factors affect the long-term reliability, so they are ranked second and third respectively. Therefore, the initial weight ratio can be set as 4:3:3, and the weighted result generates a dynamic health reference value of 0-100 points for preliminary status judgment.

[0032] When the health of any sub-index drops by more than 20% in three consecutive sampling periods, start the dynamic weight adjustment mechanism, where: If the ripple health status is abnormal or the load response health status is abnormal, the weight of the abnormal item is increased to 50%, and the weights of the remaining indicators are compressed proportionally. Since sudden increases in ripple and load surges are often accompanied by capacitor failures or damage to switching devices, increasing their weights can enhance the sensitivity to hardware failures, synchronously activate the high-frequency sampling mode, and enhance the ability to capture ripple details and load changes; If the environmental stress is abnormal, trigger an additional calibration process to recalculate the temperature and humidity compensation parameters. Environmental mutations (such as temperature rise caused by air conditioner failure) may cause other indicators to be distorted. Special calibration can eliminate systematic errors and reduce misjudgments caused by environmental interference.

[0033] Perform a baseline test every fixed time, verify the measurement accuracy of each sub-indicator through a standard load, and accordingly correct the weight allocation ratio. Regular calibration can eliminate progressive errors such as sensor aging and circuit drift, and ensure that the physical meaning of the weight allocation is consistent with the actual situation.

[0034] In this embodiment, dynamic weight adjustment can trigger an early warning in the early fault stage (such as a 10% decrease in capacitor capacitance). The special environmental calibration can eliminate more than 90% of the temperature drift false alarms (such as the falsely high ripple coefficient caused by high temperature), avoid unnecessary switching actions caused by environmental interference, and finally use periodic baseline tests to enable the system to have self-healing capabilities. Even if the sensor performance deteriorates year by year, the evaluation accuracy can still be maintained through algorithm compensation.

[0035] S20. Establish a fault prediction model based on transfer learning, pre-train it through multi-scenario failure data in the historical database, and receive the health status indicators in real time and output a reliability decay curve.

[0036] Specifically, the training process of the transfer learning fault prediction model includes: Extract voltage waveforms, load responses, and environmental data from the historical fault library, collect the operation data of different models of power supplies throughout their life cycles, including normal decay, sudden failure cases, and edge cases, and align the voltage waveforms, load response curves, and environmental parameters in time to ensure the spatio-temporal consistency of the data.

[0037] Perform joint time-frequency domain feature extraction to form a dataset containing multiple feature dimensions. Extract multi-dimensional statistics such as ripple peak difference, mean offset, and zero-crossing rate as time-domain features, calculate multi-dimensional indicators such as energy distribution entropy and harmonic distortion rate in the 0.1 - 10 kHz frequency band as frequency-domain features, and perform oversampling on rare fault type data to balance the number of various fault samples.

[0038] Construct a dual-channel neural network model, which is divided into a time-domain analysis channel and a frequency-domain analysis channel according to the above dataset. In the time-domain analysis channel, analyze the time-domain characteristics of the voltage waveform within a unit time window. In the frequency-domain analysis channel, process the frequency-domain characteristics after fast Fourier transform. Concatenate the multi-dimensional feature vectors of the time-domain channel and the multi-dimensional vectors of the frequency-domain channel, and map them to the health score through a fully connected layer.

[0039] Load the parameters of the pre-trained general image recognition model, freeze the parameters of the first five layers of the network, and only perform special training on the subsequent fully connected layer for the power health prediction task, which significantly reduces the requirement of the new device model for the amount of labeled data and reduces the data acquisition cost.

[0040] Specifically, the process of outputting the reliability decay curve according to the health index includes: Normalize the sub-health indicators of the voltage ripple, load response, and environmental parameters collected in real time, unify the dimension to the range of 0-100%, and automatically update the normalization benchmark according to the operation years of the device. For example, for a power supply used for 5 years, the parameter of "minimum allowable ripple value" will be relaxed accordingly to avoid misjudgment of aging devices; the normalization process solves the problem of unbalanced feature weights caused by inconsistent dimensions of multi-sensor data. By introducing an aging compensation factor, devices with different service periods can adopt the same set of evaluation criteria, which is the key prerequisite for achieving long-term prediction stability.

[0041] Adopt a sliding time window (default 24 hours) for feature synthesis, take the exponentially weighted average of the data within the window for each sub-health indicator, where the weight of recent data is set to be relatively heavy to highlight the impact of the latest state. Fit the short-term (6 hours) change rate of each sub-indicator by the least squares method to capture the acceleration of performance decline. For example, when the slope of the ripple health changes suddenly from -0.5% / h to -2% / h, it indicates that the capacitor may enter the rapid failure period; concatenate the current value, moving average value, and slope value of each sub-indicator into a 9-dimensional vector (3 indicators × 3 features) as the input of the LSTM network.

[0042] Input the fused feature vector into the pre-trained LSTM neural network. The forget gate in the network dynamically adjusts the memory weight according to the historical state. Adopt an encoder-decoder architecture, first encode the feature sequence of the past 24 hours, and then autoregressively predict the health values in the next 72 hours. In the decoding stage, the model will focus on the historical segment most relevant to the current prediction moment and output a sequence of health prediction values for a period of time in the future. The time-series memory ability of LSTM enables it to learn the non-linear trajectory of power health changes, while the attention mechanism solves the problem of information dilution in long-term prediction.

[0043] Connect multiple consecutive predicted value data points to form an initial decay curve, generate upper and lower boundaries of the curve based on the historical prediction error range, form a dynamic confidence interval, and construct an adaptive prediction fault tolerance space. When the environment suddenly changes or the load is abnormal, the system will automatically reduce the certainty weight of the prediction result to avoid misjudgment caused by overconfidence.

[0044] In this embodiment, normalization and feature fusion enable the system to identify early weak decay signals and trigger an alarm at the initial stage of abnormal power supply changes. Through the time series modeling ability of LSTM combined with the dynamic confidence interval, false alarms caused by power grid transient disturbances are effectively filtered. This prediction method based on a deep time series model essentially constructs a "digital twin" of the power supply health state. By continuously comparing the deviation between the predicted curve and the actual measurement value, the system can continuously correct the estimation accuracy of the remaining life of the device and update the determination of the switching timing.

[0045] S30. When the slope of the decay curve exceeds the preset threshold, activate the dual verification mechanism, including: Use wavelet packet decomposition to strip high-frequency noise from the power supply output waveform, extract the characteristic energy distribution in the frequency band of 0.1 - 10 kHz. Through high-frequency noise analysis, sudden hardware damage can be discovered, and characteristic information reflecting real faults can be accurately extracted from complex power supply signals. Perform a transient load impact test, measure the dynamic response parameters of the voltage recovery time and overshoot amplitude. Through the load test, the continuous power supply ability of the power supply can be verified.

[0046] Specifically, the process of using wavelet packet decomposition to strip high-frequency noise from the power supply output waveform includes: Use wavelet basis functions to decompose the power supply waveform into six layers. Each layer of decomposition divides the signal frequency band in half. After six-layer iteration, the original signal is divided into 64 equally wide sub-bands, obtaining 64 sub-signals in different frequency bands. The wavelet basis functions have the characteristics of compact support and 6th-order vanishing moments, which can not only effectively capture transient mutations (such as voltage spikes) but also smoothly process steady-state waveforms. The time-frequency localization characteristics of wavelet packet decomposition enable it to accurately locate the specific time and frequency range where noise appears. The depth selection of six-layer decomposition is the best balance between computational complexity and frequency resolution.

[0047] Analyze the energy distribution of each decomposed sub-band: calculate the energy distribution entropy value of each sub-band. The higher the entropy value, the more complex the signal in that frequency band. Use the Shannon entropy formula to calculate the energy distribution uncertainty of each sub-band. The entropy value in the high-frequency band (>50 kHz) should be relatively low (about 0.2 - 0.5) during normal operation. If the entropy value suddenly increases to >0.8, it indicates the existence of random noise.

[0048] Calculate the total energy proportion of the high-frequency band, and count the total energy proportion of sub-bands No. 32 - 64 (corresponding to the high-frequency band of 250 kHz - 1 MHz). For a normal power supply, this ratio < 10%. When this proportion exceeds 15%, it is determined that there is abnormal noise interference. In this embodiment, the threshold of 15% is set based on a large number of measured data statistics, which can effectively distinguish the broadband spectral characteristics of normal switching noise (concentrated in a narrow band) and faults such as arc discharge.

[0049] After filtering out the high-frequency components of the abnormal noise interference, retain the effective characteristic components in the range of 0.1 - 10 kHz for subsequent analysis. This range includes: the change characteristics of the electrolytic capacitor ESR (1 - 5 kHz), PWM modulation harmonics (switching frequency ± 5 kHz), and the oscillation characteristics of the control loop (100 Hz - 10 kHz), realizing the collaborative optimization of noise suppression and feature enhancement.

[0050] In this embodiment, wavelet packet decomposition is used to strip the high-frequency noise from the power supply output waveform, effectively suppressing the pollution of electromagnetic interference to the detection system. The energy entropy analysis of the high-frequency band can detect the early degradation of semiconductor devices (such as the quantum tunneling noise caused by MOSFET gate oxide layer defects), and can accurately obtain the fault omen signal from a large number of normal signals.

[0051] Specifically, the implementation steps of the transient load impact test include: Apply load impacts at least in three levels. Among them: apply 20% of the rated load at the first level to simulate the start and stop of small-power devices, and maintain it for 10 milliseconds to observe the initial response characteristics. Quickly increase to 50% of the load at the second level to achieve a load step, representing the switching of typical working states, and maintain it for 5 milliseconds to ensure that the system enters a quasi-steady state. Jump to 80% of the load at the third level to approach the maximum capacity of the power supply, and maintain it for 2 milliseconds to test the instantaneous overload tolerance. The three load impacts constitute a "progressive - sudden increase - limit" composite test mode, which can comprehensively stimulate different response characteristics of the power supply control loop: the 20% step mainly tests the small-signal stability, the 50% test tests the mid-frequency band adjustment ability, and the 80% impact verifies the large-signal transient response. The duration settings of 10 ms / 5 ms / 2 ms correspond to the time constants of the slow integration link, the medium-speed proportional link, and the fast differential link of the power supply system respectively.

[0052] Record the voltage response curve under each level of load impact, use synchronous measurement technology to capture the dynamic response, equip with a high-speed ADC with a sampling rate of 1 GS / s, and cooperate with 16-bit high resolution, which is sufficient to capture the subtle changes within the typical control cycle of the switching power supply (usually 10 - 100 μs), ensure that the details of fluctuations at the 10 μs level can be analyzed, and use the rising edge of the load switching signal as the acquisition trigger reference.

[0053] Extract the damping coefficient and oscillation frequency parameters through the least - squares fitting algorithm as additional indicators for evaluating the dynamic stability of the power supply. The decrease in the damping coefficient may indicate an increase in the output capacitance, while the increase in the oscillation frequency suggests parameter drift in the compensation network, upgrading the evaluation of the power supply's dynamic performance from qualitative description to quantitative indicators.

[0054] S40. Generate a switching confidence score based on the cross - processing of the dual - verification results and the decay curve, and trigger a hierarchical switching instruction when the score is lower than the safety threshold.

[0055] Specifically, the process of generating the switching confidence score based on the cross - processing of the dual - verification results and the decay curve includes: Conduct frequency - domain energy analysis on the high - frequency noise separation results, calculate the proportion of abnormal energy in the 0.1 - 10 kHz frequency band to generate a noise interference index; perform weighted scoring on the voltage recovery time and overshoot amplitude of the transient load impact test to generate a dynamic response index; intercept the reliability decay curve of the most recent 24 hours and calculate its average slope.

[0056] Normalize the noise interference index, dynamic response index, and average slope as deduction items to generate a switching confidence score; set a dynamic safety threshold. When the switching confidence score is lower than the threshold but the gap ≤ 10 points: only switch non - core loads and keep the critical equipment powered; when the switching confidence score is lower than the threshold and the gap > 10 points: immediately initiate a full - system switch.

[0057] When insulation abnormality is detected: bypass the score judgment, directly cut off the faulty line and switch to the standby power supply. Embodiment 2

[0058] Based on the above - mentioned Embodiment 1, in order to implement the optimization method for power supply switching control, as shown in Figure 2 This embodiment provides a power supply switching control system that can execute the above - mentioned method to improve the accuracy of the switching timing between the main power supply and the standby power supply, including: A multi - dimensional data fusion monitoring module for real - time monitoring of the main power supply operating status, including: a voltage ripple coefficient detection unit configured to collect and analyze the ripple characteristics of the main power supply output voltage in real time; a load transient response analysis unit configured to perform a load step test and record voltage dynamic response parameters; an environmental parameter collection unit including a distributed temperature and humidity sensor array for monitoring the power supply operating environment; A dynamic health assessment module connected to the multi - dimensional data fusion monitoring module, configured to: perform weighted fusion calculation on the collected voltage ripple, load response, and environmental parameters; generate a dynamic health index characterizing the power supply health status; A fault prediction and reliability analysis module, including: A transfer learning model unit, pre-trained on a historical fault database, is configured to receive a health index and output a reliability decay curve; a dual-verification trigger unit is configured to initiate a verification process when the slope of the decay curve exceeds a preset threshold; A dual-verification execution module includes: a high-frequency noise analysis unit that processes the power output waveform using wavelet packet decomposition technology to extract the characteristic energy in the 0.1 - 10 kHz frequency band; a dynamic load test unit that is configured to perform a transient load impact and measure the voltage recovery time and overshoot amplitude; An intelligent decision-making module is configured to: cross-process the dual-verification results and the decay curve to generate a switching confidence score; when the score is lower than a dynamic safety threshold, output a hierarchical switching control signal; A hierarchical switching execution module, in response to the control signal of the intelligent decision-making module, includes: a primary-level switching unit for the rapid transfer of non-critical loads; a secondary-level switching unit for the seamless switching of the main and backup power supplies; an emergency protection unit for millisecond-level power-off protection in case of a fault.

[0059] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0060] Obviously, the above embodiments are merely examples for clear illustration and are not limitations on the implementation manners. For those of ordinary skill in the art, other different forms of changes or variations can be made based on the above description. It is not necessary and impossible to enumerate all the implementation manners here. And the obvious changes or variations derived therefrom are still within the protection scope of the present invention.

Claims

1. A power supply switching control optimization method for improving the accuracy of the switching timing between the main power supply and the backup power supply in a power supply system, characterized in that: include: Build a multi-dimensional data fusion monitoring system that integrates the real-time voltage ripple coefficient detection module of the main power supply, the load transient response analysis module, and the environmental temperature and humidity sensor array to generate dynamic health indicators; Establish a fault prediction model based on transfer learning, pre-train with multi-scenario failure data in the historical database, receive the health index in real time and output the reliability decay curve; The double verification mechanism is activated when the slope of the decay curve exceeds the preset threshold: Wavelet packet decomposition is used to remove high-frequency noise from the power supply output waveform and extract the characteristic energy distribution in the frequency range of 0.1 to 10 kHz. Perform transient load impact test to measure the dynamic response parameters of voltage recovery time and overshoot amplitude; Based on the cross-processing of the double verification results and the attenuation curve, a switching confidence score is generated, and a graded switching instruction is triggered when the score is lower than the safety threshold.

2. The power supply switching control optimization method according to claim 1, characterized in that: The construction of the real-time voltage ripple coefficient detection module includes the following steps: High-frequency sampling is used to collect full-cycle waveforms at the output end of the main power supply; Filter out the 50Hz basic voltage component and separate the ripple voltage component; The peak value difference of the ripple voltage is calculated and compared with the difference between the preset maximum allowable ripple value and the real-time measurement value. Based on the comparison result, a ripple health percentage index is generated to reflect the health of the current ripple state.

3. The power supply switching control optimization method according to claim 1, characterized in that: The implementation of the load transient response analysis module includes the following steps: Apply a step-type load change to the main power supply, with the load step amplitude ranging from 20% to 80% of the rated value and the step rise time not exceeding 10μs; The voltage transient response waveform is recorded at a sampling rate of 1GS / s; Extract the recovery time required for the voltage to return to the steady-state value range and the maximum deviation amplitude beyond the steady-state voltage during the transient process from the waveform; The difference between the measured recovery time and the factory calibrated reference value, as well as the difference between the maximum deviation amplitude of the steady-state voltage during the transient process and the reference value are weighted and scored, and the two are added together to obtain the load response health score.

4. The power supply switching control optimization method according to claim 1, characterized in that: The deployment of the environmental temperature and humidity sensor array includes: Arrange multiple temperature sensors and humidity sensors to cover the air inlet, power component heat dissipation area, air outlet, main circuit board, wiring terminals and ground busbar; Collect data from each sensor in real time and calculate the temperature and humidity influencing factors respectively: The temperature and humidity factors are combined to generate an environmental stress index, which reflects the accelerated impact of the current environment on the life of the equipment.

5. The power supply switching control optimization method according to claim 1, characterized in that: The method for generating the dynamic health index includes: The ripple health, load response health and environmental stress index are weighted and summed according to the initial weights; When the health of any sub-indicator drops by more than 20% in three consecutive sampling periods, the dynamic weight adjustment mechanism is activated: If the ripple health is abnormal or the load response health is abnormal, the weight of the abnormal item will be increased to 50%, and the weights of other indicators will be compressed proportionally; If the environmental stress is abnormal, an additional calibration process is triggered to recalculate the temperature and humidity compensation parameters; The benchmark test is performed once at a fixed time, and the measurement accuracy of each sub-indicator is verified through the standard load, and the weight distribution ratio is corrected accordingly.

6. The power supply switching control optimization method according to claim 1, characterized in that: The process of outputting a reliability decay curve based on the health index includes: Normalize the real-time collected voltage ripple, load response and environmental parameter health sub-indicators to unify the dimensions to the range of 0 to 100%; Calculate the moving average and trend slope of each sub-indicator by sliding the window at fixed time intervals to generate a fusion feature vector; Input the fused feature vector into a pre-trained LSTM neural network to output a sequence of predicted health values ​​for a period of time in the future; Connect multiple consecutive forecast value data points to form an initial decay curve, and generate the upper and lower boundaries of the curve based on the historical forecast error range.

7. The power supply switching control optimization method according to claim 1, characterized in that: The process of removing high-frequency noise from the power output waveform by wavelet packet decomposition includes: The power waveform is decomposed into six layers using wavelet basis functions to obtain 64 sub-signals in different frequency bands. Calculate the energy distribution entropy value of each sub-band. The higher the entropy value, the more complex the signal in the frequency band. The total energy proportion of the high frequency band is calculated, and when the proportion exceeds 15%, it is determined that abnormal noise interference exists; After filtering out the high-frequency components of abnormal noise interference, the effective characteristic components in the range of 0.1 to 10 kHz are retained for subsequent analysis.

8. The power supply switching control optimization method according to claim 1, characterized in that: The implementation steps of the transient load impact test include: The load shock is applied in at least three levels, where: the first level applies 20% of the rated load and maintains it for 10 milliseconds; the second level quickly increases to 50% load and maintains it for 5 milliseconds; the third level jumps to 80% load and maintains it for 2 milliseconds; Record the voltage response curve under each level of load impact; The damping coefficient and oscillation frequency parameters are extracted by the least squares fitting algorithm as additional indicators for evaluating the dynamic stability of the power supply.

9. The power supply switching control optimization method according to claim 1, characterized in that: The process of generating a switching confidence score based on the cross processing of the double verification result and the attenuation curve includes: Perform frequency domain energy analysis on the high-frequency noise separation results, calculate the abnormal energy ratio in the 0.1-10kHz frequency band, and generate the noise interference index; The voltage recovery time and overshoot amplitude of the transient load impact test are weighted and scored to generate a dynamic response index; Intercept the reliability decay curve of the last 24 hours and calculate its average slope; The noise interference index, dynamic response index and average slope are normalized as deduction items to generate a switching confidence score; Set a dynamic safety threshold. When the switching confidence score is lower than the threshold, but the difference is ≤10 points, only non-core loads are switched to keep key equipment powered. When the switching confidence score is lower than the threshold and the difference is >10 points, the whole system is switched immediately. When insulation abnormality is detected: bypass the scoring judgment, directly cut off the fault line and switch to the backup power supply.

10. A power supply switching control system, used to improve the accuracy of the switching timing between the main power supply and the backup power supply, characterized in that: include: The multi-dimensional data fusion monitoring module is used to monitor the main power supply operation status in real time, including: a voltage ripple coefficient detection unit configured to collect and analyze the ripple characteristics of the main power supply output voltage in real time; a load transient response analysis unit configured to perform a load step test and record voltage dynamic response parameters; an environmental parameter acquisition unit, including a distributed temperature and humidity sensor array, used to monitor the power supply operation environment; A dynamic health evaluation module is connected to the multi-dimensional data fusion monitoring module and is configured to: perform weighted fusion calculation on the collected voltage ripple, load response and environmental parameters; and generate a dynamic health index representing the health status of the power supply; Fault prediction and reliability analysis module, including: A transfer learning model unit, pre-trained on a historical fault database, for receiving health indicators and outputting a reliability decay curve; a dual verification trigger unit, configured to initiate a verification process when the slope of the decay curve exceeds a preset threshold; The dual verification execution module includes: a high-frequency noise analysis unit, which uses wavelet packet decomposition technology to process the power supply output waveform and extract the characteristic energy of the 0.1-10kHz frequency band; a dynamic load test unit, which is configured to perform transient load impact and measure the voltage recovery time and overshoot amplitude; The intelligent decision-making module is configured to: cross-process the double verification results and the attenuation curve to generate a switching confidence score; when the score is lower than the dynamic safety threshold, output a graded switching control signal; The hierarchical switching execution module, in response to the control signal of the intelligent decision-making module, includes: a primary switching unit for rapid transfer of non-critical loads; a secondary switching unit for seamless switching of primary and standby power supplies; and an emergency protection unit for millisecond-level power-off protection under fault conditions.

Images