A method, device, terminal equipment and medium for predicting the electrolyte replenishment level of train batteries.

By collecting train vibration and temperature data, adaptively generating spread spectrum detection signals and performing signal constraint separation and temperature compensation, the problem of accurately extracting battery liquid shortage characteristics and predicting liquid replenishment amount under train vibration and temperature changes is solved, thus improving detection efficiency and accuracy.

CN123115652APending Publication Date: 2026-11-17CENT SOUTH UNIV
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
CN202611254920.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-18
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing technologies struggle to accurately extract the evolution characteristics of battery liquid shortage and predict the amount of liquid to replenish under conditions of train vibration and temperature changes. In particular, under conditions of multiple reflections and aliasing, as well as temperature effects, the detection efficiency and anti-interference capabilities are insufficient.

Method used

By collecting train vibration signals and temperature data, the pseudo-random code length is adaptively determined to generate a spread spectrum detection signal. The reflected signal is separated by using link parameters and reflection coefficient constraints, and then subjected to in-phase correction and weighted superposition. Combined with cross-correlation processing and temperature compensation, a liquid replenishment prediction model is established.

Benefits of technology

This improves the stability of low-liquidity feature extraction and the accuracy of liquid replenishment prediction under complex operating conditions, extending battery life and train safety.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123115652A_ABST
    Figure CN123115652A_ABST
Patent Text Reader

Abstract

This invention provides a method, device, terminal equipment, and medium for predicting the replenishment amount of train batteries. The method involves collecting train vibration signals and temperature data of the battery under test. The pseudo-random code length is determined based on the vibration intensity, and a spread spectrum detection signal is generated. This signal is injected into the battery via a pre-set detection link, simultaneously acquiring a reference incident signal and a total reflected signal. The first and subsequent reflected signals are separated based on link parameters, propagation delay, reflection coefficient range, and the successive attenuation relationship of adjacent reflections. These signals are then combined with peak polarity, delay, and reflection coefficient for in-direction correction and weighted superposition to obtain an enhanced reflection signal. This enhanced reflection signal is cross-correlated with the reference incident signal and temperature-compensated to extract the low-fluid evolution characteristics, which are then input into a pre-trained model to obtain the predicted replenishment amount. This invention improves the accuracy of predicting the replenishment amount of the battery under test, thereby extending battery life and improving train safety.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of battery status monitoring technology for rail transit vehicles, specifically to a method, device, terminal equipment, and medium for predicting the replenishment amount of train batteries. Background Technology

[0002] Train battery banks are crucial backup power sources in the DC auxiliary power supply systems of rail transit vehicles. In situations such as overhead contact line failure or auxiliary power supply malfunction, they are required to supply power to loads such as lighting, communication, control, and emergency ventilation. Liquid electrolyte batteries experience electrolyte loss during long-term charging and discharging due to factors such as high temperature, overcharging, frequent fast charging, and natural evaporation. When the electrolyte level decreases, the equivalent impedance and electrochemical characteristics of the battery terminals change, potentially leading to reduced charge acceptance, accelerated capacity decay, and poorer cell consistency, thereby affecting the reliability of the auxiliary power supply.

[0003] Current methods for detecting low electrolyte levels in train batteries typically involve manual inspection, observation of electrolyte levels, and replenishment according to maintenance schedules. These methods require opening the battery box or inspecting each cell individually, resulting in low detection efficiency. Furthermore, visual abnormalities are usually only detected after the low electrolyte level has progressed to a significant stage, making it difficult to continuously reflect the evolution of the low electrolyte condition and directly provide the required amount of electrolyte to be replenished for the target battery.

[0004] To utilize impedance changes caused by electrolyte deficiency, existing technologies also employ time-domain reflectometry (TD-RS). This involves injecting electrical pulses into the test object and analyzing the reflected waves formed at impedance discontinuities to identify the internal state or defects of the battery. For example, US 12,429,525 B2 discloses a scheme for detecting internal battery defects using time-domain reflected waves and a reference waveform. Spread-spectrum TD-RS further employs pseudo-random code spread-spectrum signals and related processing, which can improve the anti-interference capability of low-amplitude reflected signals and has been used for fault detection in transmission objects such as cables.

[0005] However, in the train operating environment, vibration causes fluctuations in terminal contact status and reflection peak position, making it difficult to simultaneously balance detection efficiency and anti-interference capability with fixed spread spectrum parameters. Multiple reflections occur at connection points, cables, and battery ports in the detection link, and aliasing reflections can distort single reflection waveforms or cross-correlation peaks. Simultaneously, temperature changes alter battery impedance, coupling temperature effects with liquid level changes. Existing technologies typically rely on single reflection waveforms or single peak values ​​for judgment, lacking physical constraints to separate aliasing reflections, enhancement of weak reflection characteristics, and temperature compensation. Therefore, it is difficult to stably extract the liquid shortage evolution characteristics that change with the detection cycle and further predict the required replenishment volume. Thus, a technical solution is needed that can adapt to train vibration and temperature changes, reduce the impact of multiple reflection aliasing, and achieve quantitative prediction of replenishment volume. Summary of the Invention

[0006] The technical problem to be solved by this invention is: how to accurately extract the characteristics of battery liquid shortage evolution and predict the amount of liquid to be added to the battery under test under the conditions of train vibration, temperature change and multiple reflection interference of the detection link.

[0007] In a first aspect, the present invention provides a method for predicting the electrolyte replenishment level of a train battery, the method comprising the following steps: Collect the train vibration signal of the target train and the temperature data of the battery under test in the target train; The vibration intensity is determined based on the train vibration signal, and the pseudo-random code length is determined based on the vibration intensity. The pseudo-random code is used to generate the spread spectrum detection signal, and the pseudo-random code length is positively correlated with the vibration intensity. A spread spectrum detection signal corresponding to the pseudo-random code length is generated, a reference incident signal corresponding to the spread spectrum detection signal is obtained, the spread spectrum detection signal is injected into the battery under test through a preset detection link, and the total reflection signal is collected simultaneously; the total reflection signal is formed by the reflection of the spread spectrum detection signal at the impedance discontinuity between the preset detection link and the battery under test. Obtain the link parameters of the preset detection link, determine the propagation delay of each reflection component in the total reflection signal based on the link parameters, and constrain and separate the total reflection signal based on the propagation delay, the preset range of reflection coefficient values, and the successive attenuation relationship between two adjacent reflections formed by the same reflection propagation path in the preset detection link, to obtain the first reflection signal and at least one subsequent reflection signal. Based on the peak polarity, propagation delay, and reflection coefficient of each subsequent reflected signal relative to the first reflected signal, the first reflected signal and each subsequent reflected signal are corrected in the same direction and weighted and superimposed to obtain the enhanced reflected signal; Cross-correlation processing is performed on the enhanced reflected signal and the reference incident signal to extract the cross-correlation peak value within the preset time delay window. Temperature compensation is then performed on the cross-correlation peak value based on the temperature data to obtain the liquid shortage evolution characteristics that characterize the change of the liquid shortage state with the detection cycle. The fluid shortage evolution characteristics are input into a pre-trained fluid replenishment prediction model to obtain the predicted fluid replenishment amount of the battery under test. The fluid replenishment prediction model is established based on spread spectrum time-domain reflectometry data under different fluid level conditions and the corresponding actual fluid replenishment amount.

[0008] Optionally, the link parameters include the transmission path length, sampling frequency, and transmission path structure of the preset detection link; Based on the propagation delay, the preset range of reflection coefficient values, and the successive attenuation relationship between two adjacent reflections formed by the same reflection propagation path in the preset detection link, the total reflection signal is constrained and separated to obtain a first reflection signal and at least one subsequent reflection signal, including: The reflection components formed by the same reflection propagation path in the preset detection link are numbered in ascending order of propagation delay. The reference incident signal is then time-shifted according to the propagation delay of each reflection component to construct a delay test matrix. The expression for the delay test matrix is ​​as follows: ,in, Represents the delay test matrix; Indicates the sampling time; Indicates the first The propagation delay of each reflected component , This represents the number of reflection components to be separated, formed by the same reflection propagation path, and Not less than 2 ; Indicates the reference incident signal; Indicates according to the propagation delay The time-shifted reference incident signal; The objective is to minimize the difference between the total reflected signal and the product of the time delay test matrix and the reflection coefficient vector. An optimization model with boundary constraints on reflection coefficients, smoothing constraints on adjacent reflection coefficients, and successive attenuation constraints is used to obtain the reflection coefficients corresponding to each reflection component. The expression of the optimization model is as follows:

[0009] in, This represents the reflection coefficient vector, which consists of the reflection coefficients corresponding to each reflection component. Indicates the total reflected signal; This represents the first-order difference matrix used to determine the difference between adjacent reflection coefficients; Indicates the first The reflection coefficient corresponding to each reflection component ; This represents the attenuation coefficient between two adjacent reflections formed by the same reflection propagation path. ; This represents the first regularization coefficient corresponding to the smoothing constraint of adjacent reflection coefficients; This represents the second regularization coefficient corresponding to the successive decay constraint; Based on the propagation delay and reflection coefficient corresponding to each reflection component, the reference incident signal is time-shifted and amplitude-adjusted to reconstruct the reflected signals corresponding to each reflection component. The reflected signal corresponding to the reflection component with the smallest propagation delay is determined as the first reflected signal, and the reflected signals corresponding to the remaining reflection components are determined as subsequent reflected signals; where, the first... The reconstructed reflection signal corresponding to each reflection component is represented as follows: ;in, Indicates the first reflected signal. exist A value not less than 2 indicates a subsequent reflected signal.

[0010] Optionally, based on the peak polarity, propagation delay, and reflection coefficient of each subsequent reflected signal relative to the first reflected signal, the first reflected signal and each subsequent reflected signal are subjected to in-direction correction and weighted superposition to obtain an enhanced reflected signal, including: Determine the signal-to-noise ratio (SNR) of each subsequent reflected signal, and identify the subsequent reflected signals with an SNR not lower than a preset compensation threshold as valid subsequent reflected signals; The peak polarity correction coefficient is determined based on the peak polarity of each effective subsequent reflected signal relative to the first reflected signal, and the compensation weight is determined based on the propagation delay and reflection coefficient of each effective subsequent reflected signal; the calculation expression for the compensation weight is as follows:

[0011] in, Indicates the first The compensation weights corresponding to each subsequent reflected signal This represents the set consisting of the sequence numbers of each valid subsequent reflected signal; This indicates the propagation delay of the first reflected signal; Indicates the first The propagation delay of each subsequent reflected signal; Represents the time decay constant; Indicates the first The reflection coefficient of the subsequent reflected signal; Indicates the first The reflection coefficient of an effective subsequent reflected signal; Indicates the first The propagation delay of an effective subsequent reflected signal; Based on the peak polarity simultaneity correction coefficient and compensation weight, each effective subsequent reflection signal is weighted and superimposed with the first reflection signal to obtain the enhanced reflection signal; the expression for the enhanced reflection signal is:

[0012] in, Indicates enhanced reflected signal; Indicates the first The peak polarity correction coefficient corresponding to the i-th subsequent reflected signal, when the i-th... When the peak polarity of each subsequent reflected signal is the same as that of the first reflected signal, When the first When the peak polarity of a subsequent reflected signal is opposite to that of the first reflected signal, ; Indicates the first A subsequent reflected signal.

[0013] Optionally, cross-correlation processing is performed on the enhanced reflected signal and the reference incident signal to extract the cross-correlation peak value within a preset time delay window, and temperature compensation is performed on the cross-correlation peak value based on temperature data, including: Through calculation formula

[0014] Get the first Each detection cycle has a time delay Cross-correlation value at the location ; Indicates the first The enhanced reflection signal corresponding to each detection cycle; Through calculation formula

[0015] Get the first The cross-correlation peak value corresponding to each detection cycle ; Indicates the preset delay window; A temperature influence function is established based on the cross-correlation peak values ​​corresponding to different temperatures under normal liquid level conditions. And through calculation formula

[0016] Get the first The temperature-compensated cross-correlation peak value corresponding to each detection cycle ; This indicates the effect of temperature on temperature data. The function value at that location.

[0017] Optionally, obtain the fluid shortage evolution characteristics that characterize the change of the fluid shortage state over the detection cycle, including: The temperature-compensated cross-correlation peak values ​​are arranged according to the detection cycle to obtain the temperature-compensated peak value sequence; Integrated empirical mode decomposition was performed on the temperature-compensated peak sequence to obtain multiple intrinsic mode function components; Select intrinsic mode function components with a correlation coefficient of not less than 0.3 with the temperature-compensated peak sequence, or select low-frequency intrinsic mode function components with a cumulative energy contribution rate of not less than 80%. The selected intrinsic mode function components are reconstructed to obtain the principal component peak curves, which are then identified as the characteristics of the fluid deficiency evolution.

[0018] Optionally, the pre-trained fluid replacement prediction model can be one of the following: calibration curve model, multinomial regression model, piecewise regression model, support vector regression model, or neural network model. The pre-trained liquid replenishment prediction model takes the peak ratio, peak drop, short-term slope, temperature data, vibration intensity, battery model parameters, and historical liquid replenishment records as inputs, and the predicted liquid replenishment volume as output; the peak ratio, peak drop, and short-term slope are determined based on the principal component peak curves. The formula for calculating the peak ratio is: ;in, Indicates the first Peak percentage corresponding to each detection cycle; Indicates the peak curve of the principal component at the th The peak value of the principal component corresponding to each detection cycle; This represents the peak value of the baseline principal component when the battery under test is in a normal liquid level state. The expression for calculating the peak descent is as follows: ;in, Indicates the first The peak value decrease corresponding to each detection cycle; The expression for calculating the short-time slope is as follows: ;in, Indicates the first The short-time slope corresponding to each detection cycle; This indicates the number of detection cycles used to determine the short-term slope; Indicates the peak curve of the principal component at the th The principal component peak value corresponding to each detection cycle.

[0019] In a second aspect, the present invention provides a train battery electrolyte replenishment prediction device, comprising: The data acquisition module is used to collect the train vibration signal of the target train and the temperature data of the battery under test in the target train, and input the train vibration signal and temperature data into the processor. The processor is used to determine the vibration intensity based on the train vibration signal, determine the pseudo-random code length based on the vibration intensity, and send the pseudo-random code length to the spread spectrum detection module; the pseudo-random code is used to generate the spread spectrum detection signal, and the pseudo-random code length is positively correlated with the vibration intensity. The spread spectrum detection module is used to generate a spread spectrum detection signal corresponding to the length of the pseudo-random code, acquire the reference incident signal corresponding to the spread spectrum detection signal, inject the spread spectrum detection signal into the battery under test through a preset detection link, simultaneously acquire the total reflection signal, and input the reference incident signal and the total reflection signal into the processor; the total reflection signal is formed by the reflection of the spread spectrum detection signal at the impedance discontinuity between the preset detection link and the battery under test; The processor is also used to acquire the link parameters of the preset detection link, determine the propagation delay of each reflection component in the total reflection signal based on the link parameters, and constrain and separate the total reflection signal based on the propagation delay, the preset reflection coefficient value range, and the successive attenuation relationship between two adjacent reflections formed by the same reflection propagation path in the preset detection link, to obtain the first reflection signal and at least one subsequent reflection signal. The processor is also used to perform in-direction correction and weighted superposition of the first reflection signal and each subsequent reflection signal based on the peak polarity, propagation delay and reflection coefficient of each subsequent reflection signal relative to the first reflection signal, so as to obtain an enhanced reflection signal; The processor is also used to perform cross-correlation processing on the enhanced reflected signal and the reference incident signal, extract the cross-correlation peak value within the preset time delay window, and perform temperature compensation on the cross-correlation peak value according to the temperature data to obtain the liquid shortage evolution characteristics that characterize the change of the liquid shortage state with the detection cycle. The processor is also used to input the liquid shortage evolution characteristics into a pre-trained liquid replenishment prediction model to obtain the predicted liquid replenishment amount of the battery under test; the liquid replenishment prediction model is established based on spread spectrum time domain reflection detection data under different liquid level states and the corresponding actual liquid replenishment amount.

[0020] Optionally, the data acquisition module includes a vibration acquisition module and a temperature acquisition module, and the spread spectrum detection module includes a signal generation module, a signal acquisition module, and a characteristic impedance of 50Ω. Coaxial cable, T-terminal and impedance of 50 The compensation terminal; The vibration acquisition module is used to acquire the acceleration signal during the operation of the target train and input the acceleration signal into the processor. The temperature acquisition module is used to acquire the temperature of the battery under test or the temperature of the environment around the battery under test and input the temperature into the processor. The signal generation module is used to generate two in-phase pseudo-random code signals modulated by binary phase shift keying. One pseudo-random code signal is directly input to the signal acquisition module as a reference incident signal, and the other pseudo-random code signal is used as a spread spectrum detection signal and injected into the battery under test through a T-type terminal and a coaxial cable. The signal acquisition module is used to synchronously acquire the reference incident signal and the total reflected signal, and input the reference incident signal and the total reflected signal into the processor. The compensation terminal is used to perform impedance matching compensation for the preset detection link.

[0021] Thirdly, the present invention provides a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described method.

[0022] Fourthly, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0023] The present invention has at least the following beneficial effects: This invention collects train vibration signals and battery temperature data, adaptively determines the pseudo-random code length for generating spread spectrum detection signals based on vibration intensity, thus matching detection efficiency with anti-vibration interference capability. It separates the total reflected signal according to the detection link parameters, reflection coefficient range, and successive attenuation relationship of adjacent reflections along the same path, reducing non-physical estimations caused by multiple reflection aliasing. Furthermore, it performs in-direction correction and weighted superposition of the first and subsequent reflected signals based on peak polarity, propagation delay, and reflection coefficient, enhancing the weak reflection characteristics caused by fluid shortage. Cross-correlation and temperature compensation are applied to the enhanced reflected signals to form fluid shortage evolution characteristics that change with the detection cycle. A model based on different fluid level detection data and actual fluid replenishment volume outputs a predicted replenishment volume, improving the stability of fluid shortage feature extraction under complex operating conditions and the accuracy of fluid replenishment prediction for the battery under test, thereby extending battery life and improving train safety. Attached Figure Description

[0024] The accompanying drawings are provided to further understand the technical solutions of the present invention and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the technical solutions of the present invention, and do not constitute a limitation on the technical solutions of the present invention.

[0025] Figure 1 This is a flowchart of a method for predicting the liquid replenishment amount of a train battery in one embodiment of this application; Figure 2 This is a schematic diagram of the structure of the train battery electrolyte replenishment prediction device in one embodiment of this application; Figure 3 This is a schematic diagram of the structure of a terminal device in one embodiment of this application. Detailed Implementation

[0026] The technical solution of the present invention will now be described in detail and completely with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0027] In the description of this invention, it should be noted that the terms "upper", "lower", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0028] It should be noted that the target train in this embodiment can be a high-speed train, an intercity train, a subway train, or other rail vehicles equipped with liquid electrolyte battery packs; the battery under test can be a single cell in the battery pack or a battery module with an independent detection port.

[0029] Example 1 This embodiment uses a high-speed train equipped with liquid electrolyte battery packs as the application object. The battery packs are installed in the battery box under the train. During train operation, the train is subjected to wheel-rail excitation, car body structural vibration, and equipment start-up and shutdown impacts. The temperature inside the battery box also varies with the season, the line, and the charging and discharging status. Maintenance personnel need to continuously identify the liquid loss evolution trend of target cells without disassembling the battery packs and obtain the recommended liquid replenishment volume when the train is brought into the depot for maintenance. Since early liquid loss only causes a slight change in port impedance, it is difficult to detect in time by conventional manual observation. At the same time, the connection ends and battery ports in the detection link will generate multiple reflections, and vibration and temperature changes will also cause fluctuations in the reflection peaks, making it difficult to stably extract the liquid loss evolution characteristics and accurately predict the liquid replenishment amount.

[0030] To address the above problems, this invention provides a method for predicting the electrolyte replenishment level of train batteries, such as... Figure 1 As shown, this method comprises steps 11 to 17.

[0031] Step 11: Collect the train vibration signal of the target train and the temperature data of the battery under test in the target train.

[0032] In one specific implementation, the liquid electrolyte battery inside the battery box of the EMU (Electric Multiple Unit) is used as the battery under test. Before the train is put into operation, a triaxial accelerometer is fixed to the inner wall of the battery box containing the battery under test or to a position rigidly connected to the mounting base of the battery under test, so that the accelerometer can collect the vibration generated by the battery as the train moves. A temperature sensor is attached to the outer surface of the battery under test near the electrolyte level area and fixed with thermally conductive adhesive to collect the surface temperature of the battery under test. When the installation space is limited and it is not possible to attach the temperature sensor to the battery under test, the temperature sensor can be set within a preset distance range around the battery under test to collect the ambient temperature around the battery under test. After the train starts running, acceleration signals and temperature data in three directions are collected synchronously according to a preset detection cycle; for example, data collection is performed every 10 minutes, with each collection lasting 10 seconds. The sampling frequency of the acceleration signal is set to 1000Hz, and the sampling frequency of the temperature data is set to 1Hz. To ensure the comparability of data from different testing cycles, each data acquisition recorded the train number, the battery number under test, the acquisition time, and the testing cycle number. A unified clock was used to time-stamp the acceleration signal and temperature data. After acquisition, data segments with accelerometer range overflow, temperature data exceeding the preset effective range, or incomplete acquisition times were discarded. For valid data segments, acceleration sampling sequences in three directions were saved, and the average temperature data within the same testing cycle was used as the corresponding temperature data for that testing cycle.

[0033] When the surface temperature of the battery under test is directly collected, the average value represents the temperature of the battery under test; when the ambient temperature is collected, the average value represents the thermal environment in which the battery under test is located.

[0034] Step 12: Determine the vibration intensity based on the train vibration signal, and determine the pseudo-random code length based on the vibration intensity.

[0035] It should be noted that, in this embodiment of the invention, the pseudo-random code is used to generate the spread spectrum detection signal, and the length of the pseudo-random code is positively correlated with the vibration intensity.

[0036] In one specific implementation, the triaxial acceleration signal obtained in step 11 is first subjected to DC removal processing. Let a detection window contain... The sampling point, the first The accelerations of the sampling points in the three directions are as follows: , , The average accelerations in the three directions are respectively , , Then the first The synthetic vibration acceleration corresponding to each sampling point It can be represented as .

[0037] Subsequently, the vibration intensity was determined based on the root mean square value of each synthesized vibration acceleration within the detection window. Specifically, vibration intensity The calculation expression is as follows .

[0038] For example, in step 11, when the sampling frequency of the acceleration signal is 1000Hz and the single continuous acquisition time is 10s, one detection window contains 10,000 sampling points, that is... Before calculating the vibration intensity, sampling points exceeding the range of the accelerometer can be discarded, and median filtering can be used to remove isolated impact points whose duration does not exceed the preset pulse width, in order to avoid the acquisition anomalies directly affecting the selection of the pseudo-random code length. Vibration components caused by turnout passage, instantaneous wheel-rail impact, etc., whose duration exceeds the preset pulse width are retained, so that the determined vibration intensity can reflect the actual train running vibration experienced by the battery under test during the detection period.

[0039] The first vibration threshold is pre-calibrated based on vibration data collected during the operation of the target train on the line, the installation location of the battery box under test, and the vibration transmission characteristics of the box. Second vibration threshold ,and Specifically, vibration intensity can be collected under conditions of constant train speed, passing through curves or turnouts, and significant wheel-rail impact. The upper limit of the vibration intensity range corresponding to low vibration conditions is determined as the first vibration threshold. The upper limit of the vibration intensity range corresponding to moderate vibration conditions is defined as the second vibration threshold. .

[0040] In one specific implementation, after determining the first vibration threshold... Second vibration threshold Then, the length of the pseudo-random code to be used in the current detection period is determined from pseudo-random codes of lengths 7, 31, and 63. , .

[0041] When the vibration intensity does not exceed the first vibration threshold, a pseudo-random code with a length of 7 is selected to reduce the amount of data to be processed in one detection cycle; when the vibration intensity is greater than the first vibration threshold but not greater than the second vibration threshold, a pseudo-random code with a length of 31 is selected to balance detection efficiency and anti-interference capability; when the vibration intensity is greater than the second vibration threshold, a pseudo-random code with a length of 63 is selected to improve the processing gain of the spread spectrum detection signal after cross-correlation processing.

[0042] To avoid frequent switching of the pseudo-random code length between adjacent detection periods due to vibration intensity fluctuations around the threshold, when the difference between the current vibration intensity and the first or second vibration threshold does not exceed the corresponding threshold... During this period, the pseudo-random code length used in the previous detection cycle is maintained. The determined pseudo-random code length serves as the control parameter for generating the spread spectrum detection signal in the current detection cycle. Therefore, increasing the pseudo-random code length as the train vibration intensity increases can improve the correlation processing gain of the reflected signal under strong vibration conditions, reducing cross-correlation peak jitter, peak position drift, and false peak interference caused by vibration; using a shorter pseudo-random code when the vibration is weak can shorten the detection and processing time.

[0043] Step 13: Generate a spread spectrum detection signal corresponding to the length of the pseudo-random code, obtain the reference incident signal corresponding to the spread spectrum detection signal, inject the spread spectrum detection signal into the battery under test through a preset detection link, and simultaneously collect the total reflected signal.

[0044] It should be noted that the total reflected signal is formed by the reflection of the spread spectrum detection signal at the impedance discontinuity between the preset detection link and the battery under test.

[0045] In one specific implementation, based on the pseudo-random code length determined in step 12 Generate binary pseudo-random code sequence ,in, And through calculation formula The binary pseudo-random code sequence Convert to bipolar code sequence , The value is or .

[0046] Subsequently, the bipolar code sequence was modulated using binary phase-shift keying to obtain the spread spectrum detection signal. .

[0047] In one feasible implementation method, ,in, Indicates the amplitude of the spread spectrum detection signal; Indicates the carrier frequency; Indicates the initial phase; , This represents the duration of a single symbol. The duration of a complete code period is... For example, when the pseudo-random code length determined in step 12 is 31, 31 symbols are output sequentially according to the same symbol duration; when the pseudo-random code length is increased to 63, one code period contains 63 symbols to obtain higher correlation processing gain. During the detection process, the amplitude of the spread spectrum detection signal is set to a low amplitude that will not affect the normal operating state of the battery under test and can be effectively identified by the subsequent acquisition circuit, and the symbol rate, carrier frequency and output amplitude are kept consistent in each detection period to make the reflected signals obtained in different detection periods comparable.

[0048] When generating the spread spectrum detection signal, two signals with identical code patterns, carrier frequencies, initial phases, and output times are simultaneously generated. One of these signals bypasses the battery under test and is directly sent to the first acquisition channel, serving as the reference incident signal. The other signal serves as the spread spectrum detection signal, injected into the battery under test via a preset detection link. The reference incident signal and the spread spectrum detection signal are controlled by the same clock source and the same trigger signal, thus having the same pseudo-random code length and symbol boundaries. The reference incident signal is used to characterize the actual output detection signal of the current detection cycle, reducing the impact of output amplitude fluctuations, clock deviations, and waveform distortion on subsequent cross-correlation processing.

[0049] The preset detection link can include, in sequence, a detection signal output terminal, a T-type terminal, and a characteristic impedance of 50Ω. The coaxial cable, the test connection terminal of the battery under test, and the impedance of 50Ω The compensation terminal is used to reduce impedance mismatch between the detection signal output and the transmission link. The spread spectrum detection signal enters the coaxial cable via the T-terminal and is transmitted to the positive and negative detection terminals of the battery under test. Before testing, the connection status between the coaxial cable and the detection terminals of the battery under test is checked to ensure a secure connection. The coaxial cable length, signal propagation speed, sampling frequency, and the position of each connection node in the transmission path are recorded for subsequent steps to determine the propagation delay of each reflected component.

[0050] When the spread spectrum detection signal propagates along the preset detection link, when it passes through impedance discontinuities such as the T-terminal connection, the coaxial cable connection, and the detection terminal of the battery under test, part of the signal continues to propagate forward, while the other part returns along the original transmission path. The equivalent impedance of the battery under test port is affected by the electrolyte level and the battery state. When this equivalent impedance is inconsistent with the characteristic impedance of the coaxial cable, reflection occurs at the detection terminal of the battery under test. The returned reflected signal may also be reflected again at the aforementioned impedance discontinuities, thus forming a first reflection and at least one subsequent reflection.

[0051] Therefore, without considering acquisition noise, the total reflected signal It can be represented as .in, This indicates the number of reflection components to be separated in the total reflected signal, and ; Indicates the first The reflection coefficient corresponding to each reflection component ; Indicates the first The propagation delay of each reflected component relative to the reference incident signal.

[0052] After considering the noise from the acquisition and interference from train operation, the total reflected signal It can be represented as . This indicates noise and interference from the operating environment. The reflection components in the total reflected signal may alias in the time domain due to their similar propagation delays. Therefore, this step only synchronously collects the unseparated total reflected signal, and then performs constraint separation based on the preset transmission path structure and propagation delay of the detection link.

[0053] During data acquisition, the trigger moment when the spread spectrum detection signal begins to output is used as the common start moment for both the reference incident signal acquisition channel and the total reflection signal acquisition channel. Data from both channels are acquired synchronously at the same sampling frequency. The single acquisition time must at least cover the maximum round-trip propagation time required for the spread spectrum detection signal to travel from the output end to the battery under test and back to the acquisition end, and further cover a predetermined number of subsequent reflections. To reduce the impact of accidental noise, multiple identical code periods can be continuously transmitted within a detection cycle, and the reference incident signal and total reflection signal corresponding to each code period are acquired separately. Code periods with sampling saturation, trigger loss, or insufficient sampling points are discarded. The remaining code periods are averaged after being aligned according to the trigger moment to obtain the reference incident signal and total reflection signal corresponding to the current detection cycle. Thus, the reference incident signal and the total reflection signal maintain temporal correspondence and retain the reflection characteristics generated by the impedance changes at the battery under test port, providing a data foundation for determining the propagation delay of subsequent reflection components, constraint separation, and reflection enhancement. Step 14: Obtain the link parameters of the preset detection link, determine the propagation delay of each reflection component in the total reflection signal based on the link parameters, and constrain and separate the total reflection signal based on the propagation delay, the preset reflection coefficient value range, and the successive attenuation relationship between two adjacent reflections formed by the same reflection propagation path in the preset detection link, to obtain the first reflection signal and at least one subsequent reflection signal.

[0054] In one specific implementation, the transmission path length, sampling frequency, and transmission path structure of the preset detection link are first read. The transmission path structure at least characterizes the path relationship of the spread spectrum detection signal propagating from the signal injection position through the T-terminal, coaxial cable, and the detection terminal of the battery under test, and being reflected by each impedance discontinuity to the signal acquisition position. The transmission path length includes the length of each coaxial cable segment and the round-trip propagation distance of the reflected signal determined by the corresponding connection relationships. Let the signal propagation speed within the detection link be... , No. The total distance that each reflection component propagates along its corresponding reflection propagation path is Then the first Each reflection component propagates along its corresponding propagation time delay. It can be represented as ,in, This represents the total propagation distance from the signal injection location to the corresponding impedance discontinuity location and back to the signal acquisition location; signal propagation speed. The sampling frequency can be determined based on the nominal propagation speed or velocity factor of the coaxial cable, or it can be calibrated using a calibration cable of known length before testing. Let the sampling frequency be... Then the propagation delay Corresponding discrete time delay sampling points It can be represented as ,in, This indicates taking the nearest integer.

[0055] Based on the actual connection relationship of the preset detection link, the possible reflection propagation paths within the total reflection signal acquisition window are enumerated. Only candidate propagation delays formed by the actual propagation relationship between the T-terminal, coaxial cable connection position, and the detection end of the battery under test are retained, thus eliminating reflection components that are impossible to occur in the link structure. For multiple reflections formed by the same reflection propagation path, each reflection component is numbered from the first to the second in ascending order of propagation delay. There are 1 reflection component, and they satisfy: Not less than 2 .

[0056] After determining each propagation delay, the reference incident signal acquired in step 13 is processed according to each propagation delay. Perform time shifting and construct a latency test matrix. , .

[0057] In the process of discretization, the matrix The The column can be shifted backward from the sampling sequence of the reference incident signal. The data is obtained from sampling points, and zeros are added to positions that exceed the original sampling range after movement. The data length of each column is then adjusted to match the total reflected signal. The data lengths are consistent. Therefore, the matrix Each column represents a candidate waveform of the reference incident signal when it reaches the acquisition position according to different reflection propagation paths.

[0058] Let the reflection coefficients corresponding to each reflection component form a reflection coefficient vector. , The total reflected signal can then be fitted by the product of the time delay test matrix and the reflection coefficient vector. To obtain a separation result that conforms to the reflection law of the detection link when the reflection components are mixed, the total reflected signal is used as the basis for this process. With fitted reflection signal With the objective of minimizing the difference between them, the following optimization model is solved:

[0059] And satisfy: , .

[0060] in, This represents the attenuation coefficient between two adjacent reflections formed by the same reflection propagation path. This represents the first regularization coefficient corresponding to the smoothing constraint of adjacent reflection coefficients; This represents the second regularization coefficient corresponding to the successive decay constraint; This represents the first-order difference matrix used to determine the difference between adjacent reflection coefficients, for example, when the number of reflection components to be separated is... hour, It can be represented as .

[0061] therefore, This represents the sequence of differences between adjacent reflection coefficients. The first term limits the fitting error between the superposition of the separated reflection components and the actual acquired total reflection signal; the second term is a smoothing constraint for adjacent reflection coefficients, used to suppress drastic jumps in adjacent reflection coefficients caused by noise; the third term is a successive attenuation constraint. When the amplitude of the reflection coefficient of the subsequent reflection satisfies... When the reflection coefficient amplitude of the subsequent reflection is greater than the product of the reflection coefficient amplitude of the previous reflection and the attenuation coefficient, this term applies a penalty, thereby suppressing non-physical separation results where the intensity of the subsequent reflection is abnormally greater than that of the previous reflection. The range of reflection coefficient values ​​is limited to [value missing]. The range of reflection coefficient values ​​corresponds to the location of the transmission line impedance discontinuity.

[0062] In the specific solution, the initial value of the reflection coefficient can be obtained first using the least squares method, and then the value exceeding the initial value can be calculated. The initial value of the range is truncated to the corresponding boundary; then, an iterative optimization method with boundary constraints is used to update the reflection coefficient vector. First regularization coefficient. Second regularization coefficient and attenuation coefficient The method can utilize calibration data under normal and known liquid level conditions to minimize the reconstruction residual of the calibration data and ensure that adjacent reflection components formed along the same path satisfy a successive attenuation relationship. During the iteration process, when the change in the objective function between two adjacent iterations is lower than a preset convergence threshold, or when the preset maximum number of iterations is reached, the current reflection coefficient vector is output as the separation result. If, after solving, a certain reflection coefficient reaches its value boundary and the reconstruction residual of the total reflection signal is still higher than the preset residual threshold, or if adjacent reflection coefficients clearly do not satisfy the successive attenuation relationship, the corresponding candidate propagation delay is re-examined, abnormal reflection components are eliminated, or the regularization coefficient is adjusted before solving again.

[0063] After obtaining the reflection coefficients corresponding to each reflection component, the reference incident signal is time-shifted and its amplitude adjusted according to the corresponding propagation delay and reflection coefficient to reconstruct the first reflection component. The reflected signal corresponding to each reflection component , .

[0064] The reflected signal corresponding to the first reflection component with the smallest propagation delay This was identified as the first reflected signal. The signal is identified as the subsequent reflected signal. By pre-defining the candidate propagation delay based on the transmission path structure, and simultaneously applying constraints such as reflection coefficient boundaries, smoothing of adjacent reflection coefficients, and successive attenuation of multiple reflections along the same path, non-physical separation results caused by noise and reflection waveform aliasing can be reduced. This ensures that the separated first and subsequent reflected signals conform to the reflection law of the preset detection link in terms of propagation delay and amplitude changes, providing a reliable data foundation for subsequent steps to perform in-direction correction and weighted superposition of multiple reflected signals.

[0065] Step 15: Based on the peak polarity, propagation delay, and reflection coefficient of each subsequent reflected signal relative to the first reflected signal, perform in-direction correction and weighted superposition on the first reflected signal to obtain the enhanced reflected signal.

[0066] Specifically, step 15 includes steps 15.1 to 15.3.

[0067] Step 15.1: Determine the signal-to-noise ratio (SNR) of each subsequent reflected signal, and determine the subsequent reflected signals with an SNR not lower than the preset compensation threshold as valid subsequent reflected signals.

[0068] Specifically, the reconstructed residual signal of the total reflected signal is calculated based on the separation results of step 14. , .

[0069] With the first Propagation delay of subsequent reflected signals Set the reflection analysis window to the center. The width of the reflection analysis window is not less than the duration of one pseudo-random code symbol, and the window covers the first... The main peak region of the subsequent reflected signal. Signal-to-noise ratio of the subsequent reflected signal Determined according to the following formula:

[0070] in, A positive number is used to prevent the denominator from being zero, for example Ensure the signal-to-noise ratio is not lower than the preset compensation threshold. The subsequent reflected signals are determined as valid subsequent reflected signals, and the indices of each valid subsequent reflected signal are grouped into a set. , In this embodiment, a preset compensation threshold is used. A threshold of 6 dB is selected. This threshold can also be predetermined based on calibration data under normal liquid level and different low-liquid conditions, so that the retained subsequent reflection signal can stably characterize the impedance change of the battery under test. This processing avoids adding the subsequent reflection component, which is mainly caused by acquisition noise or separation error, to the first reflection signal.

[0071] Step 15.2: Determine the peak polarity in the same direction correction coefficient based on the peak polarity of each effective subsequent reflected signal relative to the first reflected signal, and determine the compensation weight based on the propagation delay and reflection coefficient of each effective subsequent reflected signal.

[0072] For the One reflected signal In terms of propagation delay Centered reflection analysis window Within, determine the sampling time with the largest absolute value. , .

[0073] Determine the first [sample] based on the signal value at that sampling time. Peak polarity of each reflected signal , .

[0074] According to the The peak polarity of each effective subsequent reflected signal relative to the first reflected signal is used to determine the peak polarity in-direction correction coefficient. When the first When the peak polarity of each subsequent reflected signal is the same as that of the first reflected signal, When the first When the peak polarity of a subsequent reflected signal is opposite to that of the first reflected signal, The expression is .

[0075] In other words, when the first If the peak polarity of a valid subsequent reflected signal is the same as that of the first reflected signal, the polarity of the subsequent reflected signal remains unchanged; if the peak polarities of the two signals are opposite, the subsequent reflected signal is multiplied by [the value of the first reflected signal]. This ensures that the direction of the peak change is consistent with the first reflected signal. This processing avoids the subsequent reflected signals canceling each other out during the superposition process due to different impedance change directions at different reflection positions.

[0076] Furthermore, the compensation weights are determined based on the propagation delay and reflection coefficient of each effective subsequent reflected signal. The formula for calculating the compensation weight is:

[0077] in, Indicates the first The compensation weights corresponding to each subsequent reflected signal. Indicates the first The reflection coefficient of an effective subsequent reflected signal. Indicates the first The propagation delay of an effective subsequent reflected signal. Represents the time decay constant. In this embodiment, the time decay constant is determined based on the propagation delay difference between adjacent reflection components in the same reflection propagation path. , .

[0078] The exponential term in the compensation weighting is used to give greater weight to subsequent reflected signals whose propagation delay is closer to that of the first reflected signal. The absolute value of the reflection coefficient... This is used to assign greater weights to subsequent reflection signals with more pronounced reflection characteristics. After normalizing the weights of each valid subsequent reflection signal, the following condition is met: .

[0079] Step 15.3: Based on the peak polarity correction coefficient and compensation weight, the effective subsequent reflection signals are weighted and superimposed with the first reflection signal to obtain the enhanced reflection signal.

[0080] The expression for enhancing the reflected signal is:

[0081] in, This indicates an enhancement of the reflected signal. Indicates the first The peak polarity correction coefficient corresponding to the i-th subsequent reflected signal, when the i-th... When the peak polarity of each subsequent reflected signal is the same as that of the first reflected signal, When the first When the peak polarity of a subsequent reflected signal is opposite to that of the first reflected signal, . Indicates the first A subsequent reflected signal.

[0082] When set If the value is empty, meaning the signal-to-noise ratio of all subsequent reflected signals is lower than the preset compensation threshold, no subsequent reflection compensation will be performed, and the first reflected signal will be directly determined as the enhanced reflection signal. .

[0083] Through the above processing, the signal-to-noise ratio is first used to filter out subsequent reflection signals with insufficient reliability, and then the peak polarity in-direction correction is used to eliminate the cancellation between reflection signals of different polarities. The propagation delay and reflection coefficient are combined to normalize and weight the effective subsequent reflection signals. This can introduce subsequent reflection information that conforms to the reflection law of the transmission path while retaining the main characteristics of the first reflection signal, enhance the weak impedance reflection characteristics caused by the change of the liquid level of the battery under test, and improve the stability of subsequent cross-correlation peak extraction.

[0084] Step 16: Perform cross-correlation processing on the enhanced reflection signal and the reference incident signal, extract the cross-correlation peak value within the preset time delay window, and perform temperature compensation on the cross-correlation peak value according to the temperature data to obtain the liquid shortage evolution characteristics that characterize the change of the liquid shortage state with the detection cycle.

[0085] In one specific implementation, let the current processing detection cycle be the [number]th [number]. For each detection cycle, the enhanced reflection signal obtained in step 15 is denoted as... Before performing cross-correlation processing, the data is processed according to the common triggering time. and reference incident signal The sampling sequences are aligned, and the DC component of each sampling sequence is removed. If multiple complete pseudo-random code cycles are continuously acquired within the current detection period, the sequence is segmented according to the start time of each code cycle, and the cross-correlation results of each segment are averaged to reduce the impact of noise from a single acquisition.

[0086] The propagation delay of the first reflected signal determined in step 14 Propagation delay of the last subsequent reflected signal and the symbol duration used in step 13 Set the preset delay window Preset delay window This is used to cover the first and subsequent reflection components formed by the battery under test and its corresponding reflection propagation path, while excluding correlation peaks unrelated to the battery under test generated at the connection point of the detection link. In practical applications, the boundary of the preset delay window can also be determined based on the sampling frequency. The data is converted into corresponding discrete sampling points. To ensure consistent peak extraction rules across different detection cycles, the polarity of the total reflection signal acquisition channel is determined during initial calibration, ensuring that the main cross-correlation peak of the battery under test is positive under normal liquid level conditions.

[0087] In one feasible implementation, through calculation formula

[0088] Get the first Each detection cycle has a time delay Cross-correlation value at the location .

[0089] In the discrete processing, it is assumed that both the enhanced reflected signal and the reference incident signal contain Each sampling point, time delay The corresponding number of time-shifted sampling points is The cross-correlation value can be calculated using the following formula:

[0090] in, This indicates the sampling point number within the current detection cycle. The value range is determined by the preset delay window. Confirmed. After calculating the cross-correlation values ​​for multiple complete pseudo-random code cycles within a detection period, the values ​​are averaged at the same time delay position to obtain the cross-correlation sequence corresponding to the current detection period. Through cross-correlation processing, the reflection components in the enhanced reflection signal that are consistent with the code pattern of the reference incident signal can be coherently accumulated, while train running noise and random interference unrelated to the pseudo-random code are unlikely to form stable correlation peaks, thereby improving the identifiability of the weak impedance change characteristics of the battery under test.

[0091] In one feasible implementation, through calculation formula

[0092] Get the first The cross-correlation peak value corresponding to each detection cycle .

[0093] Simultaneously record the peak delay corresponding to the peak cross-correlation value:

[0094] in, This is used to check whether the extracted cross-correlation peak is still located at the preset reflection position corresponding to the battery under test. If The minimum difference between the candidate propagation delay determined in step 14 and the delay is greater than one symbol duration. If the detection period is not detected, it will be marked as an abnormal detection period. The cross-correlation peak value will not be used to update the liquid deficiency evolution characteristics, so as to avoid false peaks formed by strong transient interference from entering subsequent processing.

[0095] Since the equivalent impedance of the battery port and the electrical characteristics of the electrolyte both change with temperature, cross-correlation peak values ​​at different temperatures are collected beforehand when the battery is at its normal electrolyte level to establish the temperature influence function. For example, the temperature of the battery under test was stabilized at 5℃, 15℃, 25℃, 35℃, and 45℃, and at least ten testing cycles were performed continuously at each temperature. After eliminating abnormal testing cycles, the average value of the corresponding cross-correlation peaks was taken as the normal liquid level calibration peak at that temperature. A quadratic polynomial was used to fit the relationship between temperature and the normal liquid level calibration peak:

[0096] in, Indicates temperature as The peak value of the fitted cross-correlation corresponding to the normal liquid level state. The fitting coefficients were determined based on the calibration data. 25℃ was selected as the reference temperature. The change in the normal liquid level peak caused by temperature relative to the reference temperature is defined as the temperature effect function: .

[0097] Since step 12 can select different pseudo-random code lengths based on vibration intensity... In this embodiment, a corresponding temperature influence function is established for each selectable pseudo-random code length; in the... Within each detection period, the current pseudo-random code length is called. The corresponding temperature effect function, and the functions called are uniformly denoted as... This avoids the difference in correlation processing gain under different pseudo-random code lengths directly affecting the temperature calibration results.

[0098] In one feasible implementation, through calculation formula

[0099] Get the first The temperature-compensated cross-correlation peak value corresponding to each detection cycle ; This indicates the effect of temperature on temperature data. The function value at that point. When hour, The temperature-compensated cross-correlation peak value is equal to the actual extracted cross-correlation peak value. When the temperature deviates from the reference temperature, the peak offset caused by temperature changes under normal liquid level conditions is subtracted from the cross-correlation peak value. If the detected temperature exceeds the calibration temperature range of the temperature influence function, the function value corresponding to the closest calibration temperature boundary is used for compensation instead of interval extrapolation, thus avoiding overcompensation caused by extrapolation of the fitted function. Through this temperature compensation, the characteristic coupling caused by the combined effect of temperature change and liquid level change on the equivalent impedance of the battery port can be weakened, so that the remaining peak changes mainly characterize the long-term changes in electrolyte level.

[0100] Arrange the temperature-compensated cross-correlation peaks in chronological order of the detection cycles to obtain the temperature-compensated peak sequence. , ,in, This indicates the number of effective detection cycles currently used to extract features of fluid deficiency evolution. To ensure comparability of each detection cycle over time, effective detection cycles obtained at the same detection interval are selected; for data segments with multiple consecutive missing detection cycles, principal component peak curve reconstruction is not performed across these data segments.

[0101] An integrated empirical mode decomposition (EMD) is performed on the temperature-compensated peak sequence. Specifically, the standard deviation of the temperature-compensated peak sequence is calculated. The standard deviation of the added white noise Set as .in, The value range is from 0.05 to 0.3. In this embodiment, we take... Number of sets Take 100 times. For the ... This set decomposition generates sets with a mean of zero and a standard deviation of... White noise sequence And construct the sequence to be decomposed. , , .

[0102] For each Perform empirical mode decomposition. During each decomposition, construct upper and lower envelopes based on the local maxima and minima of the sequence, respectively. Calculate the mean of the two envelopes and subtract it from the original sequence. Repeat this screening process until the difference between the number of zeros and the number of extreme points of the resulting component does not exceed 1, and the mean of the local envelope approaches zero. This component is then identified as an intrinsic mode function component. Continue decomposing the remaining sequence until the remaining terms are monotonic sequences or contain only one extreme point.

[0103] Let the first The decomposition yields the th... The intrinsic mode function components are: Then, averaging the eigenmode function components of the same order in 100 set decompositions yields: ,in, , This represents the number of intrinsic mode function components obtained from integrated empirical mode decomposition. Ensemble averaging can cancel out positive and negative random white noise added to different sets and reduce mode aliasing by utilizing the uniform distribution of white noise on the frequency scale.

[0104] In one screening method, a calculation formula is used.

[0105] The correlation coefficients between each intrinsic mode function component and the temperature-compensated peak sequence were obtained. ,in, Indicates the first The average value of each intrinsic mode function component. This represents the average value of the temperature-compensated peak sequence. Intrinsic mode function components with a correlation coefficient not less than 0.3 are selected, and their indices are grouped into a set. , .

[0106] In another screening method, the intrinsic mode function components are calculated in order from low frequency to high frequency. Energy of each eigenmode function component , .

[0107] Low-frequency intrinsic mode function components are selected sequentially until the cumulative energy contribution rate of the selected components satisfies: .

[0108] This implementation method preferentially adopts a screening method with a correlation coefficient of not less than 0.3; when there are fewer than one intrinsic mode function component satisfying this condition, a screening method using low-frequency intrinsic mode function components with a cumulative energy contribution rate of not less than 80% is adopted. The selected intrinsic mode function components are reconstructed to obtain the first... The principal component peak value corresponding to each detection cycle , .

[0109] The principal component peak curve is composed of the principal component peak values ​​corresponding to each detection period. , .

[0110] Principal component peak curve The electrolyte loss evolution characteristics of the battery under test were identified. Cross-correlation processing improved the processing gain of the weak reflection characteristics consistent with the spread spectrum detection signal pattern. Temperature compensation weakened the cross-correlation peak drift caused by temperature changes. Integrated empirical mode decomposition and intrinsic mode function screening further suppressed high-frequency random fluctuations caused by train vibration, instantaneous terminal disturbances, and acquisition noise, while retaining the low-frequency change trend corresponding to the gradual loss of electrolyte. Therefore, the obtained principal component peak curve can stably characterize the evolution process of the electrolyte loss state of the battery under test with the detection cycle, and provide input data for subsequent steps to calculate the peak ratio, peak drop, short-term slope, and predict the replenishment amount.

[0111] Step 17: Input the liquid shortage evolution characteristics into the pre-trained liquid replenishment prediction model to obtain the predicted liquid replenishment amount of the battery under test.

[0112] In this embodiment of the invention, the liquid replenishment prediction model is established based on spread spectrum time-domain reflectometry data under different liquid level conditions and the corresponding actual liquid replenishment volume.

[0113] In one specific implementation, a multinomial regression model is used as the pre-trained electrolyte replenishment prediction model. Before model training, calibration samples are established for batteries of the same model as those configured on the target train. Each battery is adjusted to its normal electrolyte level and several levels below the normal level, for example, by setting the electrolyte deficit relative to the normal level as 0mL, 50mL, 100mL, 150mL, and 200mL respectively. After each level stabilizes, spread spectrum time-domain reflectometry is performed according to steps 11 to 16 to obtain the corresponding principal component peak curves. After the detection, a specified replenishment liquid is gradually added to the battery until the electrolyte level is restored to the predetermined normal level. The actual volume of liquid added is used as the actual replenishment volume label for that sample. When obtaining the actual replenishment volume, a calibrated graduated cylinder, pipette, or mass measuring device is used to record the liquid volume or mass before and after replenishment, and the volume is calculated based on the density of the replenished liquid. For each liquid level state, the test is repeated under different temperatures and vibration conditions to ensure that the training sample covers the possible operating environment of the target train.

[0114] Let the current processing detection cycle be the nth. For each detection cycle, step 16 obtains the peak value of the principal component corresponding to that detection cycle. The baseline principal component peak value of the battery under test when it is in normal liquid level condition is: .in, Calibration is performed on the battery under test before online testing: the battery is brought to its normal liquid level, and multiple test cycles are continuously executed under reference temperature and low vibration conditions. The peak values ​​of the principal components for each test cycle are obtained, and the average value after removing outliers is determined as the baseline. When the correlation processing gain differs for different pseudo-random code lengths, the corresponding baseline principal component peak value is saved for each pseudo-random code length, and the value is called during prediction based on the current pseudo-random code length. corresponding .

[0115] Calculate the first based on the principal component peak curve. Peak proportion corresponding to each detection cycle , .

[0116] Calculate the first based on the principal component peak curve. Peak decrease corresponding to each detection cycle , .

[0117] The peak ratio reflects the relative change of the current principal component peak value compared to the normal liquid level reference value, while the peak decrease reflects the absolute change of the current principal component peak value compared to the normal liquid level reference value. To avoid an abnormal peak ratio caused by the reference principal component peak value approaching zero, checks are performed during the calibration phase. The absolute value; when When the level is below the preset effective threshold, the detection link connection status is rechecked and normal liquid level calibration is re-executed instead of using the baseline value for prediction.

[0118] Furthermore, the first principal component peak curve is calculated... Short-time slope corresponding to each detection cycle , ,in, Indicates the first The short-time slope corresponding to each detection cycle; This indicates the number of detection cycles used to determine the short-term slope; Indicates the peak curve of the principal component at the th The peak value of the principal component corresponding to each detection cycle. In this embodiment, the peak value of the principal component is taken as follows: When the interval between adjacent detection cycles is 10 minutes, the short-time slope characterizes the average trend of the principal component peak values ​​within the most recent 60 minutes. When the number of currently valid detection cycles is less than... If the short-time slope is not calculated at this time, the most recent valid short-time slope is used. If no valid short-time slope exists, it is set to zero and a data validity flag is added to the input. If the time intervals between detection cycles are not equal, the principal component peak curves are resampled according to a uniform time interval, or the denominator is changed. Replace with the first The first detection cycle and the first The actual time difference between each testing cycle.

[0119] To eliminate the influence caused by the different units and value ranges of the input parameters, the first parameter is calculated based on the training samples. The average value of the training set of each input parameter and standard deviation And standardize the input parameters: ,in. The first in the input vector One input parameter, Represents the standardized first One parameter, A positive number is used to prevent the denominator from being zero. After the model training is complete, the corresponding values ​​for each input parameter are... and The data is saved along with the model parameters, and the same standardized parameters are used to process the input data during online prediction.

[0120] In this embodiment, a second-order polynomial regression relationship is used to represent the relationship between the standardized input parameters and the actual fluid replacement volume. Let the standardized input vector be... Its dimension is Then the model output can be expressed as:

[0121] in, This indicates the amount of fluid replenishment output by the model. Represents the coefficient of the constant term. Denotes the coefficient of the linear term. Represents the coefficient of the squared term. This represents the coefficient of the interaction term between different input parameters.

[0122] The standardized input parameters corresponding to each training sample are expanded into second-order polynomial features and used to form a training matrix. The actual fluid rehydration volume corresponding to each training sample is used to form a label vector. The model coefficient vector is determined by the following formula. :

[0123] in, This represents the regularization coefficient used to suppress excessively large model coefficients. This represents the identity matrix. The calibration samples can be divided into training, validation, and test sets, and the regularization coefficient can be selected based on the mean absolute error on the validation set. Data from the same battery in adjacent testing cycles are grouped into the same dataset to avoid highly correlated data from the same battery entering both the training and testing sets simultaneously.

[0124] The mean and standard deviation of each input parameter are calculated based on the training set, and the training, validation, and test sets are standardized using the same parameters. A fluid replenishment prediction model is fitted using the training set, and the model order, regularization coefficient, or other model parameters are determined using the validation set. The final model is selected based on minimizing the fluid replenishment prediction error. The generalization performance of the model is then tested using the test set. Once the test results meet the preset accuracy requirements, the model parameters, standardized input parameters, applicable battery model range, and maximum allowable fluid replenishment volume are saved to form a pre-trained fluid replenishment prediction model for use during online detection.

[0125] After model training is complete, the mean absolute error, root mean square error, and coefficient of determination between the predicted and actual fluid replacement volumes are calculated using test samples not involved in model coefficient determination. When the test results meet the preset accuracy requirements, the model coefficients are... The mean and standard deviation of each input parameter, the applicable range of the battery model, and the maximum allowable replenishment volume are saved as a pre-trained replenishment volume prediction model. For different battery models, a corresponding replenishment volume prediction model can be trained separately; alternatively, if the training samples cover multiple battery models, the battery model parameters can be used as input to train a common replenishment volume prediction model.

[0126] When making online predictions, according to the first The principal component peak value corresponding to each detection cycle calculate , as well as and combined with temperature data Vibration intensity The input vector consists of battery model parameters and historical fluid replenishment records. The input vector is standardized using the mean and standard deviation saved during the training phase. The standardized input parameters are then substituted into a pre-trained multinomial regression model to obtain the result of the test battery on the [number of]th [phase]. Initial predicted fluid replenishment volume for each detection cycle .

[0127] Considering that the replenishment volume should not be negative and should not exceed the maximum volume of liquid allowed to be added to restore the battery under test from its current state to its normal liquid level, the replenishment volume should be determined accordingly. Boundary processing is performed on the initial predicted replenishment volume to obtain the final predicted replenishment volume. , .in, The required electrolyte level is determined in advance based on the battery model, normal electrolyte level requirements, and allowable replenishment range. To reduce the impact of abnormal fluctuations in a single testing cycle on maintenance decisions, the median of the prediction results from the three most recent consecutive valid testing cycles can be taken as the current predicted electrolyte replenishment amount. When the predicted electrolyte replenishment amount continuously increases within consecutive testing cycles, it indicates that the electrolyte deficiency of the battery under test may continue to worsen.

[0128] After the prediction is completed, the battery identification, detection time, principal component peak curve, peak ratio, peak drop, short-term slope, temperature data, vibration intensity, and predicted replenishment volume are associated and saved. Maintenance personnel verify and replenish the fluid according to the predicted replenishment volume, then write the actual added fluid volume into the historical replenishment record for subsequent recalibration of the baseline principal component peak or updating the replenishment volume prediction model. Thus, the replenishment volume prediction model can comprehensively utilize the current amplitude of the low-fluid evolution characteristics, the degree of change relative to the normal fluid level, recent trends, and temperature, vibration, battery model, and historical maintenance information to convert the spread spectrum time-domain reflectometry detection results into a predicted replenishment volume usable by maintenance personnel.

[0129] Example 2 like Figure 2 As shown, the present invention also provides a train battery fluid replenishment prediction device, which includes: The data acquisition module 201 is used to acquire the train vibration signal of the target train and the temperature data of the battery under test in the target train, and input the train vibration signal and temperature data into the processor 202. The processor 202 is used to determine the vibration intensity based on the train vibration signal, determine the pseudo-random code length based on the vibration intensity, and send the pseudo-random code length to the spread spectrum detection module 203; the pseudo-random code is used to generate the spread spectrum detection signal, and the pseudo-random code length is positively correlated with the vibration intensity. The spread spectrum detection module 203 is used to generate a spread spectrum detection signal corresponding to the length of the pseudo-random code, acquire the reference incident signal corresponding to the spread spectrum detection signal, inject the spread spectrum detection signal into the battery under test through a preset detection link, synchronously acquire the total reflection signal, and input the reference incident signal and the total reflection signal into the processor 202; the total reflection signal is formed by the reflection of the spread spectrum detection signal at the impedance discontinuity between the preset detection link and the battery under test; The processor 202 is also used to acquire the link parameters of the preset detection link, determine the propagation delay of each reflection component in the total reflection signal according to the link parameters, and perform constraint separation on the total reflection signal based on the propagation delay, the preset reflection coefficient value range and the successive attenuation relationship between two adjacent reflections formed by the same reflection propagation path in the preset detection link, to obtain the first reflection signal and at least one subsequent reflection signal. The processor 202 is also used to perform in-direction correction and weighted superposition on the first reflection signal and each subsequent reflection signal according to the peak polarity, propagation delay and reflection coefficient of each subsequent reflection signal relative to the first reflection signal, so as to obtain an enhanced reflection signal; The processor 202 is also used to perform cross-correlation processing on the enhanced reflected signal and the reference incident signal, extract the cross-correlation peak value within the preset time delay window, and perform temperature compensation on the cross-correlation peak value according to the temperature data to obtain the liquid shortage evolution characteristics that characterize the change of the liquid shortage state with the detection cycle. The processor 202 is also used to input the liquid shortage evolution characteristics into a pre-trained liquid replenishment prediction model to obtain the predicted liquid replenishment amount of the battery under test; the liquid replenishment prediction model is established based on spread spectrum time domain reflection detection data under different liquid level states and the corresponding actual liquid replenishment amount.

[0130] The data acquisition module 201 includes a vibration acquisition module 2011 and a temperature acquisition module 2012, and the spread spectrum detection module 203 includes a signal generation module 2031, a signal acquisition module 2032, and a characteristic impedance of 50Ω. Coaxial cable 2033, T-terminal 2034, and impedance 50 Compensation terminal 2035; The vibration acquisition module 2011 is used to acquire the acceleration signal during the operation of the target train and input the acceleration signal into the processor 202. The temperature acquisition module 2012 is used to acquire the temperature of the battery under test or the temperature of the environment around the battery under test and input the temperature into the processor 202. The signal generation module 2031 is used to generate two in-phase pseudo-random code signals modulated by binary phase shift keying. One pseudo-random code signal is directly input to the signal acquisition module 2032 as a reference incident signal, and the other pseudo-random code signal is used as a spread spectrum detection signal and injected into the battery under test through the T-type terminal 2034 and the coaxial cable 2033. The signal acquisition module 2032 is used to synchronously acquire the reference incident signal and the total reflected signal, and input the reference incident signal and the total reflected signal into the processor 202. The compensation terminal 2035 is used to perform impedance matching compensation on the preset detection link.

[0131] It should be noted that the information interaction and execution process between the above-mentioned devices / units are based on the same concept as the method embodiments of this application. Their specific functions and technical effects can be found in the method embodiments section, and will not be repeated here. Those skilled in the art will understand that, for the sake of convenience and brevity, the division of the above-mentioned functional units and modules is only used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0132] like Figure 3 As shown, embodiments of the present invention provide a terminal device, such as... Figure 3 As shown, the terminal device D10 of this embodiment includes: at least one processor D100 ( Figure 3 The diagram shows only one processor, a memory D101, and a computer program D102 stored in the memory D101 and executable on the at least one processor D100, wherein the processor D100 executes the computer program D102 to implement the steps in any of the above method embodiments.

[0133] Specifically, when the processor D100 executes the computer program D102, it collects the train vibration signal of the target train and the temperature data corresponding to the battery under test in the target train; determines the vibration intensity based on the train vibration signal, and determines the pseudo-random code length based on the vibration intensity; generates a spread spectrum detection signal corresponding to the pseudo-random code length, obtains the reference incident signal corresponding to the spread spectrum detection signal, injects the spread spectrum detection signal into the battery under test through a preset detection link, and simultaneously collects the total reflection signal; obtains the link parameters of the preset detection link, determines the propagation delay of each reflection component in the total reflection signal based on the link parameters, and, based on the propagation delay, the preset reflection coefficient value range, and the propagation path formed by the same reflection in the preset detection link... The successive attenuation relationship between two adjacent reflections is used to constrain and separate the total reflection signal, resulting in a first reflection signal and at least one subsequent reflection signal. Based on the peak polarity, propagation delay, and reflection coefficient of each subsequent reflection signal relative to the first reflection signal, the first reflection signal and each subsequent reflection signal are corrected in the same direction and weighted and superimposed to obtain an enhanced reflection signal. The enhanced reflection signal and the reference incident signal are cross-correlated, and the cross-correlation peak value within a preset delay window is extracted. Temperature compensation is then performed on the cross-correlation peak value based on temperature data to obtain the liquid shortage evolution characteristics that characterize the change of the liquid shortage state with the detection cycle. The liquid shortage evolution characteristics are input into a pre-trained liquid replenishment prediction model to obtain the predicted liquid replenishment amount of the battery under test. The process involves collecting train vibration signals and battery temperature data, adaptively determining the pseudo-random code length for generating spread spectrum detection signals based on vibration intensity, thus matching detection efficiency with vibration interference resistance. The total reflected signal is separated according to the detection link parameters, reflection coefficient range, and successive attenuation relationship of adjacent reflections along the same path, reducing non-physical estimations caused by multiple reflection aliasing. Furthermore, the first and subsequent reflected signals are corrected in the same direction and weighted based on peak polarity, propagation delay, and reflection coefficient, enhancing the weak reflection characteristics caused by fluid shortage. Cross-correlation and temperature compensation are applied to the enhanced reflection signals to form fluid shortage evolution characteristics that change with the detection cycle. A model based on different fluid level detection data and actual fluid replenishment volume outputs a predicted fluid replenishment volume, improving the stability of fluid shortage feature extraction under complex operating conditions and increasing the accuracy of fluid replenishment prediction for the battery under test, thereby extending battery life and improving train safety.

[0134] The processor D100 can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.

[0135] In some embodiments, the memory D101 may be an internal storage unit of the terminal device D10, such as a hard disk or memory of the terminal device D10. In other embodiments, the memory D101 may be an external storage device of the terminal device D10, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the terminal device D10. Furthermore, the memory D101 may include both internal and external storage units of the terminal device D10. The memory D101 is used to store the operating system, applications, bootloader, data, and other programs, such as the program code of the computer program. The memory D101 can also be used to temporarily store data that has been output or will be output.

[0136] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps described in the various method embodiments above.

[0137] This application provides a computer program product that, when run on a terminal device, enables the terminal device to implement the steps described in the various method embodiments above.

[0138] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of protection of this application is limited to these examples; within the framework of this application, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of one or more embodiments of this application as described above, which are not provided in detail for the sake of brevity.

[0139] One or more embodiments in this application are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of this application. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of one or more embodiments in this application should be included within the protection scope of this application.

Claims

1. A method of predicting a replenishment amount of a train battery, characterized by, include: Collect the train vibration signal of the target train and the temperature data of the battery under test in the target train; The vibration intensity is determined based on the train vibration signal, and the pseudo-random code length is determined based on the vibration intensity. The pseudo-random code is used to generate the spread spectrum detection signal, and the length of the pseudo-random code is positively correlated with the vibration intensity. Generate a spread spectrum detection signal corresponding to the length of the pseudo-random code, obtain the reference incident signal corresponding to the spread spectrum detection signal, inject the spread spectrum detection signal into the battery under test through a preset detection link, and simultaneously collect the total reflected signal. The total reflected signal is formed by the reflection of the spread spectrum detection signal at the impedance discontinuity between the preset detection link and the battery under test; The link parameters of the preset detection link are obtained, the propagation delay of each reflection component in the total reflection signal is determined according to the link parameters, and the total reflection signal is constrained and separated based on the propagation delay, the preset reflection coefficient value range, and the successive attenuation relationship between two adjacent reflections formed by the same reflection propagation path in the preset detection link, to obtain a first reflection signal and at least one subsequent reflection signal. Based on the peak polarity, propagation delay, and reflection coefficient of each subsequent reflected signal relative to the first reflected signal, the first reflected signal and each subsequent reflected signal are corrected in the same direction and weighted and superimposed to obtain an enhanced reflected signal; The enhanced reflection signal and the reference incident signal are cross-correlated, the cross-correlation peak value within a preset time delay window is extracted, and the cross-correlation peak value is temperature compensated according to the temperature data to obtain the liquid shortage evolution characteristics that characterize the change of the liquid shortage state with the detection cycle. The fluid shortage evolution characteristics are input into a pre-trained fluid replenishment prediction model to obtain the predicted fluid replenishment amount of the battery under test; the fluid replenishment prediction model is established based on spread spectrum time-domain reflectometry data under different fluid level states and the corresponding actual fluid replenishment amount.

2. The train battery refill amount prediction method according to claim 1, characterized by, The link parameters include the transmission path length, sampling frequency, and transmission path structure of the preset detection link; The total reflected signal is constrained and separated based on the propagation delay, the preset range of reflection coefficient values, and the successive attenuation relationship between two adjacent reflections formed by the same reflection propagation path in the preset detection link, to obtain a first reflected signal and at least one subsequent reflected signal, including: The reflection components formed by the same reflection propagation path in the preset detection link are numbered in ascending order of propagation delay. The reference incident signal is then time-shifted according to the propagation delay of each reflection component to construct a delay test matrix. The expression for the delay test matrix is ​​as follows: ,in, This represents the delay test matrix; Indicates the sampling time; Indicates the first The propagation delay of each reflected component , This indicates the number of reflection components to be separated formed by the same reflection propagation path, and Not less than 2 ; This refers to the reference incident signal; Indicates according to the propagation delay The time-shifted reference incident signal; With the objective of minimizing the difference between the total reflected signal and the product of the time delay test matrix and the reflection coefficient vector, the reflection coefficients corresponding to each reflection component are obtained by solving an optimization model with reflection coefficient boundary constraints, adjacent reflection coefficient smoothing constraints, and successive attenuation constraints; the expression of the optimization model is: in, This represents a reflection coefficient vector composed of the reflection coefficients corresponding to each of the aforementioned reflection components; This represents the total reflected signal; This represents the first-order difference matrix used to determine the difference between adjacent reflection coefficients; Indicates the first The reflection coefficient corresponding to each reflection component ; This represents the attenuation coefficient between two adjacent reflections formed by the same reflection propagation path. ; This represents the first regularization coefficient corresponding to the smoothing constraint of the adjacent reflection coefficients; This represents the second regularization coefficient corresponding to the successive decay constraint; According to the propagation delay and the reflection coefficient corresponding to each reflection component, the reference incident signal is time-shifted and amplitude-adjusted, and a reflection signal corresponding to each reflection component is reconstructed. The reflection signal corresponding to the reflection component with the minimum propagation delay is determined as the first reflection signal, and the reflection signals corresponding to the remaining reflection components are determined as the subsequent reflection signals; wherein the reflection signal corresponding to the first reflection component is represented as after reconstruction. ; wherein, represents the first reflection signal, when is not less than 2, the subsequent reflection signal is represented.

3. The train battery refill volume prediction method according to claim 2, characterized by, The step of performing co-directional correction and weighted superposition on the first reflected signal and each of the subsequent reflected signals based on the peak polarity, propagation delay, and reflection coefficient of each of the subsequent reflected signals relative to the first reflected signal to obtain an enhanced reflected signal includes: Determine the signal-to-noise ratio (SNR) of each of the subsequent reflected signals, and determine the subsequent reflected signals with an SNR not lower than a preset compensation threshold as valid subsequent reflected signals; A peak polarity symmetry correction coefficient is determined based on the peak polarity of each effective subsequent reflected signal relative to the first reflected signal, and a compensation weight is determined based on the propagation delay and reflection coefficient of each effective subsequent reflected signal; the calculation expression for the compensation weight is as follows: in, Indicates the first The compensation weights corresponding to each subsequent reflected signal This represents the set consisting of the sequence numbers of each of the aforementioned valid subsequent reflected signals; This indicates the propagation delay of the first reflected signal; Indicates the first The propagation delay of the subsequent reflected signal; Represents the time decay constant; Indicates the first The reflection coefficient of the subsequent reflected signal; Indicates the first The reflection coefficient of an effective subsequent reflected signal; Indicates the first The propagation delay of an effective subsequent reflected signal; Based on the peak polarity simultaneity correction coefficient and the compensation weight, each of the effective subsequent reflection signals is weighted and superimposed with the first reflection signal to obtain the enhanced reflection signal; the expression of the enhanced reflection signal is: in, This refers to the enhanced reflection signal; Indicates the first The peak polarity correction coefficient corresponding to the i-th subsequent reflected signal, when the i-th... When the peak polarity of each subsequent reflected signal is the same as that of the first reflected signal, When the first When the peak polarity of a subsequent reflected signal is opposite to that of the first reflected signal, ; Indicates the first A subsequent reflected signal.

4. The train battery refill volume prediction method according to claim 3, characterized by, The step of performing cross-correlation processing on the enhanced reflected signal and the reference incident signal, extracting the cross-correlation peak value within a preset time delay window, and performing temperature compensation on the cross-correlation peak value based on the temperature data includes: Through calculation formula The first cross-correlation value of the second detection period at the time delay ; represents the enhanced reflection signal corresponding to the first detection period; Through calculation formula Get the first The cross-correlation peak value corresponding to each detection cycle ; This refers to the preset delay window; According to the cross-correlation peak values corresponding to different temperatures under the normal liquid level state, a temperature influence function is established and is calculated by a formula a first detection cycle corresponds to a temperature-compensated cross-correlation peak value of a second detection cycle ; denotes a function value of the temperature influence function at the temperature data ; 5. The train battery refill volume prediction method of claim 4, wherein, The obtained characteristics of the fluid shortage evolution, representing the change of the fluid shortage state with the detection cycle, include: The temperature-compensated cross-correlation peak values ​​are arranged according to the detection cycle to obtain the temperature-compensated peak value sequence; The temperature-compensated peak sequence is subjected to integrated empirical mode decomposition to obtain multiple intrinsic mode function components; Select an intrinsic mode function component with a correlation coefficient of not less than 0.3 with the temperature compensation peak sequence, or select a low-frequency intrinsic mode function component with a cumulative energy contribution rate of not less than 80%. The selected intrinsic mode function components are reconstructed to obtain the principal component peak curves, and the principal component peak curves are determined as the fluid deficiency evolution characteristics.

6. The train battery refill volume prediction method of claim 5, wherein, The pre-trained fluid replacement prediction model is one of the following: calibration curve model, multinomial regression model, piecewise regression model, support vector regression model, or neural network model. The pre-trained liquid replenishment prediction model takes the peak ratio, peak drop, short-term slope, temperature data, vibration intensity, battery model parameters, and historical liquid replenishment records as input, and the predicted liquid replenishment amount as output; the peak ratio, peak drop, and short-term slope are determined based on the principal component peak curve; The formula for calculating the peak ratio is as follows: ;in, Indicates the first Peak percentage corresponding to each detection cycle; The peak curve of the principal component indicates that the peak curve of the principal component is at the 1st twentieth ... The peak value of the principal component corresponding to each detection cycle; This represents the reference principal component peak value when the battery under test is in a normal liquid level state; The calculation expression of the peak drop is ; wherein, represents the peak drop corresponding to the i th detection period. The expression for calculating the short-time slope is as follows: ;in, Indicates the first The short-time slope corresponding to each detection cycle; This indicates the number of detection cycles used to determine the short-term slope; The peak curve of the principal component indicates that the peak curve of the principal component is at the 1st twentieth twenti The principal component peak value corresponding to each detection cycle.

7. A train battery replenishment amount prediction device characterized by comprising: include: The data acquisition module is used to acquire the train vibration signal of the target train and the temperature data of the battery under test in the target train, and input the train vibration signal and the temperature data into the processor. The processor is configured to determine the vibration intensity based on the train vibration signal, determine the pseudo-random code length based on the vibration intensity, and send the pseudo-random code length to the spread spectrum detection module; the pseudo-random code is used to generate the spread spectrum detection signal, and the pseudo-random code length is positively correlated with the vibration intensity. The spread spectrum detection module is used to generate a spread spectrum detection signal corresponding to the length of the pseudo-random code, obtain a reference incident signal corresponding to the spread spectrum detection signal, inject the spread spectrum detection signal into the battery under test through a preset detection link, synchronously collect the total reflection signal, and input the reference incident signal and the total reflection signal into the processor. The total reflected signal is formed by the reflection of the spread spectrum detection signal at the impedance discontinuity between the preset detection link and the battery under test; The processor is further configured to acquire the link parameters of the preset detection link, determine the propagation delay of each reflection component in the total reflection signal according to the link parameters, and constrain and separate the total reflection signal based on the propagation delay, the preset reflection coefficient value range, and the successive attenuation relationship between two adjacent reflections formed by the same reflection propagation path in the preset detection link, to obtain a first reflection signal and at least one subsequent reflection signal. The processor is further configured to perform co-directional correction and weighted superposition on the first reflected signal and each of the subsequent reflected signals based on the peak polarity, propagation delay and reflection coefficient of each of the subsequent reflected signals relative to the first reflected signal, to obtain an enhanced reflected signal; The processor is further configured to perform cross-correlation processing on the enhanced reflection signal and the reference incident signal, extract the cross-correlation peak value within a preset time delay window, and perform temperature compensation on the cross-correlation peak value according to the temperature data to obtain the liquid shortage evolution characteristics characterizing the change of the liquid shortage state with the detection cycle. The processor is further configured to input the liquid shortage evolution characteristics into a pre-trained liquid replenishment prediction model to obtain the predicted liquid replenishment amount of the battery under test; the liquid replenishment prediction model is established based on spread spectrum time-domain reflectometry data under different liquid level states and the corresponding actual liquid replenishment amount.

8. The train battery electrolyte replenishment prediction device according to claim 7, characterized in that, The data acquisition module includes a vibration acquisition module and a temperature acquisition module, and the spread spectrum detection module includes a signal generation module, a signal acquisition module, and a characteristic impedance of 50Ω. Coaxial cable, T-terminal and impedance of 50 The compensation terminal; The vibration acquisition module is used to acquire the acceleration signal during the operation of the target train and input the acceleration signal into the processor. The temperature acquisition module is used to acquire the temperature of the battery under test or the temperature of the environment around the battery under test and input the temperature into the processor. The signal generation module is used to generate two in-phase pseudo-random code signals modulated by binary phase shift keying. One of the pseudo-random code signals is directly input to the signal acquisition module as the reference incident signal, and the other pseudo-random code signal is used as the spread spectrum detection signal and injected into the battery under test through the T-terminal and the coaxial cable. The signal acquisition module is used to synchronously acquire the reference incident signal and the total reflected signal, and input the reference incident signal and the total reflected signal into the processor. The compensation terminal is used to perform impedance matching compensation on the preset detection link.

9. A terminal device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method as described in any one of claims 1 to 6.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 6.

Citation Information

Patent Citations

  • Device and method for detecting internal defect of battery cell by using tdr

    US12429525B2