Lead-acid battery charge state prediction method based on multi-source data fusion

Abstract

The invention provides a lead-acid battery state-of-charge prediction method based on multi-source data fusion, and the method comprises the steps: collecting electrical data, vibration data and acoustic data in the charging and discharging process of a lead-acid battery, and enabling the acoustic data to comprise bubble rupture sound and electrolyte flow sound; carrying out correction processing on the electrical data by adopting acoustic data, and carrying out interference suppression processing on the vibration data; performing feature extraction on the processed electrical data and vibration data to obtain an electrical feature vector and a vibration feature vector; fusing the electrical feature vector and the vibration feature vector to obtain a fusion vector, inputting the fusion vector into the trained echo state network, and predicting the state of charge; after the acoustic data and the vibration data are introduced, the electrical data and the vibration data are respectively corrected and subjected to interference suppression processing through the acoustic data, the measurement deviation of the electrical data is corrected, the accuracy is improved, meanwhile, environmental noise is eliminated, the multi-source data fusion precision is improved, and the accuracy of a state of charge prediction result is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of battery management, and in particular to a method for predicting the state of charge of a lead-acid battery based on multi-source data fusion. Background Art

[0002] Lead-acid batteries, as core components of energy storage and power systems, are widely used in electric vehicles, energy storage systems and other fields due to their low cost and high safety. Accurate prediction of their state of charge (SOC) is crucial to system safety and service life. Existing SOC prediction technologies mostly rely on electrical signals such as voltage and current, and achieve estimation through equivalent circuit models or data-driven algorithms. However, there are obvious defects: electrical signals are easily affected by polarization effects, temperature drift and battery aging, resulting in measurement deviations. For example, the sudden voltage change during high current charging and discharging has a nonlinear correlation with SOC, and it is difficult to reflect the physical and chemical processes inside the battery. At the same time, existing technologies generally ignore the potential value of acoustic signals and vibration signals: acoustic signals such as bubble burst sound and electrolyte flow sound are directly related to ion migration efficiency and plate reaction intensity, and can serve as a natural basis for correcting electrical signal deviations; although vibration signals can reflect internal states such as plate deformation, they are easily disturbed by environmental vibrations and are difficult to use effectively without targeted processing. Prediction methods that rely solely on electrical signals are not only limited in accuracy, but also fail to form a multi-source data collaborative optimization mechanism, and fail to establish the inherent correlation between different signals - especially ignoring the coupling relationship between the physical and chemical processes inside the battery and the electrical / vibration signals. The direct reflection characteristics of acoustic signals on the battery's charging and discharging status give them unique value in correcting electrical signal deviations and suppressing vibration noise, but there is currently no technology to use them in the above scenarios, resulting in insufficient robustness of SOC prediction under complex working conditions. Summary of the Invention

[0003] In view of this, the present invention proposes a lead-acid battery state of charge prediction method based on multi-source data fusion, which can improve the accuracy of state of charge prediction.

[0004] The technical solution of the present invention is achieved as follows: A method for predicting the state of charge of a lead-acid battery based on multi-source data fusion includes the following steps: Step S1, collecting electrical data, vibration data, and acoustic data during the charge and discharge process of the lead-acid battery, wherein the acoustic data includes bubble bursting sound and electrolyte flow sound; Step S2: Using acoustic data to correct electrical data and perform interference suppression on vibration data; Step S3: performing feature extraction on the processed electrical data and vibration data to obtain electrical feature vectors and vibration feature vectors; Step S4: Fusing the electrical feature vector and the vibration feature vector to obtain a fused vector, and inputting the fused vector into the trained echo state network to predict the state of charge.

[0005] Preferably, the specific steps of step S1 include: Step S11: collecting voltage and current data of the lead-acid battery during charging and discharging through the battery management system, and calculating the internal resistance based on the voltage and current data; Step S12: using a vibration sensor to collect vibration data during the charge and discharge process of the lead-acid battery; Step S13: using a circular microphone array to collect the bubble bursting sound and electrolyte flow sound during the charge and discharge process of the lead-acid battery.

[0006] Preferably, after collecting the electrical data, vibration data and acoustic data, step S1 performs data cleaning, enhancement and time synchronization on the electrical data, vibration data and acoustic data.

[0007] Preferably, the specific steps of using acoustic data to correct electrical data in step S2 include: Step S21: Calculate the coefficient of variation of the main frequency through the bubble burst sound. When the coefficient of variation is less than a preset threshold, retain the internal resistance at the same moment. Step S22: When the coefficient of variation is greater than a preset threshold, the internal resistance increment is calculated based on the main frequency of the bubble burst sound, and the internal resistance is updated according to the internal resistance increment; Step S23: Calculate the flow acoustic energy through the electrolyte flow acoustics, convert the flow acoustic energy into ion migration efficiency, and retain the voltage data at the same moment when the ion migration efficiency is greater than a preset threshold; Step S24: When the ion migration efficiency is less than a preset threshold, it is determined that ion migration is blocked, and the voltage data is corrected based on the ion migration efficiency.

[0008] Preferably, the expression for calculating the internal resistance increment based on the main frequency of the bubble burst sound in step S22 is: ,in = , is the correction factor obtained by fitting experiments on different lead-acid batteries, is the main frequency of bubble bursting sound.

[0009] Preferably, the expression for correcting the voltage data based on the ion migration efficiency in step S24 is: , They are the voltage data before and after correction, is the ion migration efficiency at the current moment, is the ion transfer efficiency of a new lead-acid battery.

[0010] Preferably, the specific steps of performing interference suppression processing on vibration data using acoustic data in step S2 include: Step S25: Calculate the sound source coordinates based on the bubble burst sound and the electrolyte flow sound using the TDOA algorithm; Step S26: The signal collected by the vibration sensor is focused on the vibration direction corresponding to the sound source coordinates through a beamforming algorithm, and the ambient vibration noise in other directions is suppressed.

[0011] Preferably, the specific steps of step S3 include: Step S31: extracting time-domain static features, time-domain dynamic features, frequency-domain features, and charge-discharge stage features from the electrical data, and concatenating the time-domain static features, time-domain dynamic features, frequency-domain features, and charge-discharge stage features into an electrical feature vector; Step S32: extracting time domain statistical features, time domain waveform features, frequency domain features, and the main vibration frequency offset under different currents from the vibration data, and splicing the time domain statistical features, time domain waveform features, frequency domain features, and the main vibration frequency offset under different currents into a vibration feature vector.

[0012] Preferably, the specific steps of step S4 include: Step S41: After dimensional alignment and feature normalization of the electrical eigenvector and the vibration eigenvector, the electrical eigenvector and the vibration eigenvector are directly concatenated in sequence to form a fusion vector; Step S42: Input the fused vector into the echo state network in time series, and generate a high-dimensional state matrix from the reserve pool of the echo state network; Step S43: Using the high-dimensional state matrix as input, ridge regression is used to solve the output layer weights and predict the state of charge.

[0013] Compared with the prior art, the present invention has the following beneficial effects: The present invention provides a lead-acid battery state of charge prediction method based on multi-source data fusion. In addition to collecting traditional electrical data, it also collects vibration data and acoustic data. The acoustic data can respectively correct the electrical data and perform interference suppression processing to improve the accuracy of the acoustic signal, while eliminating environmental vibration interference. The processed electrical data and vibration data are subjected to feature processing to obtain feature vectors and then fused. Finally, an echo state network is introduced, and the fused vector is used as input. The predicted state of charge is obtained through echo state network processing. This can not only improve the multi-source fusion accuracy, but also improve the robustness of SOC prediction under complex working conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only preferred embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0015] Figure 1 This is a flow chart of a method for predicting the state of charge of a lead-acid battery based on multi-source data fusion according to the present invention; Figure 2 Flowchart of step S1 of a method for predicting state of charge of a lead-acid battery based on multi-source data fusion according to the present invention; Figure 3 Flowchart of step S2 of a method for predicting state of charge of a lead-acid battery based on multi-source data fusion according to the present invention; Figure 4 Flowchart of step S3 of a method for predicting state of charge of a lead-acid battery based on multi-source data fusion according to the present invention; Figure 5 This is a flowchart of step S4 of a lead-acid battery state of charge prediction method based on multi-source data fusion according to the present invention. DETAILED DESCRIPTION

[0016] In order to better understand the technical content of the present invention, a specific embodiment is provided below, and the present invention is further described in conjunction with the accompanying drawings.

[0017] See also Figures 1 to 5 The present invention provides a method for predicting the state of charge of a lead-acid battery based on multi-source data fusion, comprising the following steps: Step S1, collecting electrical data, vibration data, and acoustic data during the charge and discharge process of the lead-acid battery, wherein the acoustic data includes bubble bursting sound and electrolyte flow sound; Step S2: Using acoustic data to correct electrical data and perform interference suppression on vibration data; Step S3: performing feature extraction on the processed electrical data and vibration data to obtain electrical feature vectors and vibration feature vectors; Step S4: Fusing the electrical feature vector and the vibration feature vector to obtain a fused vector, and inputting the fused vector into the trained echo state network to predict the state of charge.

[0018] The present invention provides a lead-acid battery state of charge prediction method based on multi-source data fusion, which realizes multi-source collaborative optimization by introducing acoustic signals and vibration signals. In addition to collecting traditional electrical signals such as voltage and current, two types of key auxiliary signals are synchronously collected: one is the acoustic signal, including the bubble burst sound and electrolyte flow sound generated during the charging and discharging process. Among them, the bubble burst sound originates from the water electrolysis reaction in the late stage of battery charging, which will generate oxygen and hydrogen bubbles. The frequency and pulse density of the bubble sound increase significantly with the increase of SOC, and the bubble sound frequency of the aging battery will shift to low frequency; the electrolyte flow sound presents energy fluctuations due to the difference in ion migration rate at different stages of charging and discharging. , which can directly reflect the ion migration efficiency and the state of the electrolyte. The second is the vibration signal, which covers the vibration of the battery shell and the micro-vibration of the plate. Its frequency and amplitude are related to the expansion / contraction of the active material of the plate and the impact strength of the electrolyte, and directly reflect the dynamic changes of the internal structure of the battery. Based on the above signal characteristics, the present invention uses the characteristics of the bubble burst sound and the electrolyte flow sound to correct the measurement deviation of the electrical signal caused by polarization effect and temperature drift, and improve the accuracy of the electrical data. On the other hand, through the homology of the acoustic signal and the inherent vibration of the battery, interference suppression processing is performed to eliminate the environmental noise in the vibration signal and retain the effective vibration characteristics related to the internal state of the battery.

[0019] Feature extraction is performed on the optimized electrical data and vibration data respectively. The two types of extracted feature vectors are fused to form a high-dimensional fusion vector, which is input into the pre-trained echo state network. The echo state network, relying on the efficient capture of time series features by the reserve pool, maps the fusion vector into the SOC prediction value. Electrical correction and vibration noise reduction are achieved by introducing acoustic signals. Combining multi-source feature fusion with the echo state network model, it breaks through the limitations of traditional single electrical signal prediction and significantly improves the accuracy and robustness of lead-acid battery SOC prediction under complex working conditions.

[0020] Preferably, the specific steps of step S1 include: Step S11: collecting voltage and current data of the lead-acid battery during charging and discharging through the battery management system, and calculating the internal resistance based on the voltage and current data; Step S12: using a vibration sensor to collect vibration data during the charge and discharge process of the lead-acid battery; Step S13: using a circular microphone array to collect the bubble bursting sound and electrolyte flow sound during the charge and discharge process of the lead-acid battery.

[0021] The built-in sensors in the battery management system collect voltage and current data in real time. At the same time, the internal resistance of the battery at different times can be calculated based on the collected voltage and current data. The voltage data, current data and battery internal resistance are combined into electrical data. The vibration sensor can be installed in the battery casing to collect pulse vibrations caused by bubbles impacting the plates or casing, as well as low-frequency vibrations caused by plate bending and active material shedding. A high-sensitivity MEMS ring microphone array is set at the key position of the battery shell. The ring microphone array can collect the bubble bursting sound and electrolyte flow sound during the charging and discharging process of the lead-acid battery. In order to ensure the subsequent processing of acoustic data on electrical data and vibration data, all collected data are preprocessed after data collection is completed. The preprocessing includes data cleaning, enhancement and time synchronization, removing noise data and enhancing it, and synchronizing the clocks between data from different sources. Subsequent processing based on acoustic data refers to processing data of the same time series.

[0022] Preferably, the specific steps of using acoustic data to correct electrical data in step S2 include: Step S21: Calculate the coefficient of variation of the main frequency through the bubble burst sound. When the coefficient of variation is less than a preset threshold, retain the internal resistance at the same moment. Step S22: When the coefficient of variation is greater than a preset threshold, the internal resistance increment is calculated based on the main frequency of the bubble burst sound, and the internal resistance is updated according to the internal resistance increment; Step S23: Calculate the flow acoustic energy through the electrolyte flow acoustics, convert the flow acoustic energy into ion migration efficiency, and retain the voltage data at the same moment when the ion migration efficiency is greater than a preset threshold; Step S24: When the ion migration efficiency is less than a preset threshold, it is determined that ion migration is blocked, and the voltage data is corrected based on the ion migration efficiency.

[0023] Traditional internal resistance measurement is susceptible to instantaneous interference, resulting in pseudo-high or pseudo-low internal resistance. Therefore, the internal resistance can be corrected by the consistency of bubble bursting. The main frequency variation coefficient of the bubble bursting sound is calculated in real time and compared with the preset threshold. If it is less than the preset threshold, it means that the reaction state is stable. At this time, the measured internal resistance is the true value and can be retained. If the variation coefficient is greater than the preset threshold, it means that the internal resistance at this time is an interference value and needs to be corrected. The correction process requires calling the correlation factor , whose expression is = ,in is the main frequency of bubble burst sound. After obtaining the correlation factor, according to the expression of internal resistance increment: , calculate the internal resistance increment, add the internal resistance increment to the internal resistance calculated at the current moment, and you can get the corrected resistance value. In terms of , it is a coefficient fitted by experimental data and needs to be calibrated for different battery models.

[0024] In addition, the voltage data will be corrected by the electrolyte flow sound. First, the electrolyte flow sound is converted into flow sound energy. The electrolyte flow sound energy reflects the ion migration efficiency, and the voltage data can be corrected by the ion migration efficiency. Similarly, the set threshold is compared with the ion migration efficiency. If it is greater than the preset threshold, it is determined that the ion migration is normal and the current voltage data is retained. If it is less than the preset threshold, it means that the ion migration is blocked and the voltage needs to be corrected. The correction expression is: , They are the voltage data before and after correction, is the ion migration efficiency at the current moment, is the ion transfer efficiency of a new lead-acid battery.

[0025] Preferably, the specific steps of performing interference suppression processing on vibration data using acoustic data in step S2 include: Step S25: Calculate the sound source coordinates based on the bubble burst sound and the electrolyte flow sound using the TDOA algorithm; Step S26: The signal collected by the vibration sensor is focused on the vibration direction corresponding to the sound source coordinates through a beamforming algorithm, and the ambient vibration noise in other directions is suppressed.

[0026] The vibration signal may contain environmental interference and be mixed with the internal vibration of the battery. Therefore, the present invention can perform interference suppression processing on the vibration data based on the acoustic data. First, the TDOA algorithm can be used to accurately locate the coordinates of the sound source based on the acoustic signal, and then the focusing direction of the vibration sensor is processed. Through the beamforming algorithm, the vibration direction is focused on the vibration direction corresponding to the sound source coordinates, ensuring that the collected vibration data is the vibration caused by the bursting of bubbles or the flow of electrolyte inside the battery, suppressing environmental vibration noise in other directions, ensuring the accuracy of the vibration data, and reducing the interference of external noise.

[0027] Preferably, the specific steps of step S3 include: Step S31: extracting time-domain static features, time-domain dynamic features, frequency-domain features, and charge-discharge stage features from the electrical data, and concatenating the time-domain static features, time-domain dynamic features, frequency-domain features, and charge-discharge stage features into an electrical feature vector; Step S32: extracting time domain statistical features, time domain waveform features, frequency domain features, and the main vibration frequency offset under different currents from the vibration data, and splicing the time domain statistical features, time domain waveform features, frequency domain features, and the main vibration frequency offset under different currents into a vibration feature vector.

[0028] The processed electrical data and vibration data both need to be feature extracted. The features extracted from the electrical data include time domain static features, time domain dynamic features, frequency domain features, and charge and discharge stage features. The time domain static features include the mean, peak, valley, variance, and standard deviation of voltage and current. The time domain dynamic features include the rate of change of voltage and current, pulse width, etc. The frequency domain features require Fourier transform of the voltage / current signal to extract the main frequency, harmonic component amplitude, and high-frequency energy ratio to reflect the frequency distribution characteristics of the voltage / current signal. The charge and discharge stage features include the segmented calculation of the voltage mean and variance in each interval to highlight the differences in electrical characteristics under different charge states.

[0029] The time domain statistical features of vibration data include the root mean square value, peak factor, kurtosis, etc. of the vibration signal, which quantify the vibration intensity and impact characteristics. The time domain waveform features include the waveform factor and pulse factor of the vibration signal, which are used to describe the morphological characteristics of the vibration waveform. The frequency domain features include the center frequency, frequency band energy distribution, and resonance peak frequency of the vibration signal extracted by wavelet transform or power spectrum analysis, which are used to reflect the frequency domain characteristics of the vibration. The main frequency offset of the vibration under different currents needs to be calculated in combination with the charge and discharge current size, which can be used to capture the correlation between vibration and charge and discharge intensity. After the feature extraction is completed, the extracted features are directly spliced ​​into electrical feature vectors and vibration feature vectors.

[0030] Preferably, the specific steps of step S4 include: Step S41: After dimensional alignment and feature normalization of the electrical eigenvector and the vibration eigenvector, the electrical eigenvector and the vibration eigenvector are directly concatenated in sequence to form a fusion vector; Step S42: Input the fused vector into the echo state network in time series, and generate a high-dimensional state matrix from the reserve pool of the echo state network; Step S43: Using the high-dimensional state matrix as input, ridge regression is used to solve the output layer weights and predict the state of charge.

[0031] After obtaining the electrical eigenvector and the vibration eigenvector, the two eigenvectors need to be fused. Before fusion, the dimensions need to be aligned first, and the lengths need to be unified through dimensionality reduction. Then, the two eigenvectors are scaled and mapped to the [0, 1] interval to eliminate the influence of dimensional differences on fusion. Finally, the electrical eigenvector and the vibration eigenvector are directly concatenated in sequence to obtain the fusion vector. The echo state network needs to be trained in advance, and the parameters of the echo state network are initialized. The reserve pool size, input weight, reserve pool connection weight and sparsity are set to ensure that the echo state characteristics are met. Then, the fusion vector of the existing multi-source data is input into the echo state network, and a high-dimensional state matrix is ​​generated through nonlinear transformation of the reserve pool. The high-dimensional state matrix is ​​used as input and the true state of charge is used as output. Ridge regression is used to solve the output layer weights to complete the network training. After the training is completed, the currently obtained fusion vector can be input into the echo state network, and the echo state network predicts the state of charge.

[0032] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A lead-acid battery state of charge prediction method based on multi-source data fusion, characterized in that: The following steps are involved: Step S1, collecting electrical data, vibration data, and acoustic data during the charge and discharge process of the lead-acid battery, wherein the acoustic data includes bubble bursting sound and electrolyte flow sound; Step S2: Using acoustic data to correct electrical data and perform interference suppression on vibration data; Step S3: performing feature extraction on the processed electrical data and vibration data to obtain electrical feature vectors and vibration feature vectors; Step S4: Fusing the electrical feature vector and the vibration feature vector to obtain a fused vector, and inputting the fused vector into the trained echo state network to predict the state of charge.

2. A lead-acid battery state of charge prediction method based on multi-source data fusion according to claim 1, characterized in that: The specific steps of step S1 include: Step S11: collecting voltage and current data of the lead-acid battery during charging and discharging through the battery management system, and calculating the internal resistance based on the voltage and current data; Step S12: using a vibration sensor to collect vibration data during the charge and discharge process of the lead-acid battery; Step S13: using a circular microphone array to collect the bubble bursting sound and electrolyte flow sound during the charge and discharge process of the lead-acid battery.

3. The method for predicting state of charge of a lead-acid battery based on multi-source data fusion according to claim 1, characterized in that: After collecting the electrical data, vibration data and acoustic data, the step S1 performs data cleaning, enhancement and time synchronization on the electrical data, vibration data and acoustic data.

4. The method for predicting state of charge of a lead-acid battery based on multi-source data fusion according to claim 2, characterized in that: The specific steps of using acoustic data to correct electrical data in step S2 include: Step S21: Calculate the coefficient of variation of the main frequency through the bubble burst sound. When the coefficient of variation is less than a preset threshold, retain the internal resistance at the same moment. Step S22: When the coefficient of variation is greater than a preset threshold, the internal resistance increment is calculated based on the main frequency of the bubble burst sound, and the internal resistance is updated according to the internal resistance increment; Step S23: Calculate the flow acoustic energy through the electrolyte flow acoustics, convert the flow acoustic energy into ion migration efficiency, and retain the voltage data at the same moment when the ion migration efficiency is greater than a preset threshold; Step S24: When the ion migration efficiency is less than a preset threshold, it is determined that ion migration is blocked, and the voltage data is corrected based on the ion migration efficiency.

5. The method for predicting state of charge of a lead-acid battery based on multi-source data fusion according to claim 4, characterized in that: The expression for calculating the internal resistance increment based on the main frequency of the bubble burst sound in step S22 is: ,in = , is the correction factor obtained by fitting experiments on different lead-acid batteries, is the main frequency of bubble burst sound.

6. The method for predicting state of charge of a lead-acid battery based on multi-source data fusion according to claim 4, characterized in that: The expression for correcting the voltage data based on the ion migration efficiency in step S24 is: , They are the voltage data before and after correction, is the ion migration efficiency at the current moment, is the ion transfer efficiency of a new lead-acid battery.

7. The method for predicting state of charge of a lead-acid battery based on multi-source data fusion according to claim 2, characterized in that: The specific steps of step S2 of using acoustic data to perform interference suppression processing on vibration data include: Step S25: Calculate the sound source coordinates based on the bubble burst sound and the electrolyte flow sound using the TDOA algorithm; Step S26: The signal collected by the vibration sensor is focused on the vibration direction corresponding to the sound source coordinates through a beamforming algorithm, and the ambient vibration noise in other directions is suppressed.

8. The method for predicting state of charge of a lead-acid battery based on multi-source data fusion according to claim 1, characterized in that: The specific steps of step S3 include: Step S31: extracting time-domain static features, time-domain dynamic features, frequency-domain features, and charge-discharge stage features from the electrical data, and concatenating the time-domain static features, time-domain dynamic features, frequency-domain features, and charge-discharge stage features into an electrical feature vector; Step S32: extracting time domain statistical features, time domain waveform features, frequency domain features, and the main vibration frequency offset under different currents from the vibration data, and splicing the time domain statistical features, time domain waveform features, frequency domain features, and the main vibration frequency offset under different currents into a vibration feature vector.

9. The method for predicting state of charge of a lead-acid battery based on multi-source data fusion according to claim 1, characterized in that: The specific steps of step S4 include: Step S41: After dimensional alignment and feature normalization of the electrical eigenvector and the vibration eigenvector, the electrical eigenvector and the vibration eigenvector are directly concatenated in sequence to form a fusion vector; Step S42: Input the fused vector into the echo state network in time series, and generate a high-dimensional state matrix from the reserve pool of the echo state network; Step S43: Using the high-dimensional state matrix as input, ridge regression is used to solve the output layer weights and predict the state of charge.

Images