Lithium battery pack voltage conversion control method

Abstract

The invention belongs to the technical field of voltage conversion control, and discloses a lithium battery pack voltage conversion control method, which comprises the following steps: acquiring voltage signal data, battery basic data and battery environment data; based on the voltage signal data, performing layer-by-layer decomposition by using wavelet packet transformation to obtain sub-band voltage signals; performing feature extraction based on the sub-band voltage signal to obtain feature data; based on the battery basic data, the battery environment data and the feature data, performing data fusion by using an improved PCA to obtain a fusion vector; inputting the fusion vector into a pre-trained full-connection neural network model, and predicting to obtain a health state index; setting a health state index threshold value and performing state judgment, performing voltage equalization on healthy single batteries, and cutting off the connection between unhealthy single batteries and the lithium battery pack, namely, not performing voltage equalization; performing voltage conversion based on the balanced voltage to reach the voltage required by the load; and the safety and the stability of the battery pack are obviously improved.

Description

[0001] The present invention relates to the technical field of voltage conversion control, and in particular to a lithium battery pack voltage conversion control method. Background Art

[0002] The existing lithium battery pack voltage conversion control technology has exposed many problems: First, most of the current lithium battery management systems rely on traditional temperature sensors for temperature monitoring, but these sensors can only provide a single temperature data, and cannot effectively capture the local temperature differences of each single cell inside the battery. Especially in large-scale battery packs, there may be large temperature variations between different batteries, increasing the safety risks of the system.

[0003] Secondly, the data is single and useful features are not deeply mined. It only relies on simple basic data such as battery voltage, current, temperature, etc. for monitoring and analysis. The feature extraction is not detailed enough. Most of the existing methods ignore the in-depth feature extraction of battery data, especially in the frequency domain characteristics and time series characteristics of battery voltage and current signals.

[0004] In addition, there is a lack of health prediction models based on multi-dimensional data. Current battery health predictions are mostly based on simple rules and threshold judgments, and lack intelligent modeling methods. Even using intelligent algorithms such as machine learning, many systems are simply applied to a single data source, lacking basic battery data, battery environmental data and multi-source data fusion, making it difficult to accurately capture the dynamic changes and trends of battery health, and thus unable to provide timely and accurate health warnings.

[0005] In view of this, the present invention proposes a lithium battery pack voltage conversion control method to solve the above problem. Summary of the invention

[0006] In order to overcome the above-mentioned defects of the prior art and to achieve the above-mentioned purpose, the present invention provides the following technical solution, a lithium battery pack voltage conversion control method, comprising: step S1, obtaining voltage signal data, battery basic data and battery environment data; the battery basic data includes single battery voltage data, single battery current data, and single battery actual temperature data; the battery environment data includes ambient temperature data and ambient humidity data;

[0007] Step S2: Based on the voltage signal data, perform layer-by-layer decomposition using wavelet packet transform to obtain a sub-band voltage signal; perform feature extraction based on the sub-band voltage signal to obtain feature data;

[0008] Step S3: Based on the battery basic data, the battery environment data and the characteristic data, the improved PCA is used to perform data fusion to obtain a fusion vector;

[0009] Step S4, inputting the fusion vector into a pre-trained fully connected neural network model to predict a health status index;

[0010] Step S5, setting a health status index threshold and performing status judgment, performing voltage balancing on healthy cells, and disconnecting unhealthy cells from the lithium battery pack, i.e., not performing voltage balancing;

[0011] Step S6: Perform voltage conversion based on the balanced voltage.

[0012] Furthermore, the basic battery data includes single battery voltage data, single battery current data, and single battery actual temperature data; the voltage data of the single battery is measured in real time by a voltage sensor, and the current data of the single battery is measured in real time by a Hall effect sensor;

[0013] The battery environment data includes environment temperature data and environment humidity data. The temperature data of the environment in which the single battery is located is measured in real time by a temperature sensor, and the humidity data of the environment in which the single battery is located is measured in real time by a humidity sensor.

[0014] Furthermore, the method for acquiring the actual temperature data of the single battery includes:

[0015] Use a high-resolution infrared imager to perform thermal imaging scanning on the lithium battery pack to generate a thermal image. Each pixel in the thermal image represents a specific location, and each pixel value represents the temperature of the location. Use image segmentation technology to segment the thermal image into S1 regions, each region corresponds to a single battery, and calculate the average temperature of each single battery pack based on the pixel value on the infrared thermal image.

[0016] Temperature data is obtained by installing a temperature sensor in each single cell area, and the temperature data is unified with the average temperature data obtained by the infrared imager, and weighted average is performed to obtain the actual temperature data of the single cell.

[0017] The specific methods of the image segmentation technology include:

[0018] De-noising and contrast enhancement are performed on the thermal image to obtain a pre-processed thermal image;

[0019] The Sobel operator is used to calculate the gradients of the preprocessed thermal image in the horizontal and vertical directions. The horizontal and vertical gradients are combined, and the Euclidean norm is used to calculate the gradient strength of each pixel. Then a threshold T is set to determine whether the pixel is an edge. When the gradient strength is greater than the threshold T, the pixel is considered to be an edge pixel and is represented by 1. When the gradient strength is less than or equal to the threshold T, the pixel is considered not to be an edge pixel and is represented by 0, forming a binary image.

[0020] For binary images, connected edge pixels are grouped into the same group of regions through depth-first search, and each region represents the boundary of a single cell;

[0021] For the obtained single cell boundary, the average temperature of the single cell is calculated by calculating the pixel value in each area. The formula is: Among them, T avg represents the average temperature in the single cell, R represents the number of pixels in the single cell area, and T(x,y) represents the temperature at the position (x,y).

[0022] The method of obtaining the sub-band voltage signal by performing layer-by-layer decomposition based on the voltage signal data by using wavelet packet transform includes:

[0023] The single battery voltage signal is collected by the data acquisition device ADC, and based on the voltage signal, the wavelet packet transform is used to perform layer-by-layer decomposition;

[0024] The wavelet basis function Daubechies is selected for voltage signal decomposition, and the number of decomposition layers is set to Q layers. At each decomposition layer, the voltage signal is divided into low-frequency and high-frequency parts.

[0025] Perform the first-layer decomposition on the voltage signal, use the low-pass filter and high-pass filter of Daubechies wavelet to convolve the voltage signal to obtain low-frequency sub-bands and high-frequency sub-bands, downsample the filtered low-frequency sub-bands to obtain the first-layer low-frequency output signal;

[0026] For the first layer low-frequency output signal, apply low-pass filter and high-pass filter again to obtain new low-frequency sub-band and high-frequency sub-band, and downsample, repeat the above steps until the set Q-layer decomposition level is reached;

[0027] Get the low-frequency sub-band voltage signal DP of the Q layer Q [n] and high frequency sub-band voltage signal GP Q [n].

[0028] Furthermore, the method of extracting features based on the sub-band voltage signal to obtain feature data includes:

[0029] The sub-band voltage signal includes a low-frequency sub-band voltage signal and a high-frequency sub-band voltage signal; the extracted feature data includes energy features, entropy features, degree features and skewness features of the low-frequency sub-band voltage signal and the high-frequency sub-band voltage signal;

[0030] The energy signature is expressed by calculating the sum of the squares of the low-frequency and high-frequency sub-band voltage signals: and represents the energy characteristics of the low-frequency and high-frequency sub-band voltage signals at the qth layer, DPq [n] GP q [n] represents the sampling value of the qth layer low-frequency and high-frequency sub-band voltage signal at time point n, n represents the sampling point index in the low-frequency and high-frequency sub-band voltage signal, N q Represents the length of the voltage signal of the low-frequency and high-frequency sub-bands of the qth layer;

[0031] Entropy feature extraction is performed using low-frequency sub-band voltage signals and high-frequency sub-band voltage signals:

[0032] Among them, H(DP q ) and H(GP q ) represents the entropy characteristics of the voltage signals of the low-frequency and high-frequency sub-bands in the qth layer, P(DP q [n]), P(GP q [n]) represents the probability distribution of each sample value in the low-frequency and high-frequency sub-band voltage signals;

[0033] Calculate the mean and standard deviation of the low-frequency and high-frequency sub-band voltage signals, and calculate the kurtosis and skewness characteristics of the low-frequency and high-frequency sub-band voltage signals through the mean and standard deviation. Kurtosis: in, and represents the kurtosis characteristics of the voltage signals of the low-frequency and high-frequency sub-bands in the qth layer, and represents the mean value of the voltage signal of the low-frequency and high-frequency sub-bands of the qth layer, and represents the standard deviation of the voltage signals of the low-frequency and high-frequency sub-bands of the qth layer;

[0034] Skewness: in, and Represents the skewness characteristics of the voltage signals in the low-frequency and high-frequency sub-bands at the qth layer.

[0035] Furthermore, the method of obtaining the fusion vector by using the improved PCA to perform data fusion based on the battery basic data, the battery environment data and the characteristic data includes:

[0036] Data preprocessing: Process missing values ​​and outliers on the acquired battery basic data, battery environment data, and feature data, unify timestamps, and then merge them into a comprehensive matrix according to timestamps;

[0037] Weighted normalization: Perform weighted normalization on each data in the comprehensive matrix Among them, s represents the index of the feature, w s Represents the weight of the sth feature, and the weight is calculated using the entropy weight method, Tz s,grepresents the original value of the sth feature on the gth sample, Jz s represents the mean of the sth feature, Bc s represents the standard deviation of the sth feature, Bz s,g represents the standardized data value of the sth feature on the gth sample, and integrates the weighted standardized features to obtain the weighted standardized matrix Z;

[0038] Calculate the covariance matrix: Calculate the covariance matrix using the weighted normalization matrix Z Where Z T is the transpose of the weighted normalization matrix Z, and G represents the total number of samples;

[0039] Eigenvalue decomposition: Perform eigenvalue decomposition on the covariance matrix C to obtain the eigenvalues ​​and eigenvectors of the covariance matrix: C*v s =λ s *v s , where λ s is the sth eigenvalue, representing the variance of the sth principal component, v s is the sth eigenvector, indicating the direction of the sth principal component;

[0040] Principal component selection: Arrange the eigenvalues ​​in descending order, select the eigenvectors corresponding to the first k eigenvalues ​​as the principal components, and the standard for selecting the principal components is to calculate the cumulative variance contribution rate. When CV reaches 95%, the first k eigenvalues ​​at this time are selected;

[0041] Data projection: Select the first k principal components and project the weighted normalization matrix Z onto the eigenvector matrix v corresponding to the first K principal components k On the above, we get the matrix after dimension reduction;

[0042] Get the fusion vector: After data projection, the reduced-dimensional matrix ZH is the fusion vector.

[0043] Furthermore, the method of calculating the weight using the entropy weight method includes:

[0044] Perform Z-score standardization on each feature data, and then calculate the proportion of each feature standardized data in the total sample. Among them Bz' s,g represents the standardized data of the sth feature on the gth sample, Bl s,g Represents the probability distribution of the sth feature on the gth sample;

[0045] Calculate the information entropy of each feature data

[0046] According to the information entropy XH s Calculate the redundancy Yds =1-XH s ;

[0047] Calculate the weight of each feature according to its redundancy Among them, S represents the total number of features.

[0048] Furthermore, the fusion vector is input into a pre-trained fully connected neural network model to predict the health status index, and the method includes:

[0049] Step A1: The sample set includes P_o groups of samples, each group of samples includes a fusion vector and a corresponding health status index;

[0050] Step A2: Use random initialization of weight matrix parameters and bias parameters in the fully connected neural network;

[0051] Step A2: Forward propagation, passing the input fusion vector from the input layer through each hidden layer to the output layer, and calculating the output of each layer;

[0052] From the input layer to the hidden layer 1, for the g1th sample, the output of the first layer is W 1 represents the weight matrix of the first layer, ZH g1 represents the input of the g1th sample, b 1 represents the bias term of the first layer, and f is the ReLU activation function;

[0053] For the propagation between hidden layers, we set L layers (L = 2, 3, 4, ..., l), and the formula for each layer is: in represents the output of the L-1th layer, represents the output of the Lth layer, W L represents the weight matrix of the Lth layer, b L represents the bias term of the Lth layer;

[0054] The propagation of the output layer, the output of the last layer is the predicted value W out represents the weight matrix of the output layer, b out Represents the bias term of the output layer;

[0055] Step A3: Calculate the loss function and use the average error as the loss function to measure the predicted value and the true value Y g1 The difference between:

[0056] Step A4: Back propagation, using the loss function to calculate the gradient of each parameter in each layer, and updating the weight matrix and bias term according to the gradient;

[0057] Step A5: Repeat steps A2 to A4 until the number of iterations is reached, stop the iteration, and use the trained fully connected neural network model to predict new battery health status index data.

[0058] Furthermore, the specific method of setting the health status index threshold and performing status judgment, performing voltage balancing on healthy cells and cutting off the connection between unhealthy cells and the lithium battery pack, i.e., not performing voltage balancing, includes:

[0059] The battery health index threshold is set to 80%. A health index greater than 80% indicates that the battery is healthy, and a health index less than or equal to 80% indicates that the battery is unhealthy.

[0060] For cells with a health status index greater than 80%, the voltage data is monitored in real time, and passive and active balancing are used to balance the voltage.

[0061] The threshold range of the equalization voltage is set according to the average value and standard deviation of the voltage of all healthy single cells at the current moment: DY lower =DY avg -k1*β DY , DY upper =DY avg +k1*β DY , where DY lower and DY upper Indicates the upper and lower limits of the balanced voltage, k1 is a constant, k1 = 1 or 2, DY avg Table 1: The average voltage of all healthy cells at present, β DY Table: Voltage standard deviation of all healthy cells at present;

[0062] For single cell voltages higher than the balanced voltage range, a relay control circuit is used to consume power and reduce the voltage to the balanced voltage range; for single cell voltages lower than the balanced voltage, active balancing is used to increase the voltage through external injection to bring it to the balanced voltage range;

[0063] For cells with a health status index of less than or equal to 80%, a relay is used to cut off the connection between the unhealthy cells and the lithium battery pack, and an early warning is issued to remind replacement.

[0064] Furthermore, the specific manner of performing voltage conversion based on the balanced voltage includes:

[0065] For the balanced voltage of a healthy lithium battery pack, a multi-level voltage converter is used to convert the balanced voltage into a stable voltage Dy required by the load;

[0066] In the first stage, a Buck-Boost converter is used to adjust the balanced voltage of the lithium battery to the voltage range FW1 required by the load;

[0067] In the second stage, the Buck-Boost converter is adjusted to the range FW1, and a linear regulator is used to adjust the load demand voltage Mb;

[0068] In the third stage, the linear regulator is adjusted to obtain the load demand voltage Mb, and the high-frequency noise in the voltage is removed using the low-dropout linear regulator voltage to obtain the stable voltage required by the final load.

[0069] The technical effects and advantages of the lithium battery pack voltage conversion control method of the present invention are as follows:

[0070] The present invention achieves accurate prediction of the health status of a lithium battery pack by comprehensively utilizing basic battery data, battery environment data, and voltage signal characteristics after wavelet packet transformation and feature extraction, combined with improved principal component analysis (PCA) for data fusion. Based on the prediction result, the present invention can intelligently perform balanced voltage control on healthy cells in the lithium battery pack to prevent unhealthy cells from participating in the balanced voltage operation, and cut off the connection between the unhealthy cells and the battery pack through a relay, thereby greatly improving the safety and service life of the battery pack.

[0071] In addition, the real-time monitoring of battery temperature using high-resolution infrared imaging technology improves the monitoring accuracy of single-cell temperature, effectively reduces the failure rate of lithium battery packs, and avoids safety accidents caused by single-cell overheating or voltage imbalance. Through accurate battery health prediction, the battery life cycle is extended, and the energy efficiency of the battery management system is optimized, thereby improving the performance stability and economy of the entire battery pack. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 A schematic diagram of a lithium battery pack voltage conversion control method according to the present invention;

[0073] Figure 2 This is a schematic diagram of a lithium battery pack voltage conversion control system of the present invention. DETAILED DESCRIPTION

[0074] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0075] Example 1

[0076] See also Figure 1 As shown, a lithium battery pack voltage conversion control method described in this embodiment includes:

[0077] Step S1, acquiring voltage signal data, battery basic data and battery environment data; the battery basic data includes single battery voltage data, single battery current data and single battery actual temperature data; using a voltage sensor to measure the voltage data of the single battery in real time, and using a Hall effect sensor to measure the current data of the single battery in real time;

[0078] The battery environment data includes environment temperature data and environment humidity data. The temperature data of the environment in which the single battery is located is measured in real time by a temperature sensor, and the humidity data of the environment in which the single battery is located is measured in real time by a humidity sensor.

[0079] Step S2: Based on the voltage signal data, perform layer-by-layer decomposition using wavelet packet transform to obtain a sub-band voltage signal; perform feature extraction based on the sub-band voltage signal to obtain feature data;

[0080] Step S3: Based on the battery basic data, the battery environment data and the characteristic data, the improved PCA is used to perform data fusion to obtain a fusion vector;

[0081] Step S4, inputting the fusion vector into a pre-trained fully connected neural network model to predict a health status index;

[0082] Step S5, setting a health status index threshold and performing status judgment, performing voltage balancing on healthy cells, and disconnecting unhealthy cells from the lithium battery pack, i.e., not performing voltage balancing;

[0083] Step S6: Perform voltage conversion based on the balanced voltage.

[0084] The methods for obtaining the actual temperature data of a single battery include:

[0085] The use of a combination of an infrared imager and a temperature sensor to calculate the actual temperature of a single cell is mainly to improve the accuracy and reliability of temperature measurement. The infrared imager can quickly and globally obtain the temperature distribution on the surface of the battery pack through thermal imaging technology, and provide the average temperature of each single cell area, but it has limitations in resolution and accuracy; while the temperature sensor can directly and accurately measure the actual temperature of each single cell, avoiding possible errors in infrared imaging; by unifying the timestamp and performing weighted averaging on the two types of data, it is possible to combine the advantages of the two, eliminate their respective defects, and obtain more accurate and comprehensive actual temperature data of the single cell;

[0086] Use a high-resolution infrared imager to perform thermal imaging scanning on the lithium battery pack to generate a thermal image. Each pixel in the thermal image represents a specific location, and each pixel value represents the temperature of the location. Use image segmentation technology to segment the thermal image into S1 regions, each region corresponds to a single battery, and calculate the average temperature of each single battery pack based on the pixel value on the infrared thermal image.

[0087] Temperature data is obtained by installing a temperature sensor in each single cell area, and the temperature data is unified with the average temperature data obtained by the infrared imager, and weighted average is performed to obtain the actual temperature data of the single cell.

[0088] The specific methods of image segmentation technology include:

[0089] De-noising and contrast enhancement are performed on the thermal image to obtain a pre-processed thermal image;

[0090] The Sobel operator is used to calculate the gradients of the preprocessed thermal image in the horizontal and vertical directions. The horizontal and vertical gradients are combined, and the Euclidean norm is used to calculate the gradient strength of each pixel. Then a threshold T is set to determine whether the pixel is an edge. When the gradient strength is greater than the threshold T, the pixel is considered to be an edge pixel and is represented by 1. When the gradient strength is less than or equal to the threshold T, the pixel is considered not to be an edge pixel and is represented by 0, forming a binary image.

[0091] For binary images, connected edge pixels are grouped into the same group of regions through depth-first search, and each region represents the boundary of a single cell;

[0092] For the obtained single cell boundary, the average temperature of the single cell is calculated by calculating the pixel value in each area. The formula is: Among them, T avg represents the average temperature in the single cell, R represents the number of pixels in the single cell area, and T(x,y) represents the temperature at the position (x,y);

[0093] The horizontal gradient formula is: SP x (x,y)=∑ i,j I(x+i,y+j)*SP x (i,j), where SP x (x, y) represents the horizontal gradient value of the preprocessed thermal image at the position (x, y), which reflects the rate of change in the horizontal direction of the preprocessed thermal image, that is, the edge strength in the horizontal direction. I(x+i, y+j) represents the pixel value of the input preprocessed thermal image I at the position (x+i, y+i). (x, y) represents the current pixel position. i and j represent the offset of the current pixel (x, y) in the horizontal and vertical coordinates. SP x(i, j) represents the weight of the horizontal Sobel kernel;

[0094] The vertical gradient is: SP y (x,y)=∑ i,j I(x+i,y+j)*SP y (i,j), SP y (x, y) represents the vertical gradient value of the preprocessed thermal image at position (x, y), S y (i, j) represents the weight of the vertical Sobel kernel;

[0095] The gradient strength is: Among them, SP(x,y) is the gradient strength of the pixel;

[0096] Edge judgment: 1 represents the edge part, and 0 represents the non-edge part;

[0097] Through depth-first search, connected edge pixels are grouped into the same group of regions, each of which represents the boundary of a single battery. A label matrix L1 is created to record the region to which each pixel belongs. The pixel is initialized, i.e., L1(x, y) = 0. For each pixel in the binary image, it is detected whether it is an edge pixel, i.e., the binary image value is 1. A depth-first search is used, and the 8-neighborhood connection method is used to judge whether the edge pixels around it are marked as the same group of regions. Each new unmarked edge pixel is judged and marked.

[0098] Based on the voltage signal data, wavelet packet transform is used to perform layer-by-layer decomposition to obtain the sub-band voltage signal in the following ways:

[0099] The single battery voltage signal is collected by the data acquisition device ADC, and based on the voltage signal, the wavelet packet transform is used to perform layer-by-layer decomposition;

[0100] Wavelet packet transform is a time-frequency analysis method that decomposes a signal into different frequency bands. It is an extension of wavelet transform. By further decomposing the high-frequency and low-frequency parts of the signal, the signal decomposition in the time domain and frequency domain is more detailed, so that the characteristics of the signal can be better captured in many applications.

[0101] The wavelet basis function Daubechies is selected for voltage signal decomposition, and the number of decomposition layers is set to Q layers. At each decomposition layer, the voltage signal is divided into low-frequency and high-frequency parts.

[0102] Perform the first-layer decomposition on the voltage signal, use the low-pass filter and high-pass filter of Daubechies wavelet to convolve the voltage signal to obtain low-frequency sub-bands and high-frequency sub-bands, downsample the filtered low-frequency sub-bands to obtain the first-layer low-frequency output signal;

[0103] For the first layer low-frequency output signal, apply low-pass filter and high-pass filter again to obtain new low-frequency sub-band and high-frequency sub-band, and downsample, repeat the above steps until the set Q-layer decomposition level is reached;

[0104] Get the low-frequency sub-band voltage signal DP of the Q layer Q [n] and high frequency sub-band voltage signal GP Q [n];

[0105] Among them, the low-pass filter retains the slowly changing part of the voltage signal, that is, the stable, lower frequency part, and removes the rapidly changing part of the signal, that is, the high frequency component;

[0106] A high-pass filter retains the rapidly changing portions of a voltage signal and removes the smooth, lower-frequency components;

[0107] Low-frequency sub-band voltage signal: This is the signal part after being processed by the low-pass filter, which contains the low-frequency components in the voltage signal;

[0108] High-frequency sub-band voltage signal: This is the signal part after being processed by the high-pass filter, which contains the high-frequency components in the voltage signal;

[0109] Downsampling. In wavelet transform, downsampling usually refers to the downsampling operation on the filtered signal, that is, only retaining a part of the signal samples. Specifically, if the signal is filtered (low pass or high pass), downsampling is to reduce the number of sampling points of the signal, usually by a multiple of 2. For example, if the signal was originally sampled 100 times per second, after downsampling, it may become only 50 times per second.

[0110] Methods for extracting features based on sub-band voltage signals and obtaining feature data include:

[0111] The sub-band voltage signal includes a low-frequency sub-band voltage signal and a high-frequency sub-band voltage signal; the extracted feature data includes energy features, entropy features, degree features and skewness features of the low-frequency sub-band voltage signal and the high-frequency sub-band voltage signal;

[0112] Each low-frequency sub-band and high-frequency sub-band contains the energy of the voltage signal within the frequency range. The energy is expressed by calculating the sum of the squares of the voltage signals of the low-frequency and high-frequency sub-bands: and represents the energy characteristics of the low-frequency and high-frequency sub-band voltage signals at the qth layer, DP q [n] GP q [n] represents the sampling value of the qth layer low-frequency and high-frequency sub-band voltage signal at time point n, n represents the sampling point index in the low-frequency and high-frequency sub-band voltage signal, Nq Represents the length of the voltage signal of the low-frequency and high-frequency sub-bands of the qth layer;

[0113] Assume that after the three-layer wavelet decomposition, the low-frequency sub-band voltage signal DP 3 The length of [n] is 8, that is, N 3 =8, then n will be taken from 1 to 8, corresponding to each sampling point in the low-frequency sub-band voltage signal, the energy is EDP 3 =DP 3 [1] 2 +DP 3 [2] 2 +DP 3 [3] 2 +DP 3 [4] 2 +DP 3 [5] 2 +DP 3 [6] 2 +DP 3 [7] 2 +

[0114] DP 3 [8] 2 ;

[0115] Entropy feature extraction is performed using low-frequency sub-band voltage signals and high-frequency sub-band voltage signals:

[0116] Among them, H(DP q ) and H(GP q ) represents the entropy characteristics of the voltage signals of the low-frequency and high-frequency sub-bands in the qth layer, P(DP q [n]), P(GP q [n]) represents the probability distribution of each sample value in the low-frequency and high-frequency sub-band voltage signals;

[0117] Calculate the mean and standard deviation of the low-frequency and high-frequency sub-band voltage signals, and calculate the kurtosis and skewness characteristics of the low-frequency and high-frequency sub-band voltage signals through the mean and standard deviation. Kurtosis: in, and represents the kurtosis characteristics of the voltage signals of the low-frequency and high-frequency sub-bands in the qth layer, and represents the mean value of the voltage signal of the low-frequency and high-frequency sub-bands of the qth layer, and represents the standard deviation of the voltage signals of the low-frequency and high-frequency sub-bands of the qth layer;

[0118] Skewness: in, and Represents the skewness characteristics of the voltage signals in the low-frequency and high-frequency sub-bands at the qth layer.

[0119] Based on the battery basic data, battery environment data and feature data, the improved PCA is used for data fusion to obtain the fusion vector in the following ways:

[0120] Data preprocessing: Process missing values ​​and outliers on the acquired battery basic data, battery environment data, and feature data, unify timestamps, and then merge them into a comprehensive matrix according to timestamps;

[0121] Weighted normalization: Perform weighted normalization on each data in the comprehensive matrix Among them, s represents the index of the feature, w s Represents the weight of the sth feature, and the weight is calculated using the entropy weight method, Tz s,g represents the original value of the sth feature on the gth sample, Jz s represents the mean of the sth feature, Bc s represents the standard deviation of the sth feature, Bz s,g represents the standardized data value of the sth feature on the gth sample, and integrates the weighted standardized features to obtain the weighted standardized matrix Z;

[0122] Calculate the covariance matrix: Calculate the covariance matrix using the weighted normalization matrix Z Where Z T is the transpose of the weighted normalization matrix Z, and G represents the total number of samples;

[0123] Eigenvalue decomposition: Perform eigenvalue decomposition on the covariance matrix C to obtain the eigenvalues ​​and eigenvectors of the covariance matrix: C*v s =λ s *v s , where λ s is the sth eigenvalue, representing the variance of the sth principal component, v s is the sth eigenvector, indicating the direction of the sth principal component;

[0124] Principal component selection: Arrange the eigenvalues ​​in descending order, select the eigenvectors corresponding to the first k eigenvalues ​​as the principal components, and the standard for selecting the principal components is to calculate the cumulative variance contribution rate. When CV reaches 95%, the first k eigenvalues ​​at this time are selected;

[0125] Data projection: Select the first k principal components and project the weighted normalization matrix Z onto the eigenvector matrix v corresponding to the first K principal components k The shape is S*k, and it is implemented by matrix multiplication: ZH=Z*v k , ZH represents the matrix after dimension reduction;

[0126] Get the fusion vector: After data projection, the reduced-dimensional matrix ZH is the fusion vector;

[0127] Among them, the cumulative variance contribution rate CV reaches 95% in order to reduce the data dimension and reduce the computational complexity while retaining most of the important information of the data; by selecting the principal components that can explain 95% of the variance, redundancy and noise can be effectively removed, the main features of the data can be retained, and the loss of key information due to excessive dimensionality reduction can be avoided; the 95% standard is a commonly used compromise solution that can find a reasonable balance between information retention and dimensionality compression.

[0128] Ways to calculate weights using the entropy weight method include:

[0129] Perform Z-score standardization on each feature data Then calculate the proportion of each feature standardized data to the total samples, Among them Bz' s,g represents the standardized data of the sth feature on the gth sample, Bl s,g Represents the probability distribution of the sth feature on the gth sample;

[0130]

[0131] The larger the entropy value, the less information the feature has.

[0132] According to the information entropy XH s Calculate the redundancy Yd s =1-XH s ,Redundancy reflects the amount of effective information in the feature. The higher the redundancy, the richer the feature information;

[0133] Calculate the weight of each feature according to its redundancy Among them, S represents the total number of features.

[0134] The fusion vector is input into the pre-trained fully connected neural network model, and the health status index is predicted by:

[0135] Step A1: The sample set includes P_o groups of samples, each group of samples includes a fusion vector and a corresponding health status index;

[0136] Step A2: Use random initialization of weight matrix parameters and bias parameters in the fully connected neural network;

[0137] Step A2: Forward propagation, passing the input fusion vector from the input layer through each hidden layer to the output layer, and calculating the output of each layer;

[0138] From the input layer to the hidden layer 1, for the g1th sample, the output of the first layer is W 1 represents the weight matrix of the first layer, ZH g1 represents the input of the g1th sample, b 1 represents the bias term of the first layer, and f is the ReLU activation function;

[0139] For the propagation between hidden layers, we set L layers (L = 2, 3, 4, ..., l), and the formula for each layer is: in represents the output of the L-1th layer, represents the output of the Lth layer, W L represents the weight matrix of the Lth layer, b L represents the bias term of the Lth layer;

[0140] The propagation of the output layer, the output of the last layer is the predicted value W out represents the weight matrix of the output layer, b out Represents the bias term of the output layer;

[0141] Step A3: Calculate the loss function and use the average error as the loss function to measure the predicted value and the true value Y g1 The difference between:

[0142] Step A4: Back propagation, using the loss function to calculate the gradient of each parameter in each layer, and updating the weight matrix and bias term according to the gradient;

[0143] Calculate the gradient of the output layer, that is, the partial derivative of the loss function with respect to the weights of the output layer and the partial derivative of the bias

[0144] For the hidden layer L, the chain rule is used to calculate the gradient of the loss function for each weight matrix and the gradient of the bias term For layer L-1 to layer 1, the gradient is recursively calculated using the chain rule: where f'(sh L-1 ) is the inverse of the ReLU activation function. When sh L-1 >0, then f'(sh L-1 )=1, when sh L-1 ≤0, then f'(sh L-1 )=0;

[0145] Update the weight vector of each layer using gradient descent and the bias term η represents the learning rate;

[0146] Step A5: Repeat steps A2 to A4 until the number of iterations is reached, stop the iteration, and use the trained fully connected neural network model to predict new battery health status index data.

[0147] The specific methods of setting the health status index threshold, equalizing the voltage of healthy single cells, and using relays to cut off the connection with the lithium battery pack for unhealthy single cells without equalizing the voltage include:

[0148] According to the battery factory manual, the battery health index threshold is set to 80%. A health index greater than 80% indicates that the battery is healthy, and a health index less than or equal to 80% indicates that the battery is unhealthy.

[0149] For cells with a health status index greater than 80%, the voltage data is monitored in real time, and passive and active balancing are used to balance the voltage.

[0150] The threshold range of the equalization voltage is set according to the average value and standard deviation of the voltage of all healthy single cells at the current moment: DY lower =DY avg -k1*β DY , DY upper =DY avg +k1*β DY , where DY lower and DY upper Indicates the upper and lower limits of the balanced voltage, k1 is a constant, k1 = 1 or 2, DY avg Table 1: The average voltage of all healthy cells at present, β DY Table: Voltage standard deviation of all healthy cells at present;

[0151] For single cell voltages higher than the balanced voltage range, a relay control circuit is used to consume power and reduce the voltage to the balanced voltage range; for single cell voltages lower than the balanced voltage, active balancing is used to increase the voltage through external injection to bring it to the balanced voltage range;

[0152] For cells with a health status index of less than or equal to 80%, a relay is used to cut off the connection between the unhealthy cells and the lithium battery pack, and an early warning is issued to remind replacement.

[0153] Based on the balanced voltage, the specific methods of voltage conversion include:

[0154] For the balanced voltage of a healthy lithium battery pack, a multi-level voltage converter is used to convert the balanced voltage into a stable voltage Dy required by the load;

[0155] In the first stage, a Buck-Boost converter is used to adjust the balanced voltage of the lithium battery to the voltage range FW1 required by the load;

[0156] In the second stage, the Buck-Boost converter is adjusted to the range FW1, and a linear regulator is used to adjust the load demand voltage Mb;

[0157] In the third stage, the linear regulator is adjusted to obtain the load demand voltage Mb, and the high-frequency noise in the voltage is removed using the low-dropout linear regulator voltage to obtain the stable voltage required by the final load.

[0158] In this embodiment, by comprehensively utilizing battery basic data, battery environment data, and voltage signal characteristics after wavelet packet transformation and feature extraction, combined with improved principal component analysis (PCA) for data fusion, accurate prediction of the health status of the lithium battery pack is achieved. Based on the prediction results, the healthy single cells in the lithium battery pack can be intelligently controlled to balance the voltage, to prevent unhealthy single cells from participating in the balanced voltage operation, and to cut off the connection between the unhealthy single cells and the battery pack through relays, thereby greatly improving the safety and service life of the battery pack.

[0159] In addition, the real-time monitoring of battery temperature using high-resolution infrared imaging technology improves the monitoring accuracy of single-cell temperature, effectively reduces the failure rate of lithium battery packs, and avoids safety accidents caused by single-cell overheating or voltage imbalance. Through accurate battery health prediction, the battery life cycle is extended, and the energy efficiency of the battery management system is optimized, thereby improving the performance stability and economy of the entire battery pack.

[0160] Example 2

[0161] See also Figure 2 As shown, the part not described in detail in this embodiment is described in the description of embodiment 1, and a lithium battery pack voltage conversion control system is provided, including: a data acquisition module: acquiring voltage signal data, battery basic data and battery environment data; the battery basic data includes single battery voltage data, single battery current data, and single battery actual temperature data; the battery environment data includes ambient temperature data and ambient humidity data;

[0162] Feature extraction module: Based on the voltage signal data, wavelet packet transform is used to perform layer-by-layer decomposition to obtain sub-band voltage signals; feature extraction is performed based on the sub-band voltage signals to obtain feature data;

[0163] Data fusion module: Based on battery basic data, battery environment data and feature data, the improved PCA is used to perform data fusion to obtain a fusion vector;

[0164] Battery management module: Input the fusion vector into the pre-trained fully connected neural network model to predict the health status index;

[0165] Balancing voltage module: sets the health status index threshold and performs status judgment. Healthy cells are subjected to voltage balancing, while unhealthy cells are disconnected from the lithium battery pack, i.e., voltage balancing is not performed.

[0166] Voltage conversion module: performs voltage conversion based on the balanced voltage.

[0167] Example 3

[0168] This embodiment discloses an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the operation mode of the lithium battery pack voltage conversion control method provided above is implemented.

[0169] Since the electronic device introduced in this embodiment is an electronic device used to implement a lithium battery pack voltage conversion control method in the embodiment of the present application, based on the lithium battery pack voltage conversion control method introduced in the embodiment of the present application, the technical personnel of the field can understand the specific implementation of the electronic device of the present embodiment and its various variations, so how the electronic device implements the method in the embodiment of the present application is not described in detail here. As long as the technical personnel of the field implement the electronic device used in the lithium battery pack voltage conversion control method in the embodiment of the present application, it belongs to the scope of protection of the present application.

[0170] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters and thresholds in the formula are set by technicians in this field according to actual conditions.

[0171] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technical users in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. A lithium battery pack voltage conversion control method, characterized in that: include: Step S1, acquiring voltage signal data, battery basic data and battery environment data; The basic battery data includes single battery voltage data, single battery current data, and single battery actual temperature data; the battery environment data includes ambient temperature data and ambient humidity data; Step S2: Based on the voltage signal data, perform layer-by-layer decomposition using wavelet packet transform to obtain a sub-band voltage signal; perform feature extraction based on the sub-band voltage signal to obtain feature data; Step S3: Based on the battery basic data, the battery environment data and the characteristic data, the improved PCA is used to perform data fusion to obtain a fusion vector; Step S4, inputting the fusion vector into a pre-trained fully connected neural network model to predict a health status index; Step S5, setting a health status index threshold and performing status judgment, performing voltage balancing on healthy cells, and disconnecting unhealthy cells from the lithium battery pack, i.e., not performing voltage balancing; Step S6: Perform voltage conversion based on the balanced voltage.

2. A lithium battery pack voltage conversion control method according to claim 1, characterized in that: The method for obtaining the actual temperature data of the single battery includes: Use an infrared imager to perform thermal imaging scanning on the lithium battery pack to generate a thermal image. Use image segmentation technology to segment the thermal image into S1 regions, each region corresponds to a single battery, and calculate the average temperature of each single battery pack; Temperature data is obtained by installing a temperature sensor in each single cell area, and the temperature data is unified with the average temperature data obtained by the infrared imager, and weighted average is performed to obtain the actual temperature data of the single cell.

3. A lithium battery pack voltage conversion control method according to claim 2, characterized in that: The specific methods of the image segmentation technology include: De-noising and contrast enhancement are performed on the thermal image to obtain a pre-processed thermal image; The Sobel operator is used to calculate the gradient of the preprocessed thermal image in the horizontal and vertical directions, and the Euclidean norm is used to calculate the gradient strength of each pixel. Then a threshold T is set to determine whether the pixel is an edge. When the gradient strength is greater than the threshold T, the pixel is considered to be an edge pixel and is represented by 1. When the gradient strength is less than or equal to the threshold T, the pixel is considered not to be an edge pixel and is represented by 0, forming a binary image. For binary images, connected edge pixels are grouped into the same group of regions through depth-first search, and each region represents the boundary of a single cell; For the obtained single cell boundary, the average temperature of the single cell is calculated according to the pixel value in each area. The formula is: Among them, T avg represents the average temperature in the single cell, R represents the number of pixels in the single cell area, and T(x,y) represents the temperature at the position (x,y).

4. A lithium battery pack voltage conversion control method according to claim 3, characterized in that: The method of obtaining the sub-band voltage signal by performing layer-by-layer decomposition based on the voltage signal data by using wavelet packet transform includes: The single cell voltage signal is collected by the data acquisition device ADC, and the wavelet packet transform is used to decompose it layer by layer; The wavelet basis function Daubechies is selected for voltage signal decomposition, and the number of decomposition layers is set to Q layers. At each decomposition layer, the voltage signal is divided into low-frequency and high-frequency parts. Perform the first-layer decomposition on the voltage signal, use the low-pass filter and high-pass filter of Daubechies wavelet to convolve the voltage signal to obtain low-frequency sub-bands and high-frequency sub-bands, downsample the filtered low-frequency sub-bands to obtain the first-layer low-frequency output signal; For the first layer low-frequency output signal, apply low-pass filter and high-pass filter again to obtain new low-frequency sub-band and high-frequency sub-band, and downsample, repeat the above steps until the set Q-layer decomposition level is reached; Get the low-frequency sub-band voltage signal DP of the Q layer Q [n] and high frequency sub-band voltage signal GP Q [n].

5. A lithium battery pack voltage conversion control method according to claim 4, characterized in that: The method of extracting features based on the sub-band voltage signal to obtain feature data includes: The sub-band voltage signal includes a low-frequency sub-band voltage signal and a high-frequency sub-band voltage signal; the extracted feature data includes energy features, entropy features, degree features and skewness feature data of the low-frequency sub-band voltage signal and the high-frequency sub-band voltage signal; The energy signature is expressed by calculating the sum of the squares of the low-frequency and high-frequency sub-band voltage signals: and represents the energy characteristics of the low-frequency and high-frequency sub-band voltage signals at the qth layer, DP q [n] GP q [n] represents the sampling value of the qth layer low-frequency and high-frequency sub-band voltage signal at time point n, n represents the sampling point index in the low-frequency and high-frequency sub-band voltage signal, N q Represents the length of the voltage signal of the low-frequency and high-frequency sub-bands of the qth layer; Entropy feature extraction is performed using low-frequency sub-band voltage signals and high-frequency sub-band voltage signals: Among them, H(DP q ) and H(GP q ) represents the entropy characteristics of the voltage signals of the low-frequency and high-frequency sub-bands in the qth layer, P(DP q [n]), P(GP q [n]) represents the probability distribution of each sample value in the low-frequency and high-frequency sub-band voltage signals; Calculate the mean and standard deviation of the low-frequency and high-frequency sub-band voltage signals, and calculate the kurtosis and skewness characteristics of the low-frequency and high-frequency sub-band voltage signals through the mean and standard deviation. Kurtosis: in, and represents the kurtosis characteristics of the voltage signals of the low-frequency and high-frequency sub-bands in the qth layer, and represents the mean value of the voltage signal of the low-frequency and high-frequency sub-bands of the qth layer, and represents the standard deviation of the voltage signals of the low-frequency and high-frequency sub-bands of the qth layer; Skewness: in, and Represents the skewness characteristics of the voltage signals in the low-frequency and high-frequency sub-bands at the qth layer.

6. A lithium battery pack voltage conversion control method according to claim 5, characterized in that: The method of performing data fusion based on the battery basic data, the battery environment data and the characteristic data by using the improved PCA to obtain the fusion vector includes: Data preprocessing: Process missing values ​​and outliers on the acquired battery basic data, battery environment data, and feature data, unify timestamps, and then merge them into a comprehensive matrix according to timestamps; Weighted normalization: Perform weighted normalization on each data in the comprehensive matrix Among them, s represents the index of the feature, w s Represents the weight of the sth feature, and the weight is calculated using the entropy weight method, Tz s,g represents the original value of the sth feature on the gth sample, Jz s represents the mean of the sth feature, Bc s represents the standard deviation of the sth feature, Bz s,g represents the standardized data value of the sth feature on the gth sample, and integrates the weighted standardized features to obtain the weighted standardized matrix Z; Calculate the covariance matrix: Calculate the covariance matrix using the weighted normalization matrix Z Where Z T is the transpose of the weighted normalization matrix Z, and G represents the total number of samples; Eigenvalue decomposition: Perform eigenvalue decomposition on the covariance matrix C to obtain the eigenvalues ​​and eigenvectors of the covariance matrix: C*v s =λ s *v s , where λ s is the sth eigenvalue, representing the variance of the sth principal component, v s is the sth eigenvector, indicating the direction of the sth principal component; Principal component selection: Arrange the eigenvalues ​​in descending order, select the eigenvectors corresponding to the first k eigenvalues ​​as the principal components, and the standard for selecting the principal components is to calculate the cumulative variance contribution rate. When CV reaches 95%, the first k eigenvalues ​​at this time are selected; Data projection: Select the first k principal components and project the weighted normalization matrix Z onto the eigenvector matrix v corresponding to the first K principal components k On the above, we get the matrix after dimension reduction; Get the fusion vector: After data projection, the reduced-dimensional matrix ZH is the fusion vector.

7. A lithium battery pack voltage conversion control method according to claim 6, characterized in that: The method of calculating the weight using the entropy weight method includes: Perform Z-score standardization on each feature data, and then calculate the proportion of each feature standardized data in the total sample. Among them Bz' s,g represents the standardized data of the sth feature on the gth sample, Bl s,g Represents the probability distribution of the sth feature on the gth sample; Calculate the information entropy of each feature data According to the information entropy XH s Calculate the redundancy Yd s =1-XH s ; Calculate the weight of each feature according to its redundancy Among them, S represents the total number of features.

8. A lithium battery pack voltage conversion control method according to claim 7, characterized in that: The method of inputting the fusion vector into the pre-trained fully connected neural network model to predict the health status index includes: Step A1: The sample set includes P_o groups of samples, each group of samples includes a fusion vector and a corresponding health status index; Step A2: Use random initialization of weight matrix parameters and bias parameters in the fully connected neural network; Step A2: Forward propagation, passing the input fusion vector from the input layer through each hidden layer to the output layer, and calculating the output of each layer; From the input layer to the hidden layer 1, for the g1th sample, the output of the first layer is W1 represents the weight matrix of the first layer, ZH g1 represents the input of the g1th sample, b1 represents the bias term of the first layer, and f is the ReLU activation function; For the propagation between hidden layers, we set L layers (L = 2, 3, 4, ..., l), and the formula for each layer is: in represents the output of the L-1th layer, represents the output of the Lth layer, W L represents the weight matrix of the Lth layer, b L represents the bias term of the Lth layer; The propagation of the output layer, the output of the last layer is the predicted value W out represents the weight matrix of the output layer, b out Represents the bias term of the output layer; Step A3: Calculate the loss function and use the average error as the loss function to measure the predicted value and the true value Y g1 The difference between: Step A4: Back propagation, using the loss function to calculate the gradient of each parameter in each layer, and updating the weight matrix and bias term according to the gradient; Step A5: Repeat steps A2 to A4 until the number of iterations is reached, stop the iteration, and use the trained fully connected neural network model to predict new battery health status index data.

9. A lithium battery pack voltage conversion control method according to claim 8, characterized in that: The specific method of setting the health status index threshold and performing status judgment, performing voltage balancing on healthy cells and cutting off the connection between unhealthy cells and the lithium battery pack, i.e., not performing voltage balancing, includes: The battery health index threshold is set to 80%. A health index greater than 80% indicates that the battery is healthy, and a health index less than or equal to 80% indicates that the battery is unhealthy. For cells with a health status index greater than 80%, the voltage data is monitored in real time, and passive and active balancing are used to balance the voltage. The threshold range of the equalization voltage is set according to the average value and standard deviation of the voltage of all healthy single cells at the current moment: DY lower =DY avg -k1*β DY , DY upper =DY avg +k1*β DY , where DY lower and DY upper Indicates the upper and lower limits of the balanced voltage, k1 is a constant, k1 = 1 or 2, DY avg Table 1: The average voltage of all healthy cells at present, β DY Table: Voltage standard deviation of all healthy cells at present; For single cell voltages higher than the balanced voltage range, a relay control circuit is used to consume power and reduce the voltage to the balanced voltage range; for single cell voltages lower than the balanced voltage, active balancing is used to increase the voltage through external injection to bring it to the balanced voltage range; For cells with a health status index of less than or equal to 80%, a relay is used to cut off the connection between the unhealthy cells and the lithium battery pack, and an early warning is issued to remind replacement.

10. A lithium battery pack voltage conversion control method according to claim 9, characterized in that: The specific method of performing voltage conversion based on the balanced voltage includes: For the balanced voltage of a healthy lithium battery pack, a multi-level voltage converter is used to convert the balanced voltage into a stable voltage Dy required by the load; In the first stage, a Buck-Boost converter is used to adjust the balanced voltage of the lithium battery to the voltage range FW1 required by the load; In the second stage, the Buck-Boost converter is adjusted to the range FW1, and a linear regulator is used to adjust the load demand voltage Mb; In the third stage, the linear regulator is adjusted to obtain the load demand voltage Mb, and the high-frequency noise in the voltage is removed using the low-dropout linear regulator voltage to obtain the stable voltage required by the final load.

Images