Dual-path power supply method for battery management unit and electronic device

Abstract

The present application provides a dual-power supply method and electronic device for a battery management unit, which is applied to an energy storage system. The system includes a battery cell, a charging unit, a battery management unit, a dual power supply unit (the first power supply unit is connected to an external power supply, and the second power supply unit is connected to the battery cell), a step-down unit, and a switch unit. The core of the method includes: the battery management unit collects battery operating information (temperature, voltage, current, attenuation rate) and uploads it to a server to obtain a predicted safe voltage; when there is no external power supply, the second power supply unit is enabled for power supply; when an external power supply is connected, the power supply is switched to the first power supply unit and the second power supply unit is shielded; when the external power supply is connected and the battery voltage is lower than the predicted safe voltage, the charging unit is started to charge the battery, and the first power supply unit is turned off at the same time, and the second power supply unit is enabled to maintain the operation of the BMS. Through the intelligent switching of dual power supplies and the dynamic determination of voltage thresholds, the continuous power supply of the battery management unit and the safe charging of the battery are ensured.

Description

Technical Field

[0001] The present application relates to the field of power supply technology, and in particular to a dual-path power supply method for a battery management unit and an electronic device. Background Art

[0002] Currently, conventional energy storage battery BMS systems utilize a dual power supply (12V or 24V) consisting of a battery pack (12V or 24V) and an auxiliary power supply (12V or 24V) within the charger. Under normal circumstances, energy storage batteries are always connected to the mains. This power supply approach has the following drawbacks: The battery pack and charger simultaneously provide 24V power, without optimization. This means each shares approximately 50% of the power supply, leading to frequent battery charging due to power consumption by the BMS. This reduces the battery's lifespan. Since the charger's auxiliary power supply must continuously power the BMS and cannot be disconnected, the charger remains in constant operation, shortening its lifespan. Summary of the Invention

[0003] The present application provides a dual-path power supply method for a battery management unit and an electronic device for optimizing the power supply of a BMS system.

[0004] In a first aspect, an embodiment of the present application provides a dual-path power supply method for a battery management unit, which is applied to an energy storage system. The energy storage system includes: a battery unit, a charging unit, a battery management unit, a first power supply unit, a second power supply unit, a step-down unit, and a switch unit. The first power supply unit and the second power supply unit are used to provide power to the battery management unit. The charging unit is connected to an external power supply and the battery unit, respectively. The first power supply unit is connected to the external power supply, and the second power supply unit is connected to the battery unit. The battery management unit controls the energy storage system as a whole through the switch unit. The method includes:

[0005] The battery management unit collects operating information of the battery unit and sends the operating information to a server to obtain a predicted safety voltage returned by the server, where the predicted safety voltage is determined based on the operating information;

[0006] When the external power supply is not connected, starting the second power supply unit to provide power to the battery management unit;

[0007] When the external power supply is connected, starting the first power supply unit to provide power to the battery management unit and shielding the second power supply unit;

[0008] When the external power supply is connected and the output voltage of the battery unit is lower than the predicted safety voltage, the charging unit is started to supply power to the battery unit, the first power supply unit is turned off, and the second power supply unit is started to provide power to the battery management unit.

[0009] In a second aspect, an embodiment of the present application provides an electronic device, which includes an energy storage system, and the energy storage system includes: a battery unit, a charging unit, a battery management unit, a first power supply unit, a second power supply unit, a step-down unit and a switching unit, wherein the first power supply unit and the second power supply unit are used to provide electrical energy to the battery management unit, the charging unit is respectively connected to an external power supply and the battery unit, the first power supply unit is connected to the external power supply, and the second power supply unit is connected to the battery unit, and the electronic device is used to execute the dual-path power supply method of the battery management unit as described in any one of the embodiments of the present application.

[0010] In a third aspect, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the processor enables the processor to implement a dual-path power supply method for a battery management unit as described in any one of the embodiments of the present application.

[0011] An embodiment of the present application provides a dual-path power supply method for a battery management unit, which is applied to an energy storage system. The energy storage system includes: a battery cell, a charging unit, a battery management unit, a first power supply unit, a second power supply unit, a step-down unit, and a switch unit. The first power supply unit and the second power supply unit are used to provide power to the battery management unit. The charging unit is connected to an external power supply and a battery cell, respectively. The first power supply unit is connected to the external power supply, and the second power supply unit is connected to the battery cell. The battery management unit controls the energy storage system as a whole through the switch unit. The method includes: the battery management unit collects working information of the battery cell and sends the working information to a server to obtain a predicted safety voltage returned by the server, where the predicted safety voltage is determined based on the working information; when the external power supply is not connected, the second power supply unit is started to provide power to the battery management unit; when the external power supply is connected, the first power supply unit is started to provide power to the battery management unit, and the second power supply unit is shielded; when the external power supply is connected and the output voltage of the battery cell is less than the predicted safety voltage, the charging unit is started to power the battery cell, the first power supply unit is turned off, and the second power supply unit is started to provide power to the battery management unit. Through the above method, after determining the predicted safety voltage, the first power supply unit or the second power supply unit is controlled to supply power to the battery management unit under different circumstances according to the predicted safety voltage, so as to adapt to the optimal power supply mode under the current circumstances, reduce the voltage loss of the battery pack, and extend the overall life of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the technical solutions of 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 some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0013] Figure 1 A schematic block diagram of an energy storage system provided in an embodiment of the present application;

[0014] Figure 2 A schematic flow chart of a dual-path power supply method for a battery management unit provided in an embodiment of the present application;

[0015] Figure 3 A circuit diagram of the energy storage system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0016] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0017] The flowcharts shown in the accompanying drawings are for illustrative purposes only and do not necessarily include all contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may be decomposed, combined, or partially merged, so the actual execution order may vary depending on the actual situation.

[0018] It should also be understood that the terms used in this specification are for the purpose of describing specific embodiments only and are not intended to limit the present application. As used in this specification and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms unless the context clearly indicates otherwise.

[0019] It should be further understood that the term "and / or" used in this specification and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0020] See also Figure 1 , Figure 1 This is a schematic block diagram of an energy storage system provided in an embodiment of the present application. Figure 1As shown, the energy storage system 100 includes a battery cell 11, a charging unit 12, a battery management unit 13, a first power supply unit 14, a second power supply unit 15, a step-down unit 16, and a switch unit. The external power supply 200 and the energy storage system 100 are connected via an air switch Km. The first power supply unit 14 and the second power supply unit 15 provide power to the battery management unit 13. The charging unit 12 is connected to the external power supply 200 and the battery cell 11, respectively. The first power supply unit 14 is connected to the external power supply 200, and the second power supply unit 15 is connected to the battery cell 11. The battery management unit 13 controls the energy storage system 100 as a whole via the switch unit. It should be noted that the switch unit, which is composed of multiple switches and is not shown in the figure, is provided between other units as needed. The battery management unit 13 controls the energy storage system 100 via the switch unit, for example, controlling the connection between the charging unit 12 and the external power supply 200, the connection between the external power supply 200 and the first power supply unit 14, and the connection between the battery cell 11 and the load 300.

[0021] See also Figure 2 , Figure 2 This is a schematic flow chart of a dual-path power supply method for a battery management unit provided in an embodiment of the present application. Figure 2 As shown, the specific steps of the dual-path power supply method of the battery management unit include: S101-S104.

[0022] S101 : The battery management unit 13 collects operating information of the battery unit 11 and sends the operating information to a server to obtain a predicted safety voltage returned by the server, where the predicted safety voltage is determined based on the operating information.

[0023] Exemplarily, the battery management unit 13 uses a temperature sensor (such as an NTC thermistor), a voltage sampling circuit (using a high-precision ADC chip), and a current Hall effect sensor integrated within the battery cell 11 to synchronously collect real-time data on battery temperature, discharge voltage, and discharge current with a sampling period of 10ms. The battery decay rate is calculated using the Coulomb counting method, integrating the current to calculate the capacity decay value for the current cycle. Combined with the battery cell 11's historical cycle count (stored in the battery management unit 13's EEPROM), the specific calculation formula is: Decay rate = Current capacity / Initial capacity × 100% Decay rate = Initial capacity / Current capacity × 100% The collected battery temperature, discharge voltage, discharge current, and decay rate data are then timestamped (based on the PTP clock synchronization protocol) to eliminate inter-sensor transmission delays, generating a multidimensional battery operating data set with a unified time dimension.

[0024] The battery management unit 13 uploads the multi-dimensional battery working data set to the cloud server via the CAN bus or Bluetooth 5.0 protocol. The data packet encapsulation format includes a header identifier (identifying the ID of the battery unit 11), a data segment (including temperature, voltage, current, decay rate and timestamp) and a CRC check code.

[0025] After receiving the data, the server performs the following operations: operating condition classification, feature extraction and fusion, and safe voltage prediction.

[0026] Working condition classification: Based on the amplitude of the discharge current, the data is divided into sub-datasets corresponding to different discharge rates (such as 0.5C, 1C, and 2C), and the voltage fluctuation curve and temperature drift curve at each rate are extracted.

[0027] Feature extraction and fusion: Kernel principal component analysis (KPCA) is performed on the voltage fluctuation curve to extract the first three principal component components as the voltage time-domain feature set. Wavelet packet decomposition (using the db4 wavelet basis) is performed on the temperature drift curve to generate the energy distribution of eight sub-bands as the temperature frequency-domain feature set. The two types of feature sets are cross-domain fused using the tensor folding algorithm to generate a composite feature matrix.

[0028] Safety voltage prediction: The composite feature matrix is ​​input into the pre-trained voltage safety prediction model (based on the LSTM-Transformer hybrid architecture), and the predicted safety voltage value is output through multi-dimensional regression analysis. The result is returned to the battery management unit 13 through the MQTT protocol.

[0029] After receiving the predicted safety voltage returned by the server, the battery management unit 13 compares it with the historical safety voltage threshold stored locally. If the deviation exceeds 5%, the data retransmission mechanism is triggered and the current working information is re-uploaded to the server for secondary calculation.

[0030] Among them, the server can be an independent server, a server cluster, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0031] S102 : When no external power source is connected, start the second power supply unit 15 to provide power to the battery management unit 13 .

[0032] Exemplarily, the second power supply unit 15 is used to transmit the electric energy of the battery unit 11 to the battery management unit 13 . The transmission loss of the second power supply unit 15 is extremely low, and no electric energy is wasted.

[0033] During the power supply period of the second power supply unit 15, the battery management unit 13 switches to low power consumption mode, turns off non-essential peripherals (such as the Bluetooth module), and only keeps the core control circuit (MCU and voltage / temperature sampling module) running, and the power consumption is reduced to below 10mA.

[0034] S103 , when an external power source is connected, the first power supply unit 14 is started to provide power to the battery management unit 13 , and the second power supply unit 15 is shielded.

[0035] For example, the first power supply unit 14 can use an external power source to provide power to the battery management unit 13. When the external power source is connected and the battery unit 11 is not being charged, the first power supply unit 14 provides power to the battery management unit 13, thereby reducing the loss of the battery unit 11. After the first power supply unit 14 is started, the output voltage of the first power supply unit 14 can block the output voltage of the second power supply unit 15, which is equivalent to turning off the second power supply unit 15. During this period, the second power supply unit 15 does not consume power from the battery unit 11.

[0036] S104 , when an external power source is connected and the output voltage of the battery unit 11 is lower than the predicted safety voltage, the charging unit 12 is started to supply power to the battery unit 11 , the first power supply unit 14 is turned off, and the second power supply unit 15 is started to provide power to the battery management unit 13 .

[0037] For example, during the process of charging the battery unit 11, the second power supply unit 15 can use the output voltage of the charging unit 12 to provide power to the battery management unit 13. Since the loss of the second power supply unit 15 is less than the loss of the first power supply unit 14, the first power supply unit 14 is turned off and the second power supply unit 15 is used to power the battery management unit 13.

[0038] An embodiment of the present application provides a dual-path power supply method for a battery management unit, which is applied to an energy storage system. The energy storage system includes: a battery cell, a charging unit, a battery management unit, a first power supply unit, a second power supply unit, a step-down unit, and a switch unit. The first power supply unit and the second power supply unit are used to provide power to the battery management unit. The charging unit is connected to an external power supply and a battery cell, respectively. The first power supply unit is connected to the external power supply, and the second power supply unit is connected to the battery cell. The battery management unit controls the energy storage system as a whole through the switch unit. The method includes: the battery management unit collects working information of the battery cell and sends the working information to a server to obtain a predicted safety voltage returned by the server, where the predicted safety voltage is determined based on the working information; when the external power supply is not connected, the second power supply unit is started to provide power to the battery management unit; when the external power supply is connected, the first power supply unit is started to provide power to the battery management unit, and the second power supply unit is shielded; when the external power supply is connected and the output voltage of the battery cell is less than the predicted safety voltage, the charging unit is started to power the battery cell, the first power supply unit is turned off, and the second power supply unit is started to provide power to the battery management unit. Through the above method, after determining the predicted safety voltage, the first power supply unit or the second power supply unit is controlled to supply power to the battery management unit under different circumstances according to the predicted safety voltage, so as to adapt to the optimal power supply mode under the current circumstances, reduce the voltage loss of the battery pack, and extend the overall life of the system.

[0039] In order to more clearly introduce the technical solution of the present application, the technical solution of the present application will be introduced through specific embodiments below. It should be noted that the specific embodiments are used to expand the technical solution of the present application, but are not intended to limit the present application.

[0040] In some embodiments, the working information includes: battery temperature, charge attenuation rate, discharge voltage and discharge current. When the server is used to determine the predicted safety voltage of the battery cell 11 based on the working information, it is specifically used to execute: S201-S206.

[0041] S201. Time-series alignment of battery temperature, charge decay rate, discharge voltage, and discharge current is performed to generate a multidimensional battery operating data set, and the multidimensional battery operating data set is classified by operating condition to obtain a first voltage fluctuation data set and a temperature drift data set under different discharge rates.

[0042] Exemplarily, timing alignment is achieved through a high-precision clock synchronization protocol (such as IEEE 1588 PTP), ensuring that data from the temperature sensor, voltage sampling circuit, current Hall effect sensor, and capacity calculation module have unified timestamps. The preprocessing module within the battery management unit 13 interpolates and compensates for inter-sensor transmission delays (e.g., a 50ms response lag in the temperature sensor). Using the current signal as a reference, the temperature and voltage data are remapped to the same time axis based on a time offset. The resulting multidimensional battery operating dataset is stored as a time series, containing four data channels (temperature, decay rate, voltage, and current) and nanosecond timestamps. Operating conditions are categorized based on the discharge current range. For example, 0-0.5C is defined as a low-rate operating condition, 0.5C-1.5C as a normal operating condition, and 1.5C and above as a high-rate operating condition. Each sub-operating condition dataset independently stores the corresponding voltage fluctuation curve (continuously sampled values ​​of voltage over time) and temperature drift curve (the trajectory of temperature change over discharge time).

[0043] The battery management unit 13 collects 2,000 sets of data points during the 10-minute discharge process. During the timing alignment phase, the temperature sensor data delay is compensated to synchronize with the current signal through the PTP protocol, forming a unified data set with a timestamp interval of 300ms. The operating condition classification module detects that the discharge current peak reaches 2.1C (corresponding to a current value of 4.2A) and automatically classifies the data in this period into the high-rate operating condition subset. It also extracts the continuous record of the voltage fluctuation from 3.8V to 3.5V in this interval as the first voltage fluctuation data set, and simultaneously records the temperature change trajectory from 25°C to 38°C as the temperature drift data set. The classified data is stored in CSV format, and each file header is marked with the discharge rate category and time range for subsequent feature extraction module calls.

[0044] S202 : Perform kernel principal component analysis on the first voltage fluctuation data set to extract a voltage time domain feature set, and perform wavelet packet decomposition on the temperature drift data set to generate a temperature frequency domain feature set.

[0045] For example, kernel principal component analysis (KPCA) uses a Gaussian kernel function to perform nonlinear dimensionality reduction on the time-domain features of a voltage fluctuation dataset. The kernel parameter σ is set to the standard deviation of the voltage fluctuation amplitude (e.g., 0.2V). By calculating the eigenvectors of the kernel matrix, the first three principal components (with a contribution rate >85%) are extracted as the voltage time-domain feature set, characterizing the trend, fluctuation amplitude, and mutation frequency of voltage changes. Wavelet packet decomposition of the temperature drift dataset uses the db4 wavelet basis function with four decomposition layers, generating 16 sub-bands (frequency range 0-0.5Hz). The energy contribution of each sub-band is calculated (energy value = sum of squared wavelet coefficients). The top four frequency bands (e.g., 0.1-0.2Hz and 0.3-0.4Hz) are selected as the temperature frequency-domain feature set, reflecting the periodicity and abnormal fluctuation patterns of temperature changes.

[0046] For a voltage fluctuation dataset (1000 sampling points) under high-rate operating conditions, the KPCA module calculated the Gaussian kernel matrix and extracted the first principal component (representing the overall voltage drop slope), the second principal component (reflecting the number of voltage dips), and the third principal component (describing the local fluctuation amplitude), forming a voltage time-domain feature set containing three-dimensional features. Wavelet packet decomposition of the temperature drift dataset revealed that the 0.15Hz frequency band accounted for 42% of the energy, indicating significant periodic temperature rise in this frequency band (possibly related to the chemical reaction cycle within the battery cell). The 0.35Hz frequency band had an abnormally high energy level (28%), suggesting the presence of sporadic thermal anomalies. These frequency-domain features, along with the voltage time-domain features, form the basis for subsequent cross-domain fusion.

[0047] S203 , performing cross-domain feature fusion on the voltage time domain feature set and the temperature frequency domain feature set to generate a composite feature matrix.

[0048] For example, cross-domain fusion consists of two stages: time axis alignment and feature concatenation. First, the dynamic time warping (DTW) algorithm is used to nonlinearly align voltage time domain features (300ms granularity) and temperature frequency domain features (1s granularity) to eliminate temporal misalignment caused by differences in sensor sampling frequencies. The aligned feature sequence is then segmented into time windows (e.g., 30-second windows). Within each window, the voltage time domain features (3D) and temperature frequency domain features (4D) are concatenated using tensors to form a 7-dimensional composite feature vector. These are then arranged in chronological order into an N×7 composite feature matrix (N is the number of time windows). The rows of the feature matrix represent time slices, while the columns contain cross-domain parameters such as voltage trends, fluctuations, mutations, temperature cycles, and energy in abnormal frequency bands.

[0049] During battery discharge, the voltage time-domain feature set consists of 3D data (10 minutes, 200 sampling points), and the temperature frequency-domain feature set consists of 4D data (10 minutes, 60 sampling points). After aligning the two to the same time axis using the DTW algorithm, the data is sliced ​​into 20 time slices using 30-second windows. Within each slice, the voltage features (e.g., [-0.3, 0.1, 0.05]) and the temperature frequency-domain features (e.g., [0.42, 0.28, 0.15, 0.05]) are concatenated into a 7-dimensional vector ([-0.3, 0.1, 0.05, 0.42, 0.28, 0.15, 0.05]), forming a 20×7 composite feature matrix. This matrix can be directly input into a machine learning model to correlate voltage anomalies with temperature frequency-domain changes.

[0050] S204: perform incremental feature dimension expansion on the composite feature matrix to obtain an enhanced security feature set.

[0051] For example, incremental feature dimensionality expansion employs a strategy combining an autoencoder with statistical feature stacking. First, a three-layer fully connected autoencoder (with an encoder hidden layer dimension of 5 and a decoder hidden layer dimension of 7) is used to perform unsupervised learning on the composite feature matrix, extracting latent features (5 dimensions) as supplementary dimensions. Second, statistics are calculated for the original composite feature matrix over time windows: mean (reflecting trend), standard deviation (fluctuation intensity), skewness (distribution asymmetry), and kurtosis (mutation concentration), generating 4-dimensional statistical features. Finally, the original 7-dimensional features, 5-dimensional latent features, and 4-dimensional statistical features are concatenated into a 16-dimensional enhanced security feature set, significantly improving feature expression capabilities.

[0052] For the 20×7 composite feature matrix obtained in S203, the autoencoder encodes each 7-dimensional vector to generate a 5-dimensional latent feature (e.g., [0.2, -0.1, 0.4, 0.05, -0.3]). It also calculates statistics such as the mean (e.g., [-0.2, 0.05, 0.1, 0.3, 0.2, 0.1, 0.05]) and standard deviation (e.g., [0.15, 0.08, 0.06, 0.12, 0.09, 0.04, 0.03]) of the 7-dimensional feature within each time window. The feature vector corresponding to each time slice is expanded to 16 dimensions (7+5+4), bringing the dimensionality of the enhanced security feature set to 20×16, encompassing the triple information of raw data, abstract representation, and statistical regularity.

[0053] S205 , respectively calculating the coupling coefficients of the charge decay rate and the discharge current in the enhanced safety feature set, dynamically assigning weights to the enhanced safety feature set based on the coupling coefficients, and generating a safety assessment feature vector.

[0054] For example, the coupling coefficient is calculated by the Pearson correlation coefficient to quantify the degree of linear correlation between the capacity decay rate and the discharge current. The specific process is: extract the decay rate sequence (such as the capacity retention rate of 100 cycles) and the corresponding discharge current mean sequence in the historical data from the enhanced safety feature set, and calculate the correlation coefficient r (range -1 to 1) between the two. The dynamic weight allocation adopts a piecewise function strategy: if |r|>0.7, it is considered to be strongly coupled, and the weight of the decay rate-related feature is increased by 30%; if 0.3<|r|≤0.7, the weight remains unchanged; if |r|≤0.3, the weight is reduced by 20%. The adjusted feature vector is processed by L2 normalization to ensure that the sum of the weights of each dimension is 1, forming a safety assessment feature vector.

[0055] In the historical data for a battery cell 11, the correlation coefficient between the capacity decay rate and the discharge current is r = -0.65 (a strong negative correlation, with high current accelerating capacity decay). Therefore, the weight of decay-related features (such as the capacity retention statistic) was increased from the initial 0.1 to 0.13, while the weight of discharge current-related features (such as the current mean) was reduced from 0.15 to 0.12. After dynamic adjustment, the dimension reflecting the decay-current coupling relationship among the 16-dimensional features was strengthened, and the generated safety assessment feature vector focused more on high-risk parameter combinations.

[0056] S206 , inputting the safety assessment feature vector into a pre-trained voltage safety prediction model for multi-dimensional regression analysis to obtain a predicted safety voltage of the battery unit 11 , wherein the voltage safety prediction model is trained by abnormal working data of the abnormal charging unit 12 .

[0057] For example, the voltage safety prediction model uses a hybrid LSTM-Transformer architecture. The LSTM layer (128 hidden units) captures temporal dependencies, while the Transformer encoder (with a four-head attention mechanism) extracts feature correlations across time steps. The model training data is derived from historical abnormal charging records, including datasets from scenarios such as overvoltage charging (voltage > 4.3V) and undervoltage discharge (voltage < 3.0V). The input safety assessment feature vector is encoded by the LSTM, and then an attention weight matrix is ​​used to focus on key time slices. A fully connected regression layer then outputs the predicted safe voltage value. The model is trained using the Adam optimizer with a learning rate of 0.001 and a smoothed L1 loss function. Training is repeated 500 times, with early stopping to prevent overfitting.

[0058] After inputting the safety assessment feature vector (20×16 dimensions) of the discharge process into the model, the LSTM layer extracts the implicit state of each time slice. The Transformer layer calculates the attention weights between different time slices (for example, the weight of the 5th to 10th minute slice reaches 60%). The regression layer then outputs a predicted safety voltage of 3.72V. This value is compared with the real-time monitored voltage. If the battery voltage falls below 3.72V and an external power supply is connected, the charging protection strategy is triggered. The model achieved a mean absolute error (MAE) of 0.03V on the test set, meeting engineering accuracy requirements.

[0059] In some embodiments, cross-domain feature fusion is performed on the voltage time-domain feature set and the temperature frequency-domain feature set to generate a composite feature matrix, including: S2031-S2035.

[0060] S2031 , aligning the voltage time-domain feature set and the temperature frequency-domain feature set in time dimension to generate a synchronized time-frequency feature sequence.

[0061] For example, time dimension alignment is achieved using the dynamic time warping (DTW) algorithm to address the time scale discrepancy between voltage time-domain features (high sampling rate) and temperature frequency-domain features (low sampling rate). Specifically, the voltage time-domain feature set is recorded with millisecond timestamps, while the temperature frequency-domain feature set has a time granularity of seconds due to the windowing of wavelet packet decomposition. The DTW algorithm constructs a cost matrix to find the optimal nonlinear matching path between the two feature sequences, stretching or compressing the time axis to eliminate time delay deviations. During the alignment process, redundant sampling points of the voltage time-domain features are interpolated and fitted, while sparse points of the temperature frequency-domain features are filled with cubic spline interpolation to generate a synchronized time-frequency feature sequence with strictly synchronized timestamps. In this synchronized time-frequency feature sequence, each time node contains both the voltage time-domain feature vector (such as trend component and fluctuation amplitude) and the temperature frequency-domain feature vector (such as energy value of a specific frequency band), forming multidimensional time series data and providing a temporal consistency foundation for cross-domain fusion.

[0062] S2032. Perform kernel matrix decomposition on the voltage time-domain features in the synchronized time-frequency feature sequence to extract the orthogonalized voltage principal component, perform frequency band energy normalization on the temperature frequency-domain features in the synchronized time-frequency feature sequence to generate a reduced-dimensional temperature band energy distribution set.

[0063] For example, kernel matrix decomposition of voltage time-domain features uses kernel principal component analysis (KPCA). A Gaussian kernel function is used to map the data to a high-dimensional space, and orthogonalized principal components are extracted through eigenvalue decomposition. During the KPCA process, kernel function parameters are adaptively adjusted based on the statistical distribution of voltage fluctuation amplitude to ensure effective separation of nonlinear features. The extracted principal components are sorted by contribution rate, and the top N components with cumulative contributions exceeding 90% are retained to eliminate redundant noise and reduce data dimensionality. Band energy normalization of temperature frequency-domain features normalizes the energy values ​​of each sub-band: first, the sum of the energy of each band within the time window is calculated, and then each band energy value is divided by the sum to obtain the normalized energy contribution. After normalization, high-frequency noise bands (e.g., bands with an energy contribution less than 1%) are automatically filtered out, retaining the energy distribution of key bands (e.g., those with an energy contribution greater than 5%), forming a reduced-dimensionality temperature band energy distribution set. This distribution set represents the core frequency-domain pattern of temperature variation in vector form, reducing data complexity while preserving key information.

[0064] S2033. Construct a three-dimensional tensor of the orthogonalized voltage principal component and the dimension-reduced temperature band energy distribution set along the time axis, fuse the time domain-frequency domain-energy dimension features through the tensor folding algorithm, and generate an initial composite feature tensor.

[0065] For example, a three-dimensional tensor is constructed with the time axis as the first dimension, the voltage principal component as the second dimension (time domain features), and the temperature band energy as the third dimension (frequency domain features). The outer product of the voltage principal component vector (e.g., 3D) and the temperature band energy vector (e.g., 4D) corresponding to each time slice generates a 3×4 two-dimensional matrix, representing the joint distribution of the time-frequency domain features at that moment. The two-dimensional matrices for all moments are stacked along the time axis to form a three-dimensional tensor structure of time × time domain × frequency domain. The tensor folding algorithm expands the three-dimensional tensor into a two-dimensional matrix based on the modalities. For example, the time and frequency domain dimensions are combined into the feature channel dimension, generating a two-dimensional representation of time × (time domain × frequency domain). This is then reconstructed into an initial composite feature tensor that retains the key features through low-rank approximation using singular value decomposition (SVD). This process fuses the time-domain fluctuation pattern with the frequency-domain energy distribution, establishing cross-dimensional correlations.

[0066] S2034. Perform dynamic channel weight allocation driven by a self-attention mechanism on the initial composite feature tensor to obtain cross-domain correlation.

[0067] Exemplarily, the self-attention mechanism dynamically assigns the importance of each channel by calculating similarity weights between feature channels. Specifically, each feature channel (a combination of time and frequency domains) of the initial composite feature tensor is used as the query vector, key vector, and value vector. The dot product similarity between the query and the key is calculated, and then softmax normalization is performed to generate an attention weight matrix. This weight matrix reflects the strength of cross-domain correlation between different time-domain features (such as voltage fluctuation trends) and frequency-domain features (such as temperature cyclical energy). To further enhance dynamics, a learnable scaling factor is introduced to adjust the distribution range of the attention weights and avoid the vanishing gradient problem. The cross-domain correlation is output as a weight matrix. The matrix element value represents the contribution of the corresponding feature channel to safe voltage prediction, with larger values ​​indicating stronger correlation.

[0068] S2035: Filter effective feature channels according to a preset correlation threshold and cross-domain correlation, and reconstruct and generate a composite feature matrix.

[0069] Exemplarily, the preset correlation threshold is set based on historical data statistics, for example, taking the top 20% quantile in the attention weight matrix as the threshold. During the screening process, all weight matrix elements are traversed, and feature channels with correlation exceeding the threshold are retained, while the remaining channels are set to zero to suppress noise interference. The valid feature channels are reorganized according to the original three-dimensional tensor structure, and the non-zero channel data are extracted along the time axis slice and rearranged into a two-dimensional matrix form (time × number of valid features). In the reorganized composite feature matrix, each row corresponds to a time slice, and each column is a high-correlation feature after cross-domain fusion, such as "the joint feature of the voltage principal component 1 and the temperature band energy 3." This matrix eliminates redundant information and focuses on key cross-domain correlation features, providing optimized input data for subsequent safe voltage prediction.

[0070] In some embodiments, the safety assessment feature vector is input into a pre-trained voltage safety prediction model for multi-dimensional regression analysis to obtain the predicted safety voltage of the battery cell 11, including: S2061-S2067.

[0071] S2061. Perform Gaussian kernel function mapping on the security assessment feature vector to generate a high-dimensional nonlinear feature space.

[0072] For example, the Gaussian kernel function maps low-dimensional safety assessment feature vectors to a high-dimensional space through nonlinear transformation to capture the complex correlations between features. Specifically, the standard deviation of each dimension in the safety assessment feature vector is used as the kernel parameter, and the similarity weights between any two feature vectors in the high-dimensional space are calculated to generate a kernel matrix. This matrix implicitly transforms the original features (such as voltage decay rate and temperature band energy) into points in the high-dimensional space, making linearly inseparable feature patterns (such as the correlation between voltage sag and high-frequency temperature fluctuations) linearly separable in the high-dimensional space. The resulting high-dimensional nonlinear feature space is typically several times the original dimension (e.g., 16 dimensions expanded to 256 dimensions), significantly improving the expressive power of subsequent models while preserving the topological structure of the original data.

[0073] S2062. Perform time-series slicing on the high-dimensional nonlinear feature space to obtain a multi-scale feature tensor sequence.

[0074] Exemplarily, time series slicing uses a sliding time window algorithm to segment the continuous high-dimensional feature space according to a preset window length (e.g., 30 seconds) and step size (e.g., 10 seconds). The window length setting needs to cover a typical charge and discharge cycle (e.g., the relaxation effect period of lithium batteries), and the step size ensures that adjacent windows partially overlap to avoid information loss. The feature points within each window are arranged in chronological order into a two-dimensional matrix (time step × high-dimensional feature), and then a multi-scale feature tensor sequence is generated through pyramid multi-scale sampling (e.g., using 15-second, 30-second, and 60-second windows simultaneously). This sequence contains feature distributions at different time granularities, which can capture both short-term fluctuation details and long-term trend changes, providing multi-level input for time series modeling.

[0075] S2063. The voltage decay features under the historical charge and discharge mode are extracted from the multi-scale feature tensor sequence through a bidirectional gated recurrent network, and the multi-head attention mechanism is used to calculate the association weights of different time steps in the voltage decay features to generate a dynamic attention enhanced feature set.

[0076] For example, a bidirectional gated recurrent network (BiGRU) processes multi-scale feature tensor sequences using two GRU units in the forward and backward passes, respectively, to capture the temporal dependencies of voltage decay features (e.g., the impact of discharge patterns over the previous hour on the current voltage). The GRU unit dynamically selects memory information through update and reset gates, outputting the latent state at each time step. A multi-head attention mechanism divides the latent state sequence output by the BiGRU into multiple subspaces (e.g., four heads), calculates attention weights between different time steps (e.g., the strength of the association between step t and step t-10), and generates a dynamic attention-enhanced feature set through weighted fusion. This feature set emphasizes the influence of critical time nodes (e.g., periods of voltage sag) and suppresses noise interference.

[0077] S2064. Input the dynamic attention enhanced feature set into the residual convolution module, and output a multi-level abstract feature map through the layer-by-layer residual connection in the residual convolution module.

[0078] For example, the residual convolution module consists of a cascade of multiple residual blocks, each of which includes a convolutional layer, a batch normalization layer, and a skip connection. The convolutional layer uses kernels of different sizes (e.g., 3×3, 5×5) to extract local features (e.g., voltage fluctuation segments). The batch normalization layer stabilizes the training process, and the skip connection adds the input features to the convolution output to prevent gradient vanishing. The dynamic attention-enhanced feature set is processed layer by layer, capturing detailed features (e.g., small voltage fluctuations) at shallow layers and abstract patterns (e.g., periodic attenuation trends) at deeper layers. The output multi-level abstract feature map fuses semantic information from different levels, providing multi-granularity representations for abnormal pattern matching.

[0079] S2065. Based on the preset abnormal voltage pattern library, perform abnormal pattern matching on the multi-level abstract feature map, calculate the feature similarity score between the multi-level abstract feature map and the preset pattern in the abnormal voltage pattern library, and dynamically adjust the weight distribution ratio of the voltage attenuation coefficient in the pre-trained voltage safety prediction model according to the feature similarity score.

[0080] For example, the abnormal voltage pattern library stores feature templates from historical abnormal operating conditions (such as voltage platform anomalies caused by overcharging and increased fluctuations caused by cell imbalance). Each template contains the statistical distribution of multiple levels of abstract features. Pattern matching uses a cosine similarity algorithm to calculate the similarity score between the current feature map and the template. A higher score indicates a closer match between the current state and a specific type of anomaly. The weight of the voltage attenuation coefficient is dynamically adjusted based on the score: if an overcharge template is matched, the penalty weight of the voltage attenuation coefficient is increased to mitigate the overcharging risk; if a cell imbalance template is matched, the weight is reduced to prioritize balancing issues. The weight adjustment takes effect in real time through the model's fully connected layer parameters, enabling adaptive prediction.

[0081] S2066. Perform multi-dimensional regression on the voltage safety prediction model after adjusting the weights to generate voltage safety boundary values ​​under different confidence levels.

[0082] For example, multi-dimensional regression employs a quantile regression algorithm, introducing multiple quantiles (e.g., 5%, 50%, and 95%) into the loss function to simultaneously predict voltage safety margins at different confidence levels. The model features, after weighting adjustments, are input, and the regression layer outputs three channels: a lower limit (conservative safety voltage), a median (most likely safety voltage), and an upper limit (optimistic safety voltage). During training, the quantile loss function is used to optimize the prediction accuracy of each channel, ensuring the statistical significance of the boundary values. An adaptive learning rate strategy is employed during the regression process, dynamically adjusting the parameter update step size based on the prediction error to balance convergence speed and stability.

[0083] S2067 , dynamically correct the voltage safety boundary value based on the current battery temperature and discharge rate of the battery unit 11 , and output a predicted safety voltage.

[0084] For example, dynamic correction incorporates a temperature-rate compensation coefficient table. This table experimentally calibrates the voltage offset at different temperature (e.g., 0°C-45°C) and discharge rate (e.g., 0.5°C-3°C) combinations. Based on the real-time collected battery temperature and discharge rate, the table is consulted to obtain the compensation coefficient, and the voltage safety boundary value obtained through regression is linearly interpolated and corrected. For example, in low-temperature, high-rate scenarios, the compensation coefficient is negative, lowering the safety voltage threshold to avoid over-discharge risks. The corrected boundary values ​​are weighted averaged (e.g., a 60% weight for the median and a 20% weight for the upper and lower limits) to generate a predicted safety voltage, ensuring that the output value conforms to statistical rules and adapts to real-time operating conditions.

[0085] See also Figure 3 , Figure 3 1 is a circuit diagram of an energy storage system 100 provided in an embodiment of the present application.

[0086] In some embodiments, the switch unit includes: a first switch group, a second switch group, a third switch group, and a fourth switch group.

[0087] The first switch group includes: a first switch K11, a second switch K12, a third switch K13, and a fourth switch K14. The second switch K12 and the third switch K13 are both used to connect an external power supply and the charging unit 12. The first switch K11, the second switch K12, and the third switch K13 are all normally open switches, and the fourth switch K14 is a normally closed switch.

[0088] The second switch group includes: a fifth switch K21, and the fifth switch K21 is a normally open switch.

[0089] The third switch group includes: a sixth switch K31, and the sixth switch K31 is a normally open switch.

[0090] The fourth switch group includes: a seventh switch K41, an eighth switch K42 and a ninth switch K43. The seventh switch K41 and the eighth switch K42 are both normally open switches, and the ninth switch K43 is a normally closed switch.

[0091] In some embodiments, the first power supply unit 14 includes a transformer T1 , a rectifier component DB1 , a first filter capacitor C1 , and a first diode D1 .

[0092] The first end of the transformer T1 is used to connect to the first end of the external power supply, the second end of the transformer T1 is connected to the external power supply through the fourth switch K14 and the seventh switch K41, the third end of the transformer T1 and the fourth end of the transformer T1 are both connected to the input end of the rectifier component DB1, the output positive electrode of the rectifier component DB1 is connected to the first end of the first filter capacitor C1 and the anode of the first diode D1, the output negative electrode of the rectifier component DB1 is connected to the first output end of the battery unit 11, the second end of the first filter capacitor C1, and the first input end of the step-down unit 16, the cathode of the first diode D1 is connected to the second input end of the step-down unit 16, and the output end of the step-down unit 16 is connected to the battery management unit 13.

[0093] In some embodiments, the second power supply unit 15 includes: a fuse, a tenth switch S2 , an eleventh switch S3 , and a second diode D2 . The tenth switch S2 is a normally closed switch, and the eleventh switch S3 is a normally open switch.

[0094] The first end of the FUSE component is connected to the second output end of the battery unit 11, the second end of the FUSE component is connected to the anode of the second diode D2 through the tenth switch S2 and the eleventh switch S3, the cathode of the second diode D2 is connected to the second input end of the step-down unit 16, and the two ends of the eleventh switch S3 are respectively connected to the two ends of the sixth switch K31.

[0095] In some embodiments, as Figure 3 As shown, the energy storage system 100 further includes an insulation detection unit 17. A first end of the insulation detection unit 17 is connected to the first output end of the battery unit 11, a second end of the insulation detection unit 17 is connected to the second output end of the battery unit 11, and a third end of the insulation detection unit 17 is connected to the battery management unit 13.

[0096] In some embodiments, the second switch group further includes: a twelfth switch S4. The first end of the twelfth switch S4 is connected to a preset voltage of 24V+. The second end of the twelfth switch S4 is connected to the control end of the fifth switch K21, and the control end of the fifth switch K21 is the control signal input end of the relay. The control end of the fifth switch K21 is also connected to the first end of the ninth switch K43, and the second end of the ninth switch K43 is connected to the battery management unit 13. The first end of the fifth switch K21 is connected to the second end of the FUSE component, the second end of the fifth switch K21 is connected to one end of the load, and the other end of the load is connected to the first end of the battery cell 11. Under normal discharge conditions, the battery management unit 13 controls the fifth switch K21 to close to control the battery cell 11 to provide power to the load. In the case of low battery, the battery management unit 13 is in a dormant state, and power is guaranteed in an emergency by continuously pressing the twelfth switch S4.

[0097] When the eleventh switch S3 is pressed, the battery pack voltage is transmitted through the tenth switch S2, the eleventh switch S3, and the second diode D2 to the step-down unit 16, which then outputs the driving voltage (12V or 24V) for the battery management unit 13. The battery management unit 13 receives power from the second power supply unit 15 and enters the monitoring start state. When the output voltage of the battery cell 11 is above the set sleep voltage, the insulation detection unit 17 detects whether the battery cell 11 is leaking. If no leakage is detected, the battery management unit 13 sends a drive signal to the relay of the sixth switch K31, closing it and the energy storage system 100 enters normal operation. Simultaneously, a drive signal is sent to the relay of the fifth switch K21, closing it and enabling power to be supplied to the load.

[0098] When connected to an external power source (220V AC), the air switch Km switches to the closed state. If the seventh switch K41 is not switched to the closed state, the first power supply unit 14 of the energy storage system 100 will not operate. If the eighth switch K42 is not switched to the closed state, the relay of the fourth switch K14 will not operate, and the battery cell 11 will not enter automatic charging. When the output voltage of the battery cell 11 drops to the first output warning voltage, the battery management unit 13 will issue a low voltage warning. After a delay in discharging the battery cell 11 for a period of time, the battery management unit 13 controls the fifth switch K21 to open, stopping the battery cell 11 from discharging. To enable emergency discharge of the battery cell 11 when the battery is low on charge, press and hold the twelfth switch to enable emergency discharge. When the output voltage of the battery management unit 13 drops again to the second output warning voltage, the battery management unit 13 controls the sixth switch K31 to open, depriving the battery management unit 13 of power and preventing further discharge of the battery cell 11.

[0099] If an external 220V power supply is connected, the air breaker is switched to closed, the charging switch S1 is switched to closed, and S1-1 is closed, the first power supply unit 14 enters operation, and the battery management unit 13 receives power from the first power supply unit 14. The voltage at the cathode of the first diode D1 is 10V to 20V higher than the voltage at the cathode of the second diode D2, causing the second diode D2 to be cut off (the second power supply unit 15 stops supplying power and no longer consumes power from the battery cell 11). The input to the step-down unit 16 is then entirely provided by the first power supply unit 14. When it is detected that the output voltage of the battery cell 11 has dropped below the first output warning voltage, the battery management unit 13 controls the first switch K11, the second switch K12, and the third switch K13 to close, causing the charging unit 12 to enter operation and charge the battery cell 11. Simultaneously, the fourth switch K14 is opened, disconnecting the first power supply unit 14 and switching to the second power supply unit 15 to supply power to the battery management unit 13.

[0100] When the battery cell 11 is fully charged, the battery management unit 13 controls the first switch K11, the second switch K12 and the third switch K13 to be disconnected, and the charging unit 12 stops working (to prevent the charging unit 12 from being online for a long time and affecting its life). At the same time, the fourth switch K14 is closed, the first power supply unit 14 resumes power supply, and the second power supply unit 15 stops supplying power, and no longer consumes the power of the battery cell 11. In the standby state, the battery cell 11 only has self-consumption of power, and the power decreases slowly. In the no-load discharge state of the battery cell 11, it will not enter frequent charging due to the self-consumption of power of the battery cell 11, which can extend the service life of the battery cell 11.

[0101] An embodiment of the present application provides an electronic device, which includes a memory and a processor; the memory is used to store computer programs; the processor is used to execute the computer programs and implement a dual-path power supply method for a battery management unit such as any one of the embodiments of the present application when executing the computer programs.

[0102] An embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the processor implements a dual-path power supply method for a battery management unit as described in any one of the embodiments of the present application.

[0103] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present application, and such modifications or substitutions should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A dual-circuit power supply method for a battery management unit, characterized in that: Applied to an energy storage system, the energy storage system includes: a battery unit, a charging unit, a battery management unit, a first power supply unit, a second power supply unit, a step-down unit, and a switch unit, the first power supply unit and the second power supply unit are used to provide power to the battery management unit, the charging unit is connected to an external power supply and the battery unit, respectively, the first power supply unit is connected to the external power supply, the second power supply unit is connected to the battery unit, the battery management unit controls the energy storage system as a whole through the switch unit, and the method includes: The battery management unit collects the working information of the battery cell and sends the working information to the server to obtain the predicted safety voltage returned by the server, wherein the predicted safety voltage is determined based on the working information, and the working information includes: battery temperature, power attenuation rate, discharge voltage and discharge current. When the server is used to determine the predicted safety voltage of the battery cell based on the working information, it is specifically used to perform: timing alignment of the battery temperature, the power attenuation rate, the discharge voltage and the discharge current to generate a multidimensional battery working data set, and classify the working conditions of the multidimensional battery working data set to obtain a first voltage fluctuation data set and a temperature drift data set under different discharge rates; perform kernel principal component analysis on the first voltage fluctuation data set, Extracting a voltage time-domain feature set, performing wavelet packet decomposition on the temperature drift data set to generate a temperature frequency-domain feature set; performing cross-domain feature fusion on the voltage time-domain feature set and the temperature frequency-domain feature set to generate a composite feature matrix; performing incremental feature dimension expansion on the composite feature matrix to obtain an enhanced safety feature set; calculating the coupling coefficients of the charge decay rate and the discharge current in the enhanced safety feature set respectively, and dynamically assigning weights to the enhanced safety feature set based on the coupling coefficients to generate a safety assessment feature vector; inputting the safety assessment feature vector into a pre-trained voltage safety prediction model for multi-dimensional regression analysis to obtain the predicted safety voltage, wherein the voltage safety prediction model is obtained by training abnormal working data of an abnormal charging unit; When the external power supply is not connected, starting the second power supply unit to provide power to the battery management unit; When the external power supply is connected, starting the first power supply unit to provide power to the battery management unit and shielding the second power supply unit; When the external power supply is connected and the output voltage of the battery unit is lower than the predicted safety voltage, the charging unit is started to supply power to the battery unit, the first power supply unit is turned off, and the second power supply unit is started to provide power to the battery management unit.

2. The dual-path power supply method for a battery management unit according to claim 1, wherein: The cross-domain feature fusion of the voltage time domain feature set and the temperature frequency domain feature set to generate a composite feature matrix includes: Aligning the voltage time-domain feature set and the temperature frequency-domain feature set in time dimension to generate a synchronized time-frequency feature sequence; performing kernel matrix decomposition on the voltage time-domain features in the synchronized time-frequency feature sequence to extract orthogonalized voltage principal components, performing frequency band energy normalization processing on the temperature frequency-domain features in the synchronized time-frequency feature sequence to generate a dimensionally reduced temperature band energy distribution set; Constructing a three-dimensional tensor of the orthogonalized voltage principal component and the dimensionally reduced temperature band energy distribution set along the time axis, and fusing the time-domain-frequency-domain-energy dimensional features through a tensor folding algorithm to generate an initial composite feature tensor; Performing dynamic channel weight allocation driven by a self-attention mechanism on the initial composite feature tensor to obtain cross-domain correlation; Valid feature channels are screened according to a preset correlation threshold and the cross-domain correlation, and a composite feature matrix is ​​recombined and generated.

3. The dual-path power supply method for a battery management unit according to claim 1, wherein: Inputting the safety assessment feature vector into a pre-trained voltage safety prediction model for multi-dimensional regression analysis to obtain the predicted safety voltage of the battery cell includes: Performing Gaussian kernel function mapping on the security assessment feature vector to generate a high-dimensional nonlinear feature space; Performing time-series slicing on the high-dimensional nonlinear feature space to obtain a multi-scale feature tensor sequence; Extracting voltage decay features under historical charge and discharge modes from the multi-scale feature tensor sequence using a bidirectional gated recurrent network, and using a multi-head attention mechanism to calculate the association weights of different time steps in the voltage decay features to generate a dynamic attention enhanced feature set; Inputting the dynamic attention enhanced feature set into the residual convolution module, and outputting a multi-level abstract feature map through the layer-by-layer residual connection in the residual convolution module; Based on a preset abnormal voltage pattern library, abnormal pattern matching is performed on the multi-level abstract feature graph, a feature similarity score between the multi-level abstract feature graph and a preset pattern in the abnormal voltage pattern library is calculated, and a weight distribution ratio of the voltage attenuation coefficient in the pre-trained voltage safety prediction model is dynamically adjusted according to the feature similarity score; Perform multi-dimensional regression on the voltage safety prediction model after weight adjustment to generate voltage safety boundary values ​​at different confidence levels; The voltage safety boundary value is dynamically corrected based on the current battery temperature and discharge rate of the battery unit, and a predicted safety voltage is output.

4. The dual-path power supply method for a battery management unit according to claim 1, wherein: The switch unit includes: a first switch group, a second switch group, a third switch group and a fourth switch group; The first switch group includes: a first switch (K11), a second switch (K12), a third switch (K13), and a fourth switch (K14); the second switch (K12) and the third switch (K13) are both used to connect the external power supply and the charging unit; the first switch (K11), the second switch (K12), and the third switch (K13) are all normally open switches, and the fourth switch (K14) is a normally closed switch; The second switch group includes: a fifth switch (K21), the fifth switch (K21) being a normally open switch; The third switch group includes: a sixth switch (K31), the sixth switch (K31) being a normally open switch; The fourth switch group includes: a seventh switch (K41), an eighth switch (K42) and a ninth switch (K43); the seventh switch (K41) and the eighth switch (K42) are both normally open switches, and the ninth switch (K43) is a normally closed switch.

5. The dual-path power supply method for a battery management unit according to claim 4, wherein: The first power supply unit comprises: a transformer (T1), a rectifier component (DB1), a first filter capacitor (C1) and a first diode (D1); The first end of the transformer (T1) is used to be connected to the first end of the external power supply, the second end of the transformer (T1) is connected to the external power supply through the fourth switch (K14) and the seventh switch (K41), the third end of the transformer (T1) and the fourth end of the transformer (T1) are both connected to the input end of the rectifier component (DB1), the output positive electrode of the rectifier component (DB1) is connected to the first end of the first filter capacitor (C1) and the anode of the first diode (D1), the output negative electrode of the rectifier component (DB1) is connected to the first output end of the battery unit, the second end of the first filter capacitor (C1), and the first input end of the step-down unit, the cathode of the first diode (D1) is connected to the second input end of the step-down unit, and the output end of the step-down unit is connected to the battery management unit.

6. The dual-path power supply method for a battery management unit according to claim 5, wherein: The second power supply unit comprises: a fuse, a tenth switch (S2), an eleventh switch (S3) and a second diode (D2), the tenth switch (S2) is a normally closed switch, and the eleventh switch (S3) is a normally open switch; The first end of the FUSE component is connected to the second output end of the battery unit, the second end of the FUSE component is connected to the anode of the second diode (D2) through the tenth switch (S2) and the eleventh switch (S3), the cathode of the second diode (D2) is connected to the second input end of the step-down unit, and the two ends of the eleventh switch (S3) are respectively connected to the two ends of the sixth switch (K31).

7. The dual-path power supply method for a battery management unit according to claim 6, wherein: The second switch group further includes: a twelfth switch (S4), a first end of the twelfth switch (S4) is connected to a preset voltage, a second end of the twelfth switch (S4) is connected to the control end of the fifth switch (K21), the control end of the fifth switch (K21) is a control signal input end of the relay, the control end of the fifth switch (K21) is also connected to the first end of the ninth switch (K43), the second end of the ninth switch (K43) is connected to the battery management unit, the first end of the fifth switch (K21) is connected to the second end of the FUSE component, the second end of the fifth switch (K21) is connected to one end of the load, and the other end of the load is connected to the first end of the battery unit.

8. The dual-path power supply method for a battery management unit according to claim 6, wherein: The energy storage system further includes: An insulation detection unit, wherein a first end of the insulation detection unit is connected to the first output end of the battery unit, a second end of the insulation detection unit is connected to the second output end of the battery unit, and a third end of the insulation detection unit is connected to the battery management unit.

9. An electronic device, characterized in that: The electronic device includes an energy storage system, which includes: a battery unit, a charging unit, a battery management unit, a first power supply unit, a second power supply unit, a step-down unit and a switch unit. The first power supply unit and the second power supply unit are used to provide electrical energy to the battery management unit. The charging unit is respectively connected to an external power supply and the battery unit. The first power supply unit is connected to the external power supply, and the second power supply unit is connected to the battery unit. The electronic device is used to execute the dual-path power supply method of the battery management unit as described in any one of claims 1 to 8.

Images