A Battery SOC Estimation Method and System Based on Attention Mechanism and Weighted Fusion of Multi-Source Data
By using an attention mechanism and a multi-source data weighted fusion method, the feature weights in the battery SOC estimation model are dynamically adjusted, which solves the problems of bias and response lag in SOC estimation under different operating conditions in traditional methods, and achieves more accurate and faster SOC estimation.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHINA TOWER CO LTD
- Filing Date
- 2025-10-30
- Publication Date
- 2026-05-26
AI Technical Summary
In the existing technology, the battery state of charge (SOC) estimation method lacks the ability to dynamically evaluate the importance of features, which leads to the inability to dynamically adjust feature weights during the charging cut-off stage, high temperature environment and fault conditions, resulting in SOC estimation bias and response lag.
We employ an attention-based and multi-source data weighted fusion approach, which collects battery data from multiple sensors, combines operating condition determination and a deep learning model, dynamically adjusts the weights of voltage, temperature, and current features, and uses the attention mechanism to strengthen key features to achieve accurate SOC estimation.
It improves the accuracy of SOC estimation during the charging phase, reduces the misjudgment of thermal runaway risk under high temperature conditions, enables rapid response to fault conditions, enhances the model's adaptability and overall robustness, and reduces SOC estimation errors and response time.
Smart Images
Figure CN122085115A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of battery state of charge (SOC) assessment technology, specifically to a battery SOC estimation method and system based on attention mechanism and multi-source data weighted fusion. Background Technology
[0002] Currently, fixed-weight or rule-based feature weighting methods are widely used for battery state of charge (SOC) estimation. These methods apply the same treatment to features such as voltage, temperature, and current under all operating conditions, or rely on manually preset static weight allocation strategies. However, battery systems operate under complex and variable conditions in real-world scenarios, and this fixed-weight approach leads to three main problems: First, it cannot automatically increase the weight of voltage features during the charging cutoff phase, easily causing SOC estimation errors; second, it cannot dynamically enhance the attention given to temperature-related features in high-temperature environments, making it difficult to accurately reflect the impact of thermal runaway risk on SOC; and third, it cannot quickly adjust the weight of current features when faults such as overcurrent occur, resulting in a delayed response to abnormal operating conditions. The root cause of these problems is that traditional methods lack the ability to dynamically assess the importance of features and cannot automatically optimize feature weight allocation based on real-time operating conditions. Summary of the Invention
[0003] The purpose of this invention is to overcome the shortcomings of the prior art and provide a battery SOC estimation method and system based on attention mechanism and multi-source data weighted fusion.
[0004] The objective of this invention is achieved through the following technical solution: In a first aspect, this invention discloses a battery SOC estimation method based on attention mechanism and multi-source data weighted fusion, comprising: S1. Multi-source data acquisition and operating condition determination: Continuously acquire multi-source data of the battery through multiple sensors, including the highest voltage of a single cell, total voltage, MOS temperature, ambient temperature and current, and then determine the current operating condition of the battery. S2, Working condition-driven attention weighted processing, which weights the features corresponding to the working condition based on the working condition determination. S3. Feature enhancement and model fusion prediction: First, selectively enhance the weighted features based on the ambient temperature. Then, input the processed features into a deep learning model with an embedded attention module and output the weighted SOC prediction value to achieve accurate SOC estimation.
[0005] Based on the first aspect, step S1 specifically includes the following steps: S11. Continuously collect multi-source data on battery operation using multiple sensors at a fixed sampling frequency, including collecting the highest voltage of individual cells using a voltage sensor. and total voltage The temperature of the MOS is collected by a temperature sensor. and ambient temperature Current is collected through a current sensor Then, the analog signal output by the sensor is converted by the ADC. Perform analog-to-digital conversion to convert the data into digital data for input into the system. ,in This represents the data collected by different types of sensors; S12, Working condition determination logic, which determines the working condition based on the collected data and preset rules.
[0006] Based on the first aspect, the operating conditions described in step S12 include charging operating conditions, high temperature operating conditions, and fault operating conditions. The charging condition is as follows: a charging control signal is detected and the current sensor detects current in the charging direction, i.e. and and ;in This indicates the enable command sent by the charging station. Indicates the first duration; The high-temperature operating condition is: the ambient temperature sensor continuously... Each sampling period detected The interval between each cycle ,Right now ; The fault condition is as follows: the overcurrent protection module detects current. Exceeding the preset threshold And the second duration Overcurrent trigger protection mechanism; that is and and ,in This is a fault status indicator.
[0007] Based on the first aspect, step S2 specifically includes the following steps: S21. Weighted charging conditions are used to construct a feature weight allocation model, focusing on the highest voltage of individual cells. and total voltage Assign high weight, where high weight is percentage. Let the eigenvectors be... Weight vector Where T represents transpose. express The weight, express The weight, express The weight, express The weight, express The weights are adjusted during charging. , ; in As the weighted increment, the weighted voltage characteristics are used in conjunction with the charging cutoff point model to make a judgment: when With battery full charge voltage threshold The absolute value of the difference is less than the voltage threshold. ,and The absolute value of the first derivative is less than the voltage change rate threshold. ,Right now and Identify signs that indicate charging is about to stop; S22. High-temperature operating condition weighting: Based on the thermal runaway risk correlation model, calculate the influence coefficient of temperature characteristics on SOC estimation. The weights are adjusted under high-temperature conditions: ; ; in Indicates weight Adjusted weights Indicates weight The adjusted weights are used to correct the SOC estimation bias using weighted temperature characteristics, and a temperature compensation factor is introduced. Adjustments to the SOC estimation formula: ,in This represents the original SOC estimate. This represents the SOC correction value after temperature compensation; S23, Fault condition weighting, enhanced current. Feature weights, weights Make adjustments ,in Indicates weight Adjusted weights This represents the increment of the fault weight; then the weight is reduced proportionally. ;in Indicates weight The weights after reduction, among which Indicates weight The weights after reduction, among which Indicates weight The weights after reduction, among which Indicates weight The weights after reduction, among which Indicates weight The weights are reduced; the high-weighted current characteristics are utilized in conjunction with an abnormal discharge detection algorithm to monitor the slope of current abrupt changes. , ,in This represents the change in current. Indicates the change over time; when At the same time, combined with the battery internal resistance model , This indicates the voltage change, assesses the impact of abnormal discharge on the State of Charge (SOC), and corrects the SOC estimate. : ,in For the battery's rated capacity, Integral of the discharge capacity, SOC estimate The corrected value.
[0008] Based on the first aspect, step S3 specifically includes the following steps: S31. Enhanced feature selectivity based on ambient temperature. A temperature-feature enhancement mapping table is constructed; based on a preset temperature threshold, the temperature is divided into a first temperature range and a second temperature range, each corresponding to a different enhancement strategy, to enhance the features. In the first temperature range, the enhancement factor is used. For MOS temperature and ambient temperature Perform secondary enhancement, that is , ,in Indicates MOS temperature The value after secondary enhancement Indicates ambient temperature The value after secondary enhancement; In the second temperature range, the input voltage and current characteristics are synergistically enhanced, and principal component analysis (PCA) is used for dimensionality reduction to extract the pre-temperature characteristics. Each principal component, and then through the enhancement coefficient Strengthen the principal components; that is , Where X represents the input feature, This represents the value of the feature after dimensionality reduction using Principal Component Analysis (PCA). Indicates the enhanced features; S32, Model Embedding and SOC Prediction: An attention module is embedded into a deep learning model, and then the enhanced features are... The input is fed into the model, and the importance of features is dynamically calculated through the Query-Key-Value mechanism. The weighted SOC prediction value is then output to achieve accurate SOC estimation.
[0009] Based on the first aspect, step S32 specifically includes the following steps: S321, Enhanced features Split into sub-vectors Where Q is the query vector, K is the key vector, and V is the value vector, obtained through the formula... Calculate the similarity matrix S, normalize the similarity using the Softmax function, and then calculate the attention weights A. Finally, through the formula Calculate weighted features ; S322, SOC prediction output, model weighted features Forward propagation, the output layer uses linear regression, and the loss function is... Mean Square Error (MSE): Where N represents the number of samples, Indicates the predicted value. Representing the true value, setting up a model self-correction mechanism based on the gradient of the loss function. The Adam optimizer is used to adjust model parameters and continuously improve the SOC prediction accuracy. ,in This represents the model parameters before adjustment. This represents the adjusted model parameters. This represents the learning rate.
[0010] Secondly, this invention discloses a battery SOC estimation system based on attention mechanism and multi-source data weighted fusion, used in the aforementioned battery SOC estimation method based on attention mechanism and multi-source data weighted fusion, comprising: The sensor module is used to collect multi-source data from the battery in real time. The working condition identification module is used to identify and determine the working condition based on sensor data and logic algorithms; The load condition weighting module is used to apply corresponding weights based on the type of load condition. The feature selective enhancement module is used to selectively enhance the weighted features based on the ambient temperature. The model embedding and SOC prediction module is used to input the processed features into the deep learning model embedded with the attention module, and output the weighted SOC prediction value to achieve accurate SOC estimation.
[0011] Based on the second aspect, the operating condition weighting module includes a charging characteristic weighting submodule, a high temperature characteristic weighting submodule, and a fault characteristic weighting submodule. The charging feature weighting submodule is used to weight the voltage and total voltage after entering the charging condition. The high-temperature characteristic weighting submodule is used to weight the MOS temperature and ambient temperature after entering the high-temperature operating condition. The fault feature weighting submodule is used to weight the current after entering a fault condition.
[0012] The beneficial effects of this invention are: 1) This invention automatically increases the weight of "maximum single-cell voltage" and "total voltage" to over 40% during charging. By strengthening the proportion of these two core features in the model through an attention mechanism, it improves the accuracy of SOC estimation during the charging phase and reduces misjudgment of the charging cutoff point. This solves the problems of existing technologies that use fixed weights for voltage features during the charging process, which cannot specifically increase the influence of "maximum single-cell voltage" and "total voltage" during the charging cutoff phase (such as when the battery is close to full charge), and are prone to SOC estimation deviation due to the dilution of voltage features by other data.
[0013] 2) This invention enhances the weight of "MOS temperature" and "ambient temperature" under high-temperature conditions through an attention mechanism, optimizes the accuracy of SOC estimation under high-temperature conditions, and reduces misjudgment of thermal runaway risk. It solves the problems of existing technologies using fixed weights for temperature characteristics under high-temperature conditions, which makes it difficult to reflect the strong correlation between "MOS temperature" and "ambient temperature" and thermal runaway risk, resulting in SOC estimation that cannot adapt to the impact of temperature on battery characteristics.
[0014] 3) This invention rapidly enhances the current weight through an attention mechanism under fault conditions, accelerates the response speed under fault conditions, and improves the efficiency of abnormal discharge identification. It solves the problems of existing technologies relying on static weights under fault conditions such as overcurrent, which cannot quickly increase the attention to current characteristics, resulting in a lag in the identification of abnormal discharge behavior and affecting the timeliness of battery safety protection.
[0015] 4) This invention dynamically calculates feature importance through the Query-Key-Value mechanism of the attention mechanism, realizing the transformation from "static weights" to "condition-driven dynamic weights", enhancing the model's adaptability to complex conditions, improving overall robustness, and solving the problems of existing technologies' fixed weights or simple rule weighting being unable to adapt to the dynamic switching of multiple conditions such as "charging-discharging-high temperature-fault" during battery operation, resulting in large fluctuations in SOC estimation accuracy under different scenarios. Attached Figure Description
[0016] Figure 1 This is a schematic diagram illustrating the steps of the battery SOC estimation method based on attention mechanism and multi-source data weighted fusion according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the battery SOC estimation system based on attention mechanism and multi-source data weighted fusion according to an embodiment of the present invention. Detailed Implementation
[0017] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0018] This invention discloses a battery SOC estimation method and system based on an attention mechanism and multi-source data weighted fusion. It addresses the problems of traditional models in equally processing multi-source data and failing to adapt to differences in operating conditions when estimating battery State of Charge (SOC). By introducing an attention mechanism, it dynamically allocates weights for data features such as voltage, temperature, and current based on real-time operating conditions such as charging, high temperature, and faults. Combined with feature enhancement and deep model embedding, it achieves accurate SOC estimation. A schematic diagram of the method's steps is shown below. Figure 1 As shown, the method includes several main steps such as multi-source data acquisition, working condition determination, attention weighting, feature enhancement, and model prediction. Specifically, the method includes: S1. Multi-source data acquisition and operating condition determination: Continuously acquire multi-source data of the battery through multiple sensors, including the highest voltage of a single cell, total voltage, MOS temperature, ambient temperature and current, and then determine the current operating condition of the battery. S2, Working condition-driven attention weighted processing, which weights the features corresponding to the working condition based on the working condition determination. S3. Feature enhancement and model fusion prediction: First, selectively enhance the weighted features based on the ambient temperature. Then, input the processed features into a deep learning model with an embedded attention module and output the weighted SOC prediction value to achieve accurate SOC estimation.
[0019] For example, step S1 specifically includes the following steps: S11. Continuously collect multi-source data on battery operation using multiple sensors at a fixed sampling frequency, including collecting the highest voltage of individual cells using a voltage sensor. and total voltage The temperature of the MOS is collected by a temperature sensor. and ambient temperature Current is collected through a current sensor Then, the analog signal output by the sensor is converted by the ADC. Perform analog-to-digital conversion to convert the data into digital data for input into the system. ,in This represents the data collected by different types of sensors, for example, when Time correspondence , Time correspondence wait; S12, Working condition determination logic, which determines the working condition based on the collected data and preset rules.
[0020] For example, the operating conditions described in step S12 include charging operating conditions, high temperature operating conditions, and fault operating conditions. The charging condition is as follows: a charging control signal is detected and the current sensor detects current in the charging direction, i.e. and and ;in This indicates the enable command sent by the charging station. Indicates the first duration; The high-temperature operating condition is: the ambient temperature sensor continuously... Each sampling period detected The interval between each cycle ,Right now ; The fault condition is as follows: the overcurrent protection module detects current. Exceeding the preset threshold And the second duration Overcurrent trigger protection mechanism; that is and and ,in This is a fault status indicator.
[0021] For example, step S2 specifically includes the following steps: S21. Weighted charging conditions are used to construct a feature weight allocation model, focusing on the highest voltage of individual cells. and total voltage Assign high weight, where high weight is percentage. Let the eigenvectors be... Weight vector Where T represents transpose. express The weight, express The weight, express The weight, express The weight, express The weights are adjusted during charging. , ; in For example, weight increment ,make sure Using the weighted voltage characteristics, combined with the charging cutoff point model, a judgment is made: when With battery full charge voltage threshold The absolute value of the difference is less than the voltage threshold. ,and The absolute value of the first derivative is less than the voltage change rate threshold. ,Right now and In this embodiment, the voltage threshold is... The value is set to 0.01V, and the voltage change rate threshold is set. The value is set to 0.001V / s to identify the signs of charging cutoff. The determination of the charging cutoff point is highly dependent on the voltage change trend (such as the voltage growth rate slowing down or approaching the full charge threshold). The high weight makes the model more focused on the subtle changes in voltage characteristics, thereby more accurately identifying the charging cutoff point and reducing the SOC estimation error. S22. High-temperature operating condition weighting: Based on the thermal runaway risk correlation model, calculate the influence coefficient of temperature characteristics on SOC estimation. , Thermal runaway risk model Determining, for example, by fitting a function experimentally, when , The weight increases linearly with increasing temperature; the weight is adjusted under high-temperature conditions. ; ; in Indicates weight Adjusted weights Indicates weight The adjusted weights are used to correct the SOC estimation bias using weighted temperature characteristics, and a temperature compensation factor is introduced. Adjustments to the SOC estimation formula: ,in This represents the original SOC estimate. This represents the SOC correction value after temperature compensation. Under high temperature conditions, the correlation between battery thermal runaway risk and temperature characteristics is significantly higher than that between voltage and current. Increasing the temperature weight allows the model to prioritize capturing the impact of temperature changes on battery capacity decay and internal resistance changes, thereby correcting the SOC offset caused by thermal runaway risk. S23, Fault condition weighting, enhanced current. Feature weights, weights Make adjustments ,in Indicates weight Adjusted weights This represents the fault weight increment, which in this embodiment will be... Set it to 0.2; then reduce the weight proportionally. ;in Indicates weight The weights after reduction, among which Indicates weight The weights after reduction, among which Indicates weight The weights after reduction, among which Indicates weight The weights after reduction, among which Indicates weight The weights are reduced; the high-weighted current characteristics are utilized in conjunction with an abnormal discharge detection algorithm to monitor the slope of current abrupt changes. , ,in This represents the change in current. Indicates the change over time; when At the same time, combined with the battery internal resistance model , This indicates the voltage change, assesses the impact of abnormal discharge on the State of Charge (SOC), and corrects the SOC estimate. : ,in For the battery's rated capacity, Integral of the discharge capacity, SOC estimate The corrected value.
[0022] For example, step S3 specifically includes the following steps: S31. Enhanced feature selectivity based on ambient temperature. A temperature-feature enhancement mapping table is constructed; the temperature is divided into a first temperature range and a second temperature range based on a preset temperature threshold, each corresponding to a different enhancement strategy to enhance the features; in this embodiment, the preset temperature threshold is set to 40℃, the first temperature range (high temperature) is (40℃, +∞), and the second temperature range (normal temperature) is [25℃, 40℃]. The first temperature range (high temperature range) In ), the enhancement factor is used. For MOS temperature and ambient temperature A secondary enhancement is performed; in this embodiment, the enhancement coefficient is... The value is set to 1.2, that is... , ,in Indicates MOS temperature The value after secondary enhancement Indicates ambient temperature The value after secondary enhancement; The second temperature range (normal temperature range) In this study, input voltage and current features are synergistically enhanced, and principal component analysis (PCA) is used for dimensionality reduction to extract the pre-existing characteristics. Each principal component, and then through the enhancement coefficient To enhance the principal components, in this embodiment, the enhancement coefficient is... The value is set to 1.1; that is... , Where X represents the input feature, This represents the value of the feature after dimensionality reduction using Principal Component Analysis (PCA). Indicates the enhanced features; S32. Model Embedding and SOC Prediction: An attention module is embedded in the deep learning model. This embodiment uses two types of deep learning models: one is a Deep Neural Network (DNN), constructing a 7-layer fully connected network with 128, 64, 32, 16, 8, 4, and 1 neurons per layer; the other is a Transformer model using an encoder-decoder architecture with 6 attention heads. Then, the enhanced features... The input is fed into the embedded model, and the importance of features is dynamically calculated through the Query-Key-Value mechanism. The weighted SOC prediction value is output to achieve accurate SOC estimation. The model can automatically adjust the weight allocation strategy of each feature in real time according to changes in working conditions, so that the SOC estimation always focuses on the core influencing factors in the current scenario. In this way, the system can maintain stable estimation accuracy and improve system availability in complex and ever-changing actual operating environments.
[0023] For example, step S32 specifically includes the following steps: S321, Enhanced features Split into sub-vectors Where Q is the query vector, K is the key vector, and V is the value vector, obtained through the formula... Calculate the similarity matrix S, normalize the similarity using the Softmax function, and then calculate the attention weights A. Finally, through the formula Calculate weighted features ; S322, SOC prediction output, model weighted features Forward propagation, the output layer uses linear regression, and the loss function is... Mean Square Error (MSE): Where N represents the number of samples, Indicates the predicted value. Representing the true value (collected by a high-precision coulomb counter), a model self-correction mechanism is set up based on the gradient of the loss function. The Adam optimizer is used to adjust model parameters and continuously improve the SOC prediction accuracy. ,in This represents the model parameters before adjustment. This represents the adjusted model parameters. The learning rate is used to represent the model parameters, which include weights and biases. In this embodiment, the learning rate is used. The value is set to 0.001.
[0024] This invention discloses a battery SOC estimation system based on attention mechanism and multi-source data weighted fusion, the schematic diagram of which is shown below. Figure 2 As shown, the battery SOC estimation method based on attention mechanism and multi-source data weighted fusion described above includes: The sensor module is used to collect multi-source data from the battery in real time. The working condition identification module is used to identify and determine the working condition based on sensor data and logic algorithms; The load condition weighting module is used to apply corresponding weights based on the type of load condition. The feature selective enhancement module is used to selectively enhance the weighted features based on the ambient temperature. The model embedding and SOC prediction module is used to input the processed features into the deep learning model embedded with the attention module, and output the weighted SOC prediction value to achieve accurate SOC estimation.
[0025] The operating condition weighting module includes a charging characteristic weighting submodule, a high temperature characteristic weighting submodule, and a fault characteristic weighting submodule. After the operating condition identification module determines that it is in charging condition, it starts the charging feature weighting submodule to weight the voltage and total voltage. After the operating condition identification module determines that the operating condition is high temperature, it starts the high temperature feature weighting submodule to weight the MOS temperature and the ambient temperature. After the operating condition identification module determines that the operating condition is faulty, it triggers the fault feature weighting submodule to weight the current.
[0026] In summary, this invention discloses a battery SOC estimation method and system based on an attention mechanism and multi-source data weighted fusion. It addresses the problem of existing technologies using an "equal treatment" approach for multi-source data such as voltage, temperature, and current. This approach assigns fixed or equal weights to each data feature regardless of the battery's operating condition (charging, high temperature, fault, etc.), ignoring the differences in data importance under different conditions. By introducing an attention mechanism, it achieves dynamic weighting of multi-source data based on real-time operating conditions. Specifically, by identifying the battery's current operating condition (e.g., charging, high temperature, fault), it automatically assigns differentiated weights to different data features. Under critical operating conditions, it increases the weight of corresponding core features (e.g., increasing voltage weight during charging, increasing temperature weight during high temperature), making the model more focused on data that plays a decisive role in SOC estimation under the current operating condition. It has the following characteristics: improved SOC estimation accuracy; dynamic weighting allows the model to "focus on key features" under different operating conditions, avoiding interference from irrelevant or secondary features. For example, during charging, high-weighted voltage features more accurately determine the charging cutoff point; at high temperatures, high-weighted temperature features correct the impact of thermal runaway on SOC; and during faults, high-weighted current features quickly capture abnormal discharges, ultimately significantly reducing SOC estimation errors under various operating conditions. This enhances model adaptability. Traditional models are prone to estimation biases when switching operating conditions (such as from normal discharge at room temperature to high-temperature charging) due to fixed weights; however, this invention uses dynamic weight adjustment driven by operating conditions to enable the model to adapt to changes in characteristics of different scenarios, maintaining stable performance in complex and ever-changing real-world applications. It also optimizes response speed in abnormal scenarios. In fault conditions (such as overcurrent protection triggering), by enhancing current weights, the model can quickly identify abnormal discharge behavior, shortening the SOC estimation response time and providing more timely basis for battery safety warnings and protection decisions.
[0027] The above description is merely a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the concept described herein through the above teachings or related technologies or knowledge. Modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.
Claims
1. A battery SOC estimation method based on attention mechanism and multi-source data weighted fusion, characterized in that, include: S1. Multi-source data acquisition and operating condition determination: Continuously acquire multi-source data of the battery through multiple sensors, including the highest voltage of a single cell, total voltage, MOS temperature, ambient temperature and current, and then determine the current operating condition of the battery. S2, Working condition-driven attention weighted processing, which weights the features corresponding to the working condition based on the working condition determination. S3. Feature enhancement and model fusion prediction: First, selectively enhance the weighted features based on the ambient temperature. Then, input the processed features into a deep learning model with an embedded attention module and output the weighted SOC prediction value to achieve accurate SOC estimation.
2. The battery SOC estimation method based on attention mechanism and multi-source data weighted fusion according to claim 1, characterized in that, Step S1 specifically includes the following steps: S11. Continuously collect multi-source data on battery operation using multiple sensors at a fixed sampling frequency, including collecting the highest voltage of individual cells using a voltage sensor. and total voltage The temperature of the MOS is collected by a temperature sensor. and ambient temperature Current is collected through a current sensor Then, the analog signal output by the sensor is converted by the ADC. Perform analog-to-digital conversion to convert the data into digital data for input into the system. ,in This represents the data collected by different types of sensors; S12, Working condition determination logic, which determines the working condition based on the collected data and preset rules.
3. The battery SOC estimation method based on attention mechanism and multi-source data weighted fusion according to claim 2, characterized in that, The operating conditions mentioned in step S12 include charging conditions, high-temperature conditions, and fault conditions; The charging condition is as follows: a charging control signal is detected and the current sensor detects current in the charging direction, i.e. and and ;in This indicates the enable command sent by the charging station. Indicates the first duration; The high-temperature operating condition is: the ambient temperature sensor continuously... Each sampling period detected The interval between each cycle ,Right now ; The fault condition is as follows: the overcurrent protection module detects current. Exceeding the preset threshold And the second duration Overcurrent trigger protection mechanism; that is and and ,in This is a fault status indicator.
4. The battery SOC estimation method based on attention mechanism and multi-source data weighted fusion according to claim 3, characterized in that, Step S2 specifically includes the following steps: S21. Weighted charging conditions are used to construct a feature weight allocation model, focusing on the highest voltage of individual cells. and total voltage Assign high weight, where high weight is percentage. Let the eigenvectors be... Weight vector Where T represents transpose, express The weight, express The weight, express The weight, express The weight, express The weights are adjusted during charging. ， ； in As the weighted increment, the weighted voltage characteristics are used in conjunction with the charging cutoff point model to make a judgment: when With battery full charge voltage threshold The absolute value of the difference is less than the voltage threshold. ,and The absolute value of the first derivative is less than the voltage change rate threshold. ,Right now and Identify signs that indicate charging is about to stop; S22. High-temperature operating condition weighting: Based on the thermal runaway risk correlation model, calculate the influence coefficient of temperature characteristics on SOC estimation. The weights are adjusted under high-temperature conditions: ; ； in Indicates weight Adjusted weights Indicates weight The adjusted weights are used to correct the SOC estimation bias using weighted temperature characteristics, and a temperature compensation factor is introduced. Adjustments to the SOC estimation formula: ,in This represents the original SOC estimate. This represents the SOC correction value after temperature compensation; S23, Fault condition weighting, enhanced current. Feature weights, weights Make adjustments ,in Indicates weight Adjusted weights This represents the increment of the fault weight; then the weight is reduced proportionally. ;in Indicates weight The weights after reduction, among which Indicates weight The weights after reduction, among which Indicates weight The weights after reduction, among which Indicates weight The weights after reduction, among which Indicates weight The weights are reduced; the high-weighted current characteristics are utilized in conjunction with an abnormal discharge detection algorithm to monitor the slope of current abrupt changes. , ,in This represents the change in current. Indicates the change over time; when At the same time, combined with the battery internal resistance model , This indicates the voltage change, assesses the impact of abnormal discharge on the State of Charge (SOC), and corrects the SOC estimate. : ,in For the battery's rated capacity, Integral of the discharge capacity, SOC estimate The corrected value.
5. The battery SOC estimation method based on attention mechanism and multi-source data weighted fusion according to claim 4, characterized in that, Step S3 specifically includes the following steps: S31. Enhanced feature selectivity based on ambient temperature. A temperature-feature enhancement mapping table is constructed; based on a preset temperature threshold, the temperature is divided into a first temperature range and a second temperature range, each corresponding to a different enhancement strategy, to enhance the features. In the first temperature range, the enhancement factor is used. For MOS temperature and ambient temperature Perform secondary enhancement, that is , ,in Indicates MOS temperature The value after secondary enhancement Indicates ambient temperature The value after secondary enhancement; In the second temperature range, the input voltage and current characteristics are synergistically enhanced, and principal component analysis (PCA) is used for dimensionality reduction to extract the pre-temperature characteristics. Each principal component, and then through the enhancement coefficient Strengthen the principal components; that is , Where X represents the input feature, This represents the value of the feature after dimensionality reduction using Principal Component Analysis (PCA). Indicates the enhanced features; S32, Model Embedding and SOC Prediction: An attention module is embedded into a deep learning model, and then the enhanced features are... The input is fed into the model, and the importance of features is dynamically calculated through the Query-Key-Value mechanism. The weighted SOC prediction value is then output to achieve accurate SOC estimation.
6. The battery SOC estimation method based on attention mechanism and multi-source data weighted fusion according to claim 5, characterized in that, Step S32 specifically includes the following steps: S321, Enhanced features Split into sub-vectors Where Q is the query vector, K is the key vector, and V is the value vector, obtained through the formula... Calculate the similarity matrix S, normalize the similarity using the Softmax function, and then calculate the attention weights A. Finally, through the formula Calculate weighted features ; S322, SOC prediction output, model weighted features Forward propagation, the output layer uses linear regression, and the loss function is... Mean Square Error (MSE): Where N represents the number of samples, Indicates the predicted value. Representing the true value, setting up a model self-correction mechanism based on the gradient of the loss function. The Adam optimizer is used to adjust model parameters and continuously improve the SOC prediction accuracy. ,in This represents the model parameters before adjustment. This represents the adjusted model parameters. This represents the learning rate.
7. A battery SOC estimation system based on attention mechanism and multi-source data weighted fusion, used in the battery SOC estimation method based on attention mechanism and multi-source data weighted fusion as described in any one of claims 1-6, characterized in that, include: The sensor module is used to collect multi-source data from the battery in real time. The working condition identification module is used to identify and determine the working condition based on sensor data and logic algorithms; The load condition weighting module is used to apply corresponding weights based on the type of load condition. The feature selective enhancement module is used to selectively enhance the weighted features based on the ambient temperature. The model embedding and SOC prediction module is used to input the processed features into the deep learning model embedded with the attention module, and output the weighted SOC prediction value to achieve accurate SOC estimation.
8. The battery SOC estimation system based on attention mechanism and multi-source data weighted fusion according to claim 7, characterized in that, The operating condition weighting module includes a charging characteristic weighting submodule, a high temperature characteristic weighting submodule, and a fault characteristic weighting submodule. The charging feature weighting submodule is used to weight the voltage and total voltage after entering the charging condition. The high-temperature characteristic weighting submodule is used to weight the MOS temperature and ambient temperature after entering the high-temperature operating condition. The fault feature weighting submodule is used to weight the current after entering a fault condition.