Lithium battery safety control method of electric tool

Abstract

The invention provides a lithium battery safety control method for an electric tool, which comprises the following steps: acquiring voltage, current and temperature data of a battery and user operation data through a sensor, and generating an initial operation data set; according to the initial operation data set, generating a user operation mode by adopting a clustering algorithm, and determining a normal operation habit feature vector; according to the normal operation habit feature vector, training a collected data sequence through a recurrent neural network model to generate a personalized security baseline model; extracting a threshold range from the personalized security baseline model, judging whether the currently collected voltage and current data deviate from the threshold range, determining potential abnormal behaviors, and generating an abnormal identifier; analyzing historical charging and discharging curve data according to the anomaly identifier, and determining an anomaly type; and according to the abnormal type, adopting a preset intervention rule to adjust an output power limit, and generating an optimized battery control parameter.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of information technology, and in particular to a lithium battery safety control method for power tools. BACKGROUND

[0002] As the core power source of power tools, lithium batteries have become a key technology driving industry development due to their high energy density and portability. However, lithium batteries face dual challenges of safety and lifespan in complex usage scenarios, especially under user diversification operations and extreme working conditions. Existing technologies struggle to balance safety protection and user experience, necessitating innovative solutions to enhance battery management intelligence.

[0003] Currently, battery management systems mainly rely on fixed threshold monitoring of voltage, current, and other parameters, achieving protection through passive circuit interruption. While this approach is effective in dealing with transient abnormalities, it struggles to address the hidden risks posed by long-term non-standard user operations. For example, frequent high-load operations or the use of substandard chargers may not trigger protection thresholds, but they accelerate battery aging and create potential safety hazards. Additionally, the system lacks long-term analysis capabilities for user operation habits, making it difficult to identify potentially high-risk usage patterns, resulting in delayed protective measures that impact user experience.

[0004] The key reason lies in the inability of existing technologies to dynamically adapt to user individualized operation habits and identify complex abnormal behaviors. First, battery management systems struggle to establish dynamic safety baselines based on user actual operation patterns. For example, some users are accustomed to short-term high-power operations, while others may use low power for extended periods. Static thresholds cannot accurately distinguish between normal and abnormal behavior, leading to false positives or negatives. Second, due to the lack of continuous learning of operation habits, the system struggles to capture abnormal charging and discharging behaviors that deviate from normal patterns in real-time, such as abnormal charging curves caused by long-term use of non-standard chargers, making it difficult to intervene early to reduce risks. These two technical factors are interrelated: the absence of dynamic baselines directly leads to difficulties in identifying abnormal behaviors, and the failure to promptly identify abnormal behaviors exacerbates the accumulation of safety hazards. For example, a user charges the battery using a substandard charger in a high-temperature environment for an extended period, and the battery management system fails to identify the abnormal charging pattern, resulting in an increase in battery internal resistance, which may eventually lead to thermal runaway.

[0005] Therefore, how to design an intelligent safety control method that can dynamically learn user operation habits to establish individualized safety baselines and identify abnormal charging and discharging behaviors in real-time to implement differentiated interventions has become a key issue in improving the safety of lithium batteries for power tools and user experience. SUMMARY

[0006] The present application provides a lithium battery safety control method for power tools, mainly including: The voltage, current, temperature data and user operation data of the battery are acquired by the sensor to generate an initial operation data set; a user operation mode is generated by using a clustering algorithm according to the initial operation data set, and a normal operation habit feature vector is determined; a personalized safety baseline model is generated by training a data sequence collected through a recurrent neural network model according to the normal operation habit feature vector; a threshold range is extracted from the personalized safety baseline model, it is judged whether the current collected voltage and current data deviate from the threshold range, a potential abnormal behavior is determined, and an abnormal identifier is generated; according to the abnormal identifier, historical charge and discharge curve data are analyzed to determine an abnormal type; according to the abnormal type, a preset intervention rule is used to adjust the output power limit to generate optimized battery control parameters; according to the optimized battery control parameters, real-time feedback signals are extracted, it is judged whether the real-time feedback signals meet a preset condition, the personalized safety baseline model is updated, and an enhanced model is generated.

[0007] Further, the voltage, current, temperature data and user operation data of the battery are acquired by the sensor to generate an initial operation data set, including: real-time acquisition of voltage data of the battery by a voltage sensor; real-time acquisition of current data of the battery by a current sensor; real-time acquisition of temperature data of the battery by a temperature sensor; acquisition of user operation time and power change data by a user operation recording module; time sequence correlation of the voltage data, current data, temperature data and user operation time, power change data is performed to generate an initial operation data set, wherein the initial operation data set contains multi-dimensional time sequence features.

[0008] Further, the user operation mode is generated by using a clustering algorithm according to the initial operation data set, and the normal operation habit feature vector is determined, including: preprocessing the initial operation data set to extract a multi-dimensional feature vector; grouping the multi-dimensional feature vector by using a clustering algorithm to generate a plurality of user operation modes; for each user operation mode, calculating the statistical characteristics of the feature vector, including mean and variance; according to the statistical characteristics, the normal operation habit feature vector is determined, wherein the normal operation habit feature vector represents a typical behavior mode of user operation.

[0009] Further, the normal operation habit feature vector is used to train a data sequence collected by a recurrent neural network model to generate a personalized safety baseline model, including: the normal operation habit feature vector is used as input to initialize a recurrent neural network model; subsequent collected voltage, current and temperature data sequences are acquired; the data sequence is trained by the recurrent neural network model to optimize model parameters; a personalized safety baseline model is generated according to the optimized model parameters, wherein the personalized safety baseline model is used to predict the safe running state of the battery.

[0010] Further, the extracting a threshold range from the personalized safety baseline model, judging whether the current collected voltage and current data deviate from the threshold range, determining a potential abnormal behavior, and generating an abnormal identifier include: extracting a voltage threshold range and a current threshold range from the personalized safety baseline model; obtaining current collected voltage data and current data; if the current collected voltage data or current data deviates from the threshold range, determining a potential abnormal behavior; and generating an abnormal identifier according to the potential abnormal behavior, wherein the abnormal identifier represents an abnormal degree of a battery operating state.

[0011] Further, the analyzing historical charging and discharging curve data according to the abnormal identifier and determining an abnormal type include: obtaining a time point corresponding to the abnormal identifier; extracting charging and discharging curve data associated with the time point from a historical database; determining an abnormal type by comparing the charging and discharging curve data with a preset abnormal mode feature; if the charging and discharging curve data conforms to a high-temperature charging mode, determining a high-temperature charging abnormality; and if the charging and discharging curve data conforms to a high-load mode, determining a high-load aging abnormality.

[0012] Further, the adjusting an output power limit according to the abnormal type using a preset intervention rule and generating optimized battery control parameters include: obtaining the abnormal type; extracting a corresponding power adjustment strategy from a preset intervention rule library according to the abnormal type; if the abnormal type is a high-temperature charging abnormality, reducing an upper limit of the output power; if the abnormal type is a high-load aging abnormality, adjusting a power output curve; and generating optimized battery control parameters according to the power adjustment strategy, wherein the optimized battery control parameters are used to control the operating state of the battery.

[0013] Further, the extracting real-time feedback signals according to the optimized battery control parameters, judging whether the real-time feedback signals meet a preset condition, updating the personalized safety baseline model, and generating an enhanced model include: collecting real-time feedback signals according to the optimized battery control parameters; judging whether the real-time feedback signals meet a preset risk reduction condition; if the real-time feedback signals meet the preset risk reduction condition, inputting a data sequence of the real-time feedback signals into the personalized safety baseline model; incrementally training the data sequence by a recurrent neural network model, updating model parameters, and generating an enhanced model.

[0014] The technical scheme provided by the embodiments of the present application can include the following beneficial effects: The application discloses a lithium battery safety control method of an electric tool, and aims at safety and service life problems caused by user operation habit difference in battery operation, and fuses sensor data collection, user operation mode analysis, individualized safety baseline modeling and dynamic optimization to construct an intelligent battery management scheme. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 A flowchart of the lithium battery safety control method of the electric tool. DETAILED DESCRIPTION

[0016] The technical solutions in the embodiments of the application will be described clearly and in detail below with reference to the drawings in the embodiments of the application. The described embodiments are only some of the embodiments of the application.

[0017] As shown in the figure, the lithium battery safety control method of the electric tool can specifically include the following steps. Figure 1 Step S1, acquiring voltage, current and temperature data of the battery and user operation data through a sensor, and generating an initial operation data set, including: acquiring voltage data of the battery in real time through a voltage sensor; acquiring current data of the battery in real time through a current sensor; acquiring temperature data of the battery in real time through a temperature sensor; acquiring user operation time length and power change data through a user operation record module; and associating the voltage data, current data, temperature data, user operation time length and power change data in time sequence to generate the initial operation data set, wherein the initial operation data set contains multi-dimensional time sequence features.

[0018] In an embodiment, the step of acquiring voltage, current and temperature data of the battery and user operation data through a sensor, and generating an initial operation data set includes the following sub-steps.

[0019] Step S101, acquiring voltage data of the battery in real time through a voltage sensor.

[0020] ​Specifically, the voltage sensor is installed between the positive and negative electrodes of the battery, and the voltage value is collected once per second to ensure that the data accuracy is within 0.01V.

[0021] Step S102, real-time collection of current data of the battery by the current sensor.

[0022] Specifically, the current sensor is connected in series in the battery circuit, and the current value is collected once per second, covering a range of -10A to 10A to capture the charging and discharging process.

[0023] Step S103, real-time collection of temperature data of the battery by the temperature sensor.

[0024] Specifically, the temperature sensor is attached to the surface of the battery, and the temperature value is collected once per second, with an accuracy of 0.1 degrees Celsius and a monitoring range of -20 to 80 degrees Celsius.

[0025] Step S104, collection of user operation duration and power change data by the user operation recording module.

[0026] Specifically, the user operation recording module is integrated in the device control system, and records the start time, end time and power fluctuation curve of each operation, such as capturing the change process of power from 0W to 500W in the use of electric tools. This collection method helps to identify user habit patterns and improve the relevance of subsequent data analysis.

[0027] Step S105, time series association of the voltage data, current data, temperature data and user operation duration and power change data to generate an initial operation data set, wherein the initial operation data set contains multi-dimensional time series features.

[0028] Specifically, step S1051, aligning each data point collected according to a unified time stamp, for example, synchronizing voltage, current, temperature and operation data to the same millisecond level time axis. Step S1052, constructing a multi-dimensional vector sequence, each time point vector including voltage value, current value, temperature value, operation duration accumulation and power change rate, forming a time series matrix. Step S1053, extracting sequence features, such as calculating the mean, variance and correlation coefficient of voltage and current to ensure that the data set reflects the dynamic state of the battery.

[0029] In one embodiment, the time series association process in step S105 is applied to the electric screwdriver battery monitoring scenario.

[0030] Specifically, first, collect data within one hour: voltage from 4.2V to 3.7V, current from 2A to 0.5A, temperature from 25 degrees Celsius to 35 degrees Celsius, user operation time for 45 minutes, power change from 200W peak to 50W. Then, generate a matrix according to the timestamp association, each row represents a second of data points, and the multi-dimensional features include a voltage-current correlation coefficient of 0.85 and a temperature-power change rate of 0.6. This association can early detect abnormal patterns, such as power peaks when the temperature abnormally rises, which is beneficial to improve the accuracy of battery safety warnings.

[0031] For example, in another possible implementation, for the application of an electric angle grinder battery, step S105 is extended to include sliding window analysis: a 10-second window is applied to the associated sequence to calculate the local mean, generating features such as temperature gradient, ensuring that the data set captures the instantaneous changes of battery parameters in dynamic changing loads, thereby optimizing the input quality of the subsequent clustering algorithm and reducing the risk of misjudgment.

[0032] It should be noted that the initial operation data set formed by the time series association in the above embodiment can provide a reliable basis for subsequent clustering and neural network training. In the field of lithium batteries, this multi-dimensional feature extraction process simplifies data preprocessing, shortens the response time from collection to analysis, and improves the real-time performance of the overall system.

[0033] Step S2, generating user operation patterns from the initial operation data set using a clustering algorithm to determine normal operation habit feature vectors, includes: preprocessing the initial operation data set to extract multi-dimensional feature vectors; grouping the multi-dimensional feature vectors using a clustering algorithm to generate multiple user operation patterns; for each user operation pattern, calculating the statistical properties of the feature vectors, including mean and variance; and determining normal operation habit feature vectors according to the statistical properties, wherein the normal operation habit feature vectors represent typical behavior patterns of user operation.

[0034] In one embodiment, step S2 generates user operation patterns from the initial operation data set using a clustering algorithm to determine normal operation habit feature vectors, specifically including step S201, preprocessing the initial operation data set to extract multi-dimensional feature vectors, wherein preprocessing includes removing noise data and normalization processing, and the extracted feature vectors cover voltage, current, temperature, operation time, and power change dimensions.

[0035] Step S202, grouping the multi-dimensional feature vectors using a clustering algorithm to generate multiple user operation patterns, wherein the clustering algorithm is a K-means algorithm, and by setting the number of clusters K and iteratively calculating the cluster centers, similar feature vectors are grouped into the same group to form operation patterns such as daily low-power use or high-load charging mode.

[0036] Step S203, for each user operation mode, calculate the statistical properties of the feature vectors, including mean and variance, wherein the mean and variance values of each dimension are calculated for the vectors within each mode to quantify the data distribution within the mode.

[0037] Step S204, according to the statistical properties, determine the normal operation habit feature vector, wherein the normal operation habit feature vector represents the typical behavior mode of user operation, and the vector representing user habits is formed by taking the mean as the center point and defining the boundary in combination with the variance.

[0038] For example, in step S201, when preprocessing the initial operation data set, first filter out abnormal points with voltage values exceeding the normal range of the sensor, and then perform minimum-maximum normalization on the remaining data to make the values of each dimension between 0 and 1, which facilitates subsequent processing, thus improving data quality and reducing calculation bias.

[0039] In one embodiment, step S202 uses the K-means algorithm for grouping, first randomly initializes K cluster centers, for example, K is set to 3, corresponding to low, medium and high power operation modes; then calculate the Euclidean distance of each multi-dimensional feature vector to the cluster center, and assign it to the nearest cluster; iteratively update the cluster center until convergence, and the generated user operation mode can reflect the voltage, current and temperature distribution of the lithium battery under different user habits, for example, a mode shows that the temperature is always between 25-35 degrees Celsius, and the power changes little, representing normal daily use.

[0040] For example, for lithium battery voltage data, if the vectors of a cluster show that the voltage is stable at about 3.7 volts and the current is below 1 ampere, then this mode is identified as low-load operation, which is helpful for subsequent anomaly detection.

[0041] In one embodiment, step S202 can adjust the K value to 4 to cover more scenarios, such as operation modes in high temperature environments. By increasing the number of clusters, the algorithm can refine the grouping, for example, to distinguish the power change patterns of users in summer high temperature charging from those in winter, thereby improving the accuracy of mode generation. This adjustment can bring better personalized adaptation and reduce the potential for misjudging normal behavior as abnormal.

[0042] For example, when calculating the statistical properties in step S203, for all feature vectors within an operation mode, the mean of the voltage dimension is calculated as 3.6 volts and the variance is calculated as 0.1 volts squared, and similarly for the current and temperature dimensions. These statistical values quantify the stability and variability of the mode.

[0043] In an embodiment, step S204 determines the normal operation habit feature vector according to the statistical characteristics, takes the mean vector of each mode as the core, and defines the confidence interval with the variance, for example, the feature vector is [voltage mean 3.7, current mean 0.5, temperature mean 30, variance boundary ±0.2], so the user's typical behavior mode is used for comparison with real-time data to determine whether it deviates from the habit, thereby enhancing the accuracy of battery safety monitoring. The beneficial effect of this method is that it can dynamically adapt to user behavior changes and improve the reliability of abnormal identification.

[0044] Step S3, according to the normal operation habit feature vector, the collected data sequence is trained by a recurrent neural network model to generate a personalized safety baseline model, including: taking the normal operation habit feature vector as input, initializing the recurrent neural network model; obtaining subsequent collected voltage, current, temperature data sequence; training the data sequence through the recurrent neural network model, optimizing the model parameters; generating a personalized safety baseline model according to the optimized model parameters, wherein the personalized safety baseline model is used to predict the safe running state of the battery.

[0045] Step S301, taking the normal operation habit feature vector as input, initializing the recurrent neural network model.

[0046] In an embodiment, the normal operation habit feature vector includes the user's typical voltage range, current fluctuation mode and temperature threshold, which are extracted from the initial data set by clustering algorithm. These vectors are input to the initial layer of the recurrent neural network, the hidden state of the network is set to zero vector, and the weight matrix of the input gate and the forgetting gate is adjusted according to the vector dimension, so that the model is initially adapted to the user's operation mode.

[0047] Step S302, obtaining subsequent collected voltage, current, temperature data sequence.

[0048] For example, the voltage value of the lithium battery such as 3.7V, the current value such as 1.5A and the temperature value such as 25°C are collected by the sensor every second to form time series data, which are directly obtained from real-time monitoring for subsequent training.

[0049] The sampling interval of the data sequence is adjusted according to the collection frequency, for example, the interval is shortened to 0.5 seconds during high-power operation to capture rapid changes.

[0050] Step S303, training the data sequence through the recurrent neural network model to optimize the model parameters.

[0051] In one embodiment, the recurrent neural network model adopts a long short-term memory network structure, including input gates, forget gates, and output gates, for processing the time dependency of sequence data. The data sequence is divided into training batches, each containing 100 time steps of voltage, current, and temperature values. The deviation between predicted output and actual safety is calculated through forward propagation, and the weights are updated using the backpropagation through time algorithm, such as adjusting the bias of the forget gate to retain long-term operating mode memory. The optimization process is repeated for 10 cycles until the loss function, such as mean squared error, is reduced to below 0.01, thereby adapting the model parameters to the personalized data.

[0052] Step S3031, a long short-term memory network is selected as a specific form of recurrent neural network, where the input gate controls the entry of new information, the forget gate decides which historical state to retain, and the output gate generates the current prediction, and the parameter matrices of these gates are optimized through gradient descent.

[0053] Step S3032, a dropout layer is introduced in training to prevent overfitting, such as dropping 20% of neurons in the hidden layer, ensuring the robustness of the model to noisy data.

[0054] Step S3033, for the daily charging scenario of lithium batteries, the training data sequence focuses on the gradual increase of voltage from 3.2V to 4.2V, and the parameters are optimized to predict the normal state where the temperature does not exceed 40°C; in high-voltage discharge scenarios such as power tool use, the sequence emphasizes the case where the current peak reaches 5A, and the parameters are adjusted to identify potential overload risks.

[0055] For example, in the daily charging embodiment, the trained model parameters can accurately capture user habits such as low current fluctuations caused by slow charging, resulting in the beneficial effects of improving prediction accuracy and reducing false alarms.

[0056] In another embodiment, for high-load aging scenarios, the training process uses historical discharge curves as auxiliary data to optimize parameters to learn current decay patterns, with the beneficial effects of early warning of battery degradation and prolonging service life.

[0057] Step S304, according to the optimized model parameters, a personalized safety baseline model is generated, wherein the personalized safety baseline model is used to predict the safe operating state of the battery.

[0058] In one embodiment, the optimized parameters are used to build a final model, which is deployed as a real-time prediction tool, such as inputting the current voltage 3.8V, current 2A, and temperature 30°C, the model outputs the safety probability such as 95%, and if it is below 80%, it is identified as a risk. After the model is generated, the accuracy rate reaches 98% through the validation set test, ensuring its reliability in predicting the safe operating state of the battery.

[0059] Step S3041, define the threshold range of the model based on the optimization parameters, such as the voltage safety interval 3.0V-4.2V, generate the baseline for subsequent anomaly detection.

[0060] Step S3042, incorporate user-specific habits into the model when generating the model, such as long-time low-power operation, adjust the prediction function to favor conservative estimates, and the beneficial effect is to adapt to different user scenarios and improve safety protection.

[0061] For example, in one embodiment, for battery use in high-temperature environments, the model predicts the state considering the impact of temperature on voltage, and outputs a warning if the temperature exceeds 45°C, which has the beneficial effect of preventing thermal runaway accidents.

[0062] In another embodiment, for fast charging users, the model generates a prediction that emphasizes current peaks, and after optimization, it can evaluate the running state in real time, and the beneficial effect is to dynamically adjust to avoid overcharging damage.

[0063] Step S4, the threshold range is extracted from the personalized safety baseline model, whether the current collected voltage and current data deviates from the threshold range is judged, the potential abnormal behavior is determined, and the abnormal identifier is generated, including: extracting the voltage threshold range and the current threshold range from the personalized safety baseline model; acquiring the current collected voltage data and current data; if the current collected voltage data or current data deviates from the threshold range, it is determined as a potential abnormal behavior; according to the potential abnormal behavior, an abnormal identifier is generated, wherein the abnormal identifier represents the abnormal degree of the battery running state.

[0064] In one embodiment, step S401, the voltage threshold range and the current threshold range are extracted from the personalized safety baseline model, specifically including, step S4011, accessing the output layer of the personalized safety baseline model, wherein the model is trained by a recurrent neural network on normal operation habit feature vectors and subsequent data sequences, and the output layer contains statistical boundary values based on historical data. Step S4012, read the upper and lower limit values of the voltage from the output layer to form the voltage threshold range, for example, the upper limit is 4.2V and the lower limit is 3.0V, and the upper and lower limit values of the current to form the current threshold range, for example, the upper limit is 2A and the lower limit is 0.5A. These threshold ranges reflect the user's personalized operation mode, such as the stable interval under daily low-power use.

[0065] Specifically, this extraction process ensures that the threshold adapts to individual differences, for example, the model for frequent high-power users will set a wider current range, thereby improving the accuracy of anomaly detection.

[0066] Step S402, acquiring the current collected voltage data and current data, specifically including, reading the voltage and current values of the lithium battery in real time through the sensor, for example, collecting data once every second during user operation.

[0067] In an embodiment, step S403, if the current collected voltage data or current data deviates from the threshold range, it is determined as a potential abnormal behavior, specifically including, step S4031, comparing the current voltage data with the voltage threshold range, if it exceeds the upper limit or is lower than the lower limit, it is marked as a voltage abnormality. Step S4032, compare the current current data with the current threshold range, if it exceeds the upper limit or is lower than the lower limit, it is marked as a current abnormality. Step S4033, if there is a voltage abnormality or a current abnormality, it is determined as a potential abnormal behavior. This judgment logic forms a strict chain, and the threshold range extracted previously is directly used for comparison of the current data, ensuring real-time performance.

[0068] Specifically, in the high-temperature charging scenario of lithium batteries, if the voltage exceeds 4.2V, it is judged as a potential abnormality, which helps to identify the risk as soon as possible.

[0069] It should be noted that this step is beneficial to prevent battery overcharging or overdischarging, and improves the robustness of the safety baseline model through deviation detection.

[0070] In an embodiment, step S404, generating an abnormality identifier according to the potential abnormal behavior, wherein the abnormality identifier represents the abnormality degree of the battery operating state, specifically including, step S4041, calculating the deviation degree, for example, using the deviation formula to calculate the difference percentage of the current voltage and the threshold median value, and the difference percentage of the current, wherein the deviation formula is (current value-median value) / median value Step S4042, grading the abnormality degree according to the deviation percentage, for example, a deviation of less than 10% is mild, a deviation of 10% to 30% is moderate, and a deviation of more than 30% is severe. Step S4043, generating an abnormality identifier, for example, a numerical code 1 represents a mild voltage abnormality, and the degree grading represents the operating state.

[0071] Specifically, this generation process extends from the previously determined potential abnormal behavior, uses deviation calculation to quantify the degree, forms a logical chain, for example, in the high-load aging scenario, a severe current deviation is identified as 3, indicating a high risk.

[0072] It should be noted that this quantification is beneficial to the accurate adjustment of subsequent intervention rules and improves the battery life.

[0073] For example, in the daily charging scenario of lithium batteries, assuming that the voltage threshold range extracted by the personalized safety baseline model is 3.0V to 4.2V, and the current collected voltage is 4.5V, it is judged as a deviation, and the deviation is calculated as About 25%, generating a moderate abnormality identification 2. This way forms a chain from deviation calculation to identification generation, which is beneficial to real-time feedback to reduce risks.

[0074] In an embodiment, for high-power use scenarios such as power tool batteries, the model extracts a current threshold range of 1A to 3A, the current is 3.5A, and determines a potential abnormality, the deviation calculation is About 75%, generating a severe identification 4. This embodiment extends the judgment logic, emphasizing high-load detection under an aging battery, which is beneficial to prevent sudden failures.

[0075] For example, although not directly related to temperature data, if the voltage deviates with high temperature, the abnormality identification can be integrated into a composite degree, supporting the progression from simple deviation to multi-dimensional quantification, which is beneficial to enhanced model updates.

[0076] Step S5, according to the abnormality identification, analyzing historical charge and discharge curve data to determine the abnormality type, including: obtaining the time point corresponding to the abnormality identification; extracting the charge and discharge curve data associated with the time point from the historical database; determining the abnormality type by comparing the charge and discharge curve data with the preset abnormality mode characteristics; if the charge and discharge curve data conforms to the high-temperature charging mode, it is determined as a high-temperature charging abnormality; if the charge and discharge curve data conforms to the high-load mode, it is determined as a high-load aging abnormality.

[0077] In an embodiment, obtaining the time point corresponding to the abnormality identification specifically includes, step S501, analyzing the timestamp information in the abnormality identification and converting it into a standard time format for subsequent query.

[0078] Step S502, extracting the charge and discharge curve data associated with the time point from the historical database, specifically including, step S5021, locating the relevant record in the database according to the converted time format; step S5022, extracting the curve data including voltage, current and temperature sequence to ensure data integrity.

[0079] In an embodiment, step S503, by comparing the charge and discharge curve data with the preset abnormality mode characteristics, determining the abnormality type, specifically including, step S5031, loading the preset abnormality mode characteristics, these characteristics are the curve shape parameters such as slope and peak value obtained by historical data statistics; step S5032, calculating the similarity between the extracted charge and discharge curve data and each abnormality mode characteristic, using the cosine similarity method, where the similarity higher than 0.8 is considered as matching; step S5033, classifying the abnormality type according to the similarity result. This comparison method can improve the accuracy of abnormality detection, for example, in the scenario of lithium batteries used in power tools, quickly identifying the safety hazards caused by curve deviation, which is beneficial to prevent overheating risks.

[0080] For example, in step S504, if the charge-discharge curve data conforms to the high-temperature charging mode, it is determined to be a high-temperature charging anomaly.

[0081] Specifically, if the temperature peak of the comparison display curve exceeds the preset 45 degrees Celsius and the current is stable at a high level, the anomaly is confirmed, which helps to intervene in time to avoid battery swelling.

[0082] In one embodiment, in step S505, if the charge-discharge curve data conforms to the high-load mode, it is determined to be a high-load aging anomaly, specifically including: in step S5051, checking whether the power change record in the curve shows that the continuous high load exceeds the preset time length, such as 30 minutes; in step S5052, if the voltage drop rate is higher than the normal aging curve, it is determined to be this type of anomaly. This method is particularly effective in high-load applications of lithium batteries, such as electric impact drill scenarios, and can prolong the battery life by identifying aging signs early, for example, in actual operation, if the curve shows that the power peak is 80 watts and is maintained, causing the voltage to drop from 4.2 volts to 3.8 volts at a faster rate, it is determined to be abnormal, so as to adjust the use parameters to reduce the loss, which is beneficial to improve the overall safety and durability.

[0083] For example, in the charging scenario of lithium batteries for electric impact drills, the comparison process in step S503 can be expanded to first extract the curve data associated with the time point, and then compare it with the high-temperature charging mode characteristics. If the temperature curve rises at a slope greater than 2 degrees Celsius per minute, it is determined to be abnormal, which can effectively prevent accidents caused by overcharging overheating; in another high-load aging scenario, such as the use of electric angle grinders, if the curve shows that repeated high-power discharge causes capacity decay to accelerate, it is determined by step S505, which is beneficial to users to obtain optimized control parameters to maintain battery performance.

[0084] In step S6, the output power limit is adjusted according to the preset intervention rule to generate optimized battery control parameters, including: obtaining the anomaly type; according to the anomaly type, extracting the corresponding power adjustment strategy from the preset intervention rule library; if the anomaly type is a high-temperature charging anomaly, reducing the upper limit of the output power; if the anomaly type is a high-load aging anomaly, adjusting the power output curve; according to the power adjustment strategy, generating optimized battery control parameters, wherein the optimized battery control parameters are used to control the running state of the battery.

[0085] In one embodiment, the anomaly type is obtained, specifically including: in step S601, the anomaly type identifier is parsed from the foregoing anomaly identifier, wherein the anomaly type identifier is obtained based on the comparison of the historical charge-discharge curve.

[0086] Step S602, according to the abnormal type, extract the corresponding power adjustment strategy from the preset intervention rule library, specifically including, step S6021, input the abnormal type identifier to the intervention rule library, which is a database constructed by pre-collecting intervention data under multiple lithium battery abnormal scenarios, containing the mapping relationship between abnormal type and power strategy; step S6022, query the entries matching the abnormal type in the rule library, and extract the corresponding strategy description, for example, the strategy description includes adjustment amplitude and condition threshold.

[0087] In an embodiment, the intervention rule library construction process of step S602 can be obtained by statistical analysis of the historical operation data set of the lithium battery, for example, collecting power change records under different abnormal types, calculating the average adjustment value, forming a mapping table, so as to ensure that the extracted strategy has data support.

[0088] Step S603, if the abnormal type is high-temperature charging abnormality, reduce the output power upper limit, specifically including, step S6031, compare the current temperature value with the temperature threshold in the normal operation habit feature vector, if it exceeds the threshold, calculate the reduction amplitude, for example, the amplitude is determined based on the temperature deviation ratio; step S6032, update the power upper limit by applying the reduction amplitude, for example, the original upper limit is 100W, the deviation ratio is 20%, then the new upper limit is 80W; step S6033, integrate the updated upper limit into the power adjustment strategy to prevent safety risks caused by battery overheating.

[0089] For example, in the scenario of lithium battery used for electric blower, if high-temperature charging abnormality occurs at an ambient temperature of 40 degrees Celsius, reducing the output power upper limit can reduce the risk from high to medium, which is beneficial to prolong the battery life.

[0090] In an embodiment, step S603 can be extended to multiple levels of temperature abnormality, for example, if the temperature is above 50 degrees Celsius, the upper limit is further reduced to 50% of the original value, which is obtained by simulating the high-temperature charging curve reasoning, to ensure stable operation of the battery in extreme scenarios.

[0091] Step S604, if the abnormal type is high-load aging abnormality, adjust the power output curve, specifically including, step S6041, analyze the historical charge and discharge curve data, and identify aging indicators such as capacity attenuation rate; step S6042, fit a new power output curve according to the aging indicator, for example, use linear interpolation method to smooth the curve, so that the peak power gradually decreases; step S6043, generate the adjusted curve parameters, such as adjusting the original curve peak value of 150W to a curve gradually decreasing to 120W, to alleviate the aging acceleration under high load.

[0092] For example, for long-term high-load lithium batteries, if the aging anomaly detects that the capacity attenuation rate reaches 15%, adjusting the power output curve can increase the battery cycle life by 20%, which is beneficial to maintain stable output.

[0093] In an embodiment, the curve adjustment of step S604 can be for different aging stages, such as fine-tuning the curve slope in the early aging stage, and greatly compressing the output range in the late aging stage, which is based on the inference expansion of historical curve comparison to form multi-scene coverage.

[0094] Step S605, generating optimized battery control parameters according to the power adjustment strategy, wherein the optimized battery control parameters are used to control the running state of the battery, specifically including encapsulating the adjustment results in step S603 or S604 as a parameter set, for example, including new power upper limit and curve parameters, and directly applying to the battery controller for real-time regulation and control output.

[0095] For example, after the application of the optimized parameters, the battery running state can be converted from abnormal to normal, which is beneficial to improve the overall safety and efficiency.

[0096] Step S7, according to the optimized battery control parameters, extracting real-time feedback signals, judging whether the real-time feedback signals meet the preset conditions, updating the individualized safety baseline model, and generating an enhanced model, including: collecting real-time feedback signals according to the optimized battery control parameters; judging whether the real-time feedback signals meet the preset risk reduction conditions; if the real-time feedback signals meet the preset risk reduction conditions, inputting the data sequence of the real-time feedback signals into the individualized safety baseline model; updating the model parameters by incremental training of the data sequence through the recurrent neural network model, and generating an enhanced model.

[0097] Step S701, collecting real-time feedback signals according to the optimized battery control parameters.

[0098] In an embodiment, step S701 specifically includes: step S7011, extracting the current output power limit value and voltage and current adjustment value from the optimized battery control parameters as basic data; step S7012, using sensors to monitor the voltage, current and temperature changes of the lithium battery in real time, and generating a feedback signal sequence, wherein the feedback signal includes voltage stability and current fluctuation rate. In this way, it is ensured that the feedback signal directly reflects the battery state after parameter adjustment, forming a data input chain.

[0099] Step S702, judging whether the real-time feedback signals meet the preset risk reduction conditions.

[0100] Exemplarily, in step S702, the preset risk reduction condition is defined as that the voltage fluctuation rate in the feedback signal is lower than 0.5% and the temperature rise rate is less than 1 degree Celsius per minute; in step S7021, the voltage fluctuation rate of the real-time feedback signal is calculated, and the sequence stability is evaluated by a standard deviation formula; in step S7022, the calculation result is compared with the preset threshold value, and if the condition is met, the risk reduction is confirmed. This judgment logic helps to quickly identify the effectiveness of parameter optimization and avoid invalid updates.

[0101] In step S703, if the real-time feedback signal meets the preset risk reduction condition, the data sequence of the real-time feedback signal is input into the personalized safety baseline model.

[0102] In a possible implementation, step S703 involves converting the feedback signal sequence meeting the condition into a standardized vector, which is input into the input layer of the model to ensure data compatibility. This step briefly bridges the judgment result and the subsequent training, forming a continuous processing chain.

[0103] In step S704, the data sequence is incrementally trained by a recurrent neural network model, the model parameters are updated, and an enhanced model is generated.

[0104] In an embodiment, step S704 specifically includes step S7041, in which a long short-term memory network is used as a specific form of recurrent neural network, which processes sequence data through a forgetting gate, an input gate and an output gate mechanism. The forgetting gate decides which historical information to retain, the input gate adds new information, and the output gate generates the current output; step S7042, in which the input data sequence is processed in batches, each batch containing the latest voltage, current and temperature records; step S7043, in which a back propagation algorithm is used to calculate the gradient, and only the parameters affected by the new data are updated to realize incremental training; and step S7044, in which an enhanced model is generated based on the updated parameters, which can better predict potential abnormalities.

[0105] Specifically, the forgetting gate formula of the long short-term memory network is wherein is a weight matrix, is a previous hidden state, is a current input, is a bias, is an activation function. Through this mechanism, the model gradually integrates new feedback data and improves the adaptability to user operation patterns. In the lithium battery charging scenario.

[0106] For example, when the feedback signal shows that the temperature decreases from 45 degrees Celsius to 40 degrees Celsius, and the risk reduction condition is met, the incremental training makes the model adjust the high-temperature threshold to 42 degrees Celsius, and the safety prediction accuracy is improved by 5%.

[0107] Exemplarily, in a high-load aging scenario, the incremental training of step S4 can cover the power change record.

[0108] For example, the feedback signal sequence includes data of current exceeding 2 amperes for 5 consecutive minutes, and the model parameter update after training reduces the aging threshold from 80% capacity to 75%, thereby detecting the high-load risk in advance, and the beneficial effect is to prolong the battery life by 10%.

[0109] In a possible implementation, step S704 is extended to multiple sequence inputs for different user operation durations.

[0110] For example, short-time operation such as 10-minute charging vs. long-time such as 2-hour use, the learning rate is adjusted to 0.001 and 0.0005 respectively in the training process, to generate a more adaptive enhanced version model, and to improve the abnormal detection sensitivity.

[0111] With the above embodiments according to the present application as inspiration, through the above description, relevant personnel can certainly make various changes and modifications within the scope of not deviating from the technical idea of the present application.

Claims

1. A method for safety control of lithium batteries in power tools, characterized in that, include: The system acquires battery voltage, current, and temperature data, as well as user operation data, using sensors to generate an initial operation dataset. Based on the initial operation dataset, a clustering algorithm is used to generate user operation patterns and determine the feature vector of normal operation habits. Based on the characteristic vector of normal operating habits, a personalized safety baseline model is generated by training the collected data sequence through a recurrent neural network model. The threshold range is extracted from the personalized safety baseline model to determine whether the currently collected voltage and current data deviate from the threshold range, identify potential abnormal behaviors, and generate an anomaly identifier. Based on the anomaly identifier, analyze historical charge-discharge curve data to determine the anomaly type; Based on the type of anomaly, the output power limit is adjusted using preset intervention rules to generate optimized battery control parameters; Based on the optimized battery control parameters, extract real-time feedback signals, determine whether the real-time feedback signals meet preset conditions, update the personalized safety baseline model, and generate an enhanced model.

2. The method as described in claim 1, characterized in that, The process of acquiring battery voltage, current, and temperature data, as well as user operation data, through sensors to generate an initial operation dataset includes: The battery voltage data is collected in real time using a voltage sensor; The battery current data is collected in real time using a current sensor; The battery temperature data is collected in real time using a temperature sensor. The user operation recording module collects user operation duration and power change data. The voltage data, current data, temperature data, user operation time, and power change data are correlated over time to generate an initial operation dataset, which contains multi-dimensional time series features.

3. The method as described in claim 1, characterized in that, The step of generating user operation patterns and determining normal operation habit feature vectors based on the initial operation dataset using a clustering algorithm includes: The initial operation dataset is preprocessed to extract multidimensional feature vectors; The multidimensional feature vectors are grouped using a clustering algorithm to generate multiple user operation modes; For each user's operation mode, calculate the statistical properties of the feature vector, including the mean and variance; Based on the statistical characteristics, a normal operating habit feature vector is determined, wherein the normal operating habit feature vector represents the typical behavioral pattern of user operation.

4. The method as described in claim 1, characterized in that, The step of generating a personalized safety baseline model by training the collected data sequence using a recurrent neural network model based on the feature vector of normal operating habits includes: The recurrent neural network model is initialized by taking the feature vector of the normal operating habits as input. Acquire the subsequent voltage, current, and temperature data sequences; The data sequence is trained using the recurrent neural network model to optimize the model parameters; Based on the optimized model parameters, a personalized safety baseline model is generated, which is used to predict the safe operating state of the battery.

5. The method as described in claim 1, characterized in that, The step of extracting a threshold range from the personalized safety baseline model, determining whether the currently collected voltage and current data deviates from the threshold range, identifying potential abnormal behavior, and generating an anomaly identifier includes: Extract the voltage threshold range and current threshold range from the personalized safety baseline model; Acquire the currently collected voltage and current data; If the currently collected voltage or current data deviates from the threshold range, it is determined to be a potential abnormal behavior; Based on the potential abnormal behavior, an anomaly identifier is generated, wherein the anomaly identifier characterizes the degree of abnormality in the battery's operating state.

6. The method as described in claim 1, characterized in that, The step of analyzing historical charge-discharge curve data based on the anomaly identifier to determine the anomaly type includes: Obtain the time point corresponding to the anomaly identifier; Extract charge / discharge curve data associated with the stated time point from the historical database; The anomaly type is determined by comparing the charge / discharge curve data with preset anomaly pattern characteristics; If the charge / discharge curve data matches the high-temperature charging mode, then it is determined to be a high-temperature charging anomaly; If the charge / discharge curve data matches the high-load mode, it is determined to be a high-load aging anomaly.

7. The method as described in claim 1, characterized in that, The step of adjusting the output power limit according to the anomaly type using preset intervention rules to generate optimized battery control parameters includes: Obtain the exception type; Based on the type of anomaly, the corresponding power adjustment strategy is extracted from the preset intervention rule base; If the abnormality type is high-temperature charging abnormality, then reduce the upper limit of output power; If the anomaly type is high-load aging anomaly, then adjust the power output curve; Based on the power adjustment strategy, optimized battery control parameters are generated, wherein the optimized battery control parameters are used to control the operating state of the battery.

8. The method as described in claim 1, characterized in that, The step of extracting real-time feedback signals based on the optimized battery control parameters, determining whether the real-time feedback signals meet preset conditions, updating the personalized safety baseline model, and generating an enhanced model includes: Based on the optimized battery control parameters, real-time feedback signals are collected; Determine whether the real-time feedback signal meets the preset risk reduction conditions; If the real-time feedback signal meets the preset risk reduction condition, then the data sequence of the real-time feedback signal is input into the personalized safety baseline model; The data sequence is incrementally trained using a recurrent neural network model to update the model parameters and generate an enhanced model.

Images