A lithium battery power bank temperature alarm device and method

Abstract

This invention discloses a temperature alarm device and method for a lithium battery power bank, relating to the field of lithium battery power bank technology. The method includes: pre-setting several sub-segments for the charging and discharging processes of the lithium battery power bank; responding to the current charging and discharging operation of the lithium battery power bank covering the power range of any sub-segment QR, updating the data recording unit corresponding to QR, and overwriting the original data with the data collected in QR during the current charging and discharging operation; extracting charging feature vectors and discharging feature vectors; obtaining a fused feature vector; inputting the fused feature vector into a preset risk prediction model to obtain an initial risk probability value; obtaining a target risk probability value; and dynamically adjusting the safety monitoring parameters of the power bank according to the target risk probability value; the safety monitoring parameters include a temperature change rate threshold and a temperature threshold; this invention enables the safety protection strategy to adapt to battery aging state, environmental changes, and real-time risk levels.

Description

Technical Field

[0001] This invention relates to the field of lithium battery power bank technology, and in particular to a lithium battery power bank temperature alarm device and method. Background Technology

[0002] In the current technological context, the safety management of lithium battery power banks mainly relies on real-time monitoring of voltage, current, and temperature, triggering alarms or protection when these values ​​exceed preset fixed thresholds. However, this method based on static thresholds has significant limitations. First, battery performance and safety boundaries dynamically change with aging cycles, ambient temperature, and user-specific charging and discharging habits (such as frequent shallow charging and discharging, and high-rate discharging). Fixed thresholds cannot adapt to these changes, potentially leading to underreporting of risks even after battery aging, or false alarms due to overly conservative thresholds when the battery is in good condition, impacting user experience. Second, to achieve more accurate dynamic risk assessment, some advanced solutions attempt to introduce learning models based on historical data. However, this typically requires storing and processing large amounts of time-series historical data on the device side, placing enormous pressure on the limited computing and storage resources within the power bank, making practical deployment difficult. Furthermore, the accuracy of risk assessment heavily depends on the timeliness of the data used. If the feature data used by the model has a large time span, is mixed with old and new data, or is interrupted by long time intervals during charging and discharging, the reliability of its prediction results will decrease significantly. Existing methods lack effective quantification and compensation mechanisms for data timeliness. Summary of the Invention

[0003] To overcome the shortcomings of the prior art, the present invention provides a lithium battery power bank temperature alarm device and method.

[0004] To achieve the above objectives, the present invention adopts the following technical solution, including: A method for temperature alarm of lithium battery power bank includes the following steps: S100, a number of sub-segments divided by power range are preset for the charging and discharging processes of the lithium battery power bank; wherein, each sub-segment corresponds to a data recording unit, which is used to store the data collected in the most recent charging and discharging process of the corresponding sub-segment; the data includes the power value sequence, voltage value sequence, current value sequence, temperature value sequence and corresponding timestamp of the lithium battery power bank in the sub-segment. S200, in response to the current charging and discharging operation of the lithium battery power bank covering the power range of any sub-segment, updates the data recording unit corresponding to the sub-segment, and overwrites the original data with the data collected in the corresponding sub-segment during the current charging and discharging operation; S300 extracts charging feature vectors based on the latest data of each charging segment and discharge feature vectors based on the latest data of each discharging segment. S400, the charging feature vector and the discharging feature vector are concatenated to obtain a fused feature vector; S500, the fused feature vector is input into a preset risk prediction model to obtain an initial risk probability value that characterizes the safety risk level of the current battery state and charging / discharging mode; S600, the initial risk probability value is corrected according to the time interval between the timestamps corresponding to each segment to obtain the target risk probability value; S700, dynamically adjust the safety monitoring parameters of the power bank according to the target risk probability value; the safety monitoring parameters include a temperature change rate threshold and a temperature threshold.

[0005] Preferably, step S300 includes the following steps: S310, obtain each charging sub-segment and each discharging sub-segment, and obtain a list A of charging sub-segments and a list B of discharging sub-segments; A = (A1, A2, ..., A...). i A n ), i=1,2,…,n; A i Let n be the number of charging sub-segments, where B is the i-th charging sub-segment; B = (B1, B2, ..., B...). j B m ), j=1,2,…,m; B j Let m be the j-th electron-emitting segment, and m be the number of electron-emitting segments. S320, from A i From the stored sequences of electrical quantity, voltage, current, and temperature values, a set of k preset feature values ​​is extracted to form A. i eigenvectors XA i =(XA i，1 XA i，2 , ..., XA i，r , ..., XA i，k ), r=1,2,…,k;XA i，r For A i The r-th eigenvalue; S330, from B i From the stored sequences of electrical quantity, voltage, current, and temperature values, a set of k preset feature values ​​is extracted to form B. i eigenvectors XB i =(XB i，1 XB i，2 , ..., XB i，r , ..., XB i，k ); XB i，r For B i The r-th eigenvalue; S340, Based on the feature sub-vectors corresponding to each charging segment in A, generate the charging feature vector Charge_Vec = (XA1, XA2, ..., XA...). i , ..., XA n Based on the characteristic sub-vectors corresponding to each electron-discharging segment in B, a discharge characteristic vector Discharge_Vec = (XB1, XB2, ..., XB) is generated. j , ..., XB m ).

[0006] Preferably, the k characteristic values ​​include: duration, average charge change rate, average voltage, voltage change, average current, current standard deviation, temperature change, and temperature change rate.

[0007] Preferably, step S600 includes the following steps: S610, for the p-th subsegment ZD p Get ZD p The corresponding latest timestamp T p ; S620, according to T p ZD confirmed p Data freshness decay factor F p =exp(-λ×ΔT); ΔT=T now -T p ΔT represents the current time T. now With T p The time interval between them, where λ is a preset attenuation coefficient; S630, Based on the data freshness decay factor corresponding to each segment, determine the comprehensive confidence coefficient C_total, which reflects the overall data freshness. Among them, w p For ZD p The corresponding weighting coefficient, w p According to ZD p This pertains to the charging or discharging process and ZD p The range of the current power level is set; q is the number of sub-segments; S640, the initial risk probability value P_initial is corrected according to C_total to obtain the target risk probability value P_target=P_initial×[C_min+(1-C_min)×C_total]; where C_min is the preset minimum confidence threshold.

[0008] Preferably, step S700 includes the following steps: S710, obtain the current application's temperature threshold T_base, temperature change rate threshold ΔT_base, and the battery health state coefficient SOH of the lithium battery power bank; where 0≤SOH≤1, and SOH is positively correlated with the battery state of the lithium battery power bank; S720, determine the main adjustment coefficient K_main=1-η×tanh(γ×P_target) based on the target risk probability value P_target; where η is the preset maximum adjustment amplitude coefficient, 0<η<1; γ is the preset sensitivity adjustment coefficient, γ>0; tanh() is the hyperbolic tangent function; S730, obtain the inertial smoothing coefficient K_inertia related to the historical adjustment trend; K_inertia is determined based on the moving average of the main adjustment coefficient K_main obtained from the most recent consecutive calculations; S740, based on SOH, K_main and K_inertia, determine the comprehensive adjustment coefficient K_final=[SOH×K_main+(1-SOH)×K_inertia] / [SOH+(1-SOH)]; S750, the base threshold is scaled using the comprehensive adjustment coefficient K_final to obtain the real-time application temperature threshold T_alarm and temperature change rate threshold ΔT_alarm: where T_alarm=T_base×K_final; ΔT_alarm=ΔT_base×K_final.

[0009] Preferably, after step S750, the method further includes the following steps: S760, preset safety lower limit values ​​T_min and ΔT_min and safety upper limit values ​​T_max and ΔT_max for T_alarm and ΔT_alarm respectively, so that the adjusted threshold always satisfies: T_min≤T_alarm≤T_max and ΔT_min≤ΔT_alarm≤ΔT_max.

[0010] Preferably, the preset risk prediction model is a gradient boosting decision tree model or a random forest model.

[0011] The present invention also provides a lithium battery power bank temperature alarm device, the device comprising: a main control MCU, a temperature sensor connected to the main control MCU for monitoring the temperature of different parts of the lithium battery power bank, a battery management chip, and a multimodal alarm module controlled by the main control MCU; the multimodal alarm module includes a light alarm unit, an sound alarm unit, and a tactile alarm unit; the main control MCU is used to determine whether to drive the multimodal alarm module to trigger an alarm based on the data from the temperature sensor, and to execute the lithium battery power bank temperature alarm method.

[0012] The present invention also provides a readable storage medium having a computer program stored thereon, which, when executed, implements the lithium battery power bank temperature alarm method.

[0013] The present invention also provides a computer program product, which includes a computer program / instructions that, when executed by a processor, implement the aforementioned lithium battery power bank temperature alarm method.

[0014] The advantages of this invention are: The lithium battery power bank temperature alarm method of the present invention firstly constructs a user's recent charging and discharging behavior model with extremely low storage and computing resource consumption by pre-setting power sub-segments and adopting an overlay update strategy, under the premise of storing only the key data sequence of the most recent charging and discharging process. This solves the feasibility problem of realizing personalized dynamic monitoring on resource-constrained embedded devices. Secondly, a risk probability confidence correction mechanism based on data time intervals is introduced, quantifying the freshness of data as a clear parameter affecting the final risk assessment result. This effectively eliminates the reliability decline problem caused by using outdated or excessively long interval data for prediction, ensuring the timeliness and credibility of the judgment conclusion. Finally, based on the time-corrected target risk probability value, the temperature and temperature change rate threshold are dynamically and flexibly adjusted, enabling the safety protection strategy to adapt to battery aging, environmental changes, and real-time risk levels. This improves the warning sensitivity in high-risk scenarios while reducing false alarm interference in normal conditions, achieving an optimized balance between safety protection accuracy and user experience. Attached Figure Description

[0015] Figure 1 A flowchart of a lithium battery power bank temperature alarm method provided in an embodiment of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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.

[0017] The following will refer to Figure 1 The flowchart shown illustrates a method for alarming the temperature of a lithium battery power bank.

[0018] The temperature alarm method for this lithium battery power bank may include the following steps: S100 pre-sets several sub-segments divided by power range for the charging and discharging processes of the lithium battery power bank; wherein, each sub-segment corresponds to a data recording unit, which is used to store the data collected in the most recent charging and discharging process of the corresponding sub-segment; the data includes the power value sequence, voltage value sequence, current value sequence, temperature value sequence and corresponding timestamp of the lithium battery power bank in the sub-segment.

[0019] In this embodiment, a series of continuous power ranges are preset as sub-segments for both the charging and discharging processes. Typically, these can be divided in 20% increments. For example, the discharging process can be preset into five sub-segments: 100%-80%, 80%-60%, 60%-40%, 40%-20%, and 20%-0%. The charging process follows the same principle. In the power bank's microprocessor memory, an independent "data recording unit" is established for each sub-segment. This unit is initially empty and is used to store all raw data collected during the most recent effective charging and discharging process for that sub-segment. This data includes sequences of power, voltage, current, and temperature values ​​sampled at a fixed frequency (e.g., once per second) within the sub-segment's power range, as well as a timestamp recording the end of the process. For example, a storage unit is allocated for the discharging segment "80%-60%". When the user discharges the power from 81% to 59%, all voltage, current, and temperature sampling points falling within the 80% to 60% range during this discharge, along with their end times, will be recorded in this unit.

[0020] This step abandons the traditional approach of continuously recording massive amounts of historical time-series data, maintaining only one set of data recording units with a fixed capacity corresponding to preset sub-segments. This greatly reduces the storage space requirements of embedded devices, technically solving the key challenge of behavioral data retention on resource-constrained power banks, and providing a lightweight data structure foundation for subsequent analysis.

[0021] S200, in response to the current charging and discharging operation of the lithium battery power bank covering the power range of any sub-segment, updates the data recording unit corresponding to the sub-segment (QR), and overwrites the original data with the data collected in the QR during the current charging and discharging operation.

[0022] During the real-time operation of the power bank, the current battery level is continuously monitored. When the battery level change range of a single charging / discharging operation (such as a single discharge) completely covers or passes through a preset sub-segment (for example, the battery level drops from 85% to 55%, covering the "80%-60%" sub-segment), a data update is triggered. The system extracts all sampled data (battery level, voltage, current, temperature sequence) generated within the corresponding sub-segment (QR) and uses it to completely overwrite all previously stored old data in the corresponding data recording unit of that sub-segment. Simultaneously, the moment the operation leaves the sub-segment (or the entire operation ends) is stored as a new timestamp. This process is "single-overwrite," meaning each unit always retains only the most recent relevant data. For example, if a user discharges the battery from 90% to 50% in the morning, updating the data for the "80%-60%" and "60%-40%" sub-segments; and then discharges the battery from 70% to 30% in the afternoon, the data for these two sub-segments is again overwritten and updated using the afternoon's data.

[0023] Through an overlay update mechanism, the system can dynamically and cost-effectively track users' latest charging and discharging behavior patterns, ensuring that the data used for analysis always reflects recent battery usage. This not only saves storage space but also enables feature extraction to be based on the latest and most representative behavioral data, avoiding interference from outdated historical data in judging the current state.

[0024] S300 extracts charging feature vectors based on the latest data of each charging segment and discharge feature vectors based on the latest data of each discharging segment.

[0025] Furthermore, step S300 may include the following steps: S310, obtain each charging sub-segment and each discharging sub-segment, and obtain a list A of charging sub-segments and a list B of discharging sub-segments; A = (A1, A2, ..., A...). i A n ), i=1,2,…,n; A i Let n be the number of charging sub-segments, where B is the i-th charging sub-segment; B = (B1, B2, ..., B...). j B m ), j=1,2,…,m; B j Let m be the j-th electron-electron ...

[0026] This step is data preparation before feature extraction. The system systematically reads all data recording units established for the charging process from the preset structure of the memory. These units are arranged in a preset order of power range (e.g., from low power to high power: 0%-20%, 20%-40%, ..., 80%-100%), forming an ordered list A of charging sub-segments. Each element A in the list... i(i=1,…,n) represents a specific charging segment and its associated latest data. Similarly, all data recording units established for the discharge process are read to form an ordered list B of discharging segments, where each element B... j (j=1,…,m) represents a discharge segment. The order of the list is fixed and preset, which ensures the structural consistency of the data organization each time features are extracted. For example, a typical implementation might set n=5, m=5, corresponding to five 20% charge and discharge segments, respectively.

[0027] This step organizes the scattered data records into a structured list, providing a clear input framework for subsequent batch and automated feature calculations. This systematic data organization method is a key bridge connecting raw data storage and advanced feature calculations, ensuring the logical structure and repeatability of the processing logic.

[0028] S320, from A i From the stored sequences of electrical quantity, voltage, current, and temperature values, a set of k preset feature values ​​is extracted to form A. i eigenvectors XA i =(XA i，1 XA i，2 , ..., XA i，r , ..., XA i，k ), r=1,2,…,k;XA i，r For A i The r-th eigenvalue.

[0029] Furthermore, the k characteristic values ​​include: duration, average rate of change of charge, average voltage, voltage change, average current, current standard deviation, temperature change, and temperature change rate.

[0030] For each charging segment A in list A i It retrieves four stored time series from its data recording unit: a power value series, a voltage value series, a current value series, and a temperature value series. Based on these raw sequences, it calculates a set of k pre-defined feature values ​​with clear physical or statistical significance. These k features (XA) i,1 To XA i,k This may include, but is not limited to: duration, average rate of change of charge, average voltage, voltage change, average current, current standard deviation, temperature change and rate of change of temperature.

[0031] Suppose we analyze a charging sub-segment A i Its power range is 40%-60%. This sub-segment data recording unit stores all timestamps, power values, voltage values, current values, and temperature value sequences collected during this charging process from 40% to 60% power.

[0032] Duration: Retrieve the start time t_start (when the battery reaches 40%) and end time t_end (when the battery reaches 60%) from the timestamp sequence stored in this sub-segment. Duration T_duration = t_end - t_start. For example, if t_start is 10:00:00 and t_end is 10:15:30, then T_duration is 930 seconds.

[0033] This feature directly reflects the user's charging (discharging) speed within this power range. An abnormally short duration may correspond to high-risk behaviors such as high-current fast charging or load short circuits, providing key time-dimensional information for the model to identify "rapid charging and discharging" patterns.

[0034] Average rate of change of energy: Given the energy change ΔSOC for this segment (here, 60% - 40% = 20%), and the calculated duration T_duration. Average rate of change of energy Rate_SOC = ΔSOC / T_duration. For example: ΔSOC = 20% (i.e., 0.2), T_duration = 930 seconds, then Rate_SOC ≈ 0.000215 / second or equivalently approximately 0.774 / hour.

[0035] This feature is a normalized complement to "duration," which directly quantifies the power intensity level of charging and discharging. A high rate of change typically indicates high-power operation and is one of the main drivers of heat generation rate, providing a direct input for model assessment of heat load.

[0036] Average voltage: The arithmetic mean of all sampled points in the voltage value sequence stored in this sub-segment is calculated as V_avg = (V1 + V2 + ... + VN) / N, where N is the number of voltage sampled points in this sub-segment. For example, during the 40%-60% charging process, the voltage may slowly rise from 3.7V to 3.8V. Calculating the average of all sampled points yields V_avg = 3.75V.

[0037] Average voltage reflects the typical terminal voltage of a battery within its operating range. Abnormal voltage values ​​(such as too low during charging or too high during discharging) may indicate battery aging, increased internal resistance, or connection problems. It is a fundamental characteristic for assessing whether a battery's electrical operating point is normal.

[0038] Voltage change: From the voltage value sequence of this segment, subtract the initial voltage value V_start from the final voltage value V_end to obtain ΔV = V_end - V_start. For example: During charging, V_start = 3.70V, V_end = 3.80V, then ΔV = +0.10V.

[0039] Voltage changes visually represent the degree of voltage drift within a given charge range. Excessive voltage increases during charging or excessive voltage drops during discharging may indicate severe battery polarization or overloading, serving as important signals for identifying abnormal operating conditions.

[0040] Average current: The arithmetic mean of all sampled points in the current value sequence stored in this segment, i.e., I_avg = (I1 + I2 + ... + I N I_avg is positive during charging and negative during discharging. For example, during the constant current charging phase, I_avg might be 2.0A.

[0041] Average current is one of the core determinants of heat generation (Joule heat is proportional to I²R). It is the most direct electrical characteristic for assessing the current operating stress of a battery, predicting temperature rise, and determining whether it is in an overcharged or over-discharged state.

[0042] Current Standard Deviation: Calculate the standard deviation I_std of the current value sequence for this segment, used to measure the degree of current fluctuation. For example, during charging, if the current is stable, I_std may be less than 0.1A; if load fluctuations cause current jumps, I_std may be greater than 0.5A.

[0043] This feature characterizes the stability of the current. An excessively high standard deviation of the current may indicate poor contact, drastic load changes, or abnormal power management. Drastic fluctuations in current can lead to pulsed heat generation and mechanical stress, which may accelerate battery aging and pose potential risks. This feature helps the model capture this unstable mode.

[0044] Temperature change: From the temperature value sequence of this segment, take the end temperature value Temp_end and subtract the start temperature value Temp_start to get ΔTemp = Temp_end - Temp_start. For example: Temp_start = 25°C, Temp_end = 28°C, then ΔTemp = +3°C.

[0045] This is the most direct indicator of heat accumulation. It quantifies the temperature rise of the battery itself during this charge-discharge process. A significant temperature rise is a clear physical manifestation of accumulating thermal risk and is one of the core monitoring targets of this safety method.

[0046] Rate of temperature change: Divide the calculated temperature change ΔTemp by the duration T_duration to get Rate_Temp = ΔTemp / T_duration. For example: ΔTemp = 3°C, T_duration = 930 seconds, then Rate_Temp ≈ 0.0032°C / second.

[0047] This feature is a dynamic representation of the "rate of temperature change," revealing the speed of temperature rise. A rapid rate of temperature increase (even if the final absolute value of the temperature rise is not large) is more dangerous than a slow temperature rise because it may indicate emergencies such as heat dissipation runaway or internal short circuits. Using this rate as a feature makes the model more sensitive to the dynamic evolution of thermal risks, enabling earlier and more proactive warnings.

[0048] After the calculation is completed, these k scalar characteristic values ​​are arranged in a fixed, preset order to form the charging sub-segment A. i eigenvectors XA i For example, if k=6, and the sequence is set as [duration, average current, voltage change, current standard deviation, temperature change, temperature change rate], then XA i It is a 6-dimensional vector.

[0049] These eight features constitute a multi-dimensional and complementary feature system encompassing time (duration, rate of charge change), electrical (average / variable voltage, average / fluctuating current), and thermal (temperature change and rate of change). Together, they paint a complete picture for the risk prediction model: how intensely (current, rate of charge) the battery operated for how long (time), what its electrical response (voltage) was, and what thermal results (temperature rise and rate) were ultimately achieved. This comprehensive feature engineering enables lightweight machine learning models to effectively learn the deep correlation between complex charging and discharging behaviors and potential safety risks, thereby achieving a fundamental performance improvement over traditional single temperature or voltage threshold monitoring methods.

[0050] S330, from B i From the stored sequences of electrical quantity, voltage, current, and temperature values, a set of k preset feature values ​​is extracted to form B. i eigenvectors XB i =(XB i，1 XB i，2 , ..., XB i，r , ..., XB i，k ); XB i，r For B i The r-th eigenvalue.

[0051] This step is the same as S320 in terms of logic and calculation method, but the operation object is each element B in the electron emission segment list B. j Similarly, from B j The data recording unit extracts the sequence of charge, voltage, current, and temperature, and calculates k characteristic values ​​according to a set of k characteristic definitions and calculation rules identical to those of S320. These values ​​are then arranged in the same fixed order to form the electron discharge segment B. j eigenvectors XBj This symmetry ensures that charging and discharging behaviors can be fairly compared and integrated within the same set of characteristics.

[0052] By characterizing the discharge process to the same depth, the system can quantify discharge behavior patterns, such as voltage drops and temperature rises caused by rapid discharge. Combined with the S320, the system has completed independent modeling of the two core stages of the user's charge-discharge cycle, providing a foundation for subsequent comprehensive analysis.

[0053] S340, Based on the feature sub-vectors corresponding to each charging segment in A, generate the charging feature vector Charge_Vec = (XA1, XA2, ..., XA...). i , ..., XA n And based on the characteristic sub-vectors corresponding to each electron-discharging segment in B, the discharge characteristic vector Discharge_Vec = (XB1, XB2, ..., XB) is generated. j , ..., XB m ).

[0054] After generating feature sub-vectors for all segments, this step concatenates the vectors to form a global view.

[0055] For the charging process, each segment of the charging sub-segment list A is taken out sequentially according to its inherent order (e.g., from low battery range to high battery range). i The corresponding feature vector XA i They are then concatenated end-to-end to generate a long charging feature vector Charge_Vec=(XA1,XA2,…,XA…). n The total dimension of this vector is n×k.

[0056] Similarly, following the order of electron-emitting segment list B, concatenate the eigenvectors XB of all electron-emitting segments. j Generate the discharge feature vector Discharge_Vec=(XB1,XB2,…,XB m The total dimension is m×k. For example, if n=m=5, k=6, then Charge_Vec and Discharge_Vec are both 30-dimensional vectors.

[0057] The concatenation operation integrates the feature sub-vectors describing the behavior of each local charge range into a composite feature vector that characterizes the global pattern of the entire charging and discharging process. `Charge_Vec` depicts the panoramic features of the most recent complete charging behavior, while `Discharge_Vec` depicts the panoramic features of the most recent complete discharging behavior. The generation of these two vectors marks a shift from local data to a global behavioral pattern, providing the next step of the risk prediction model with highly integrated input that simultaneously reflects electrical characteristics, thermal behavior, and their distribution patterns along the charge axis. This structured feature representation method is a key prerequisite for the effective operation of the lightweight model in this scheme.

[0058] S400, the charging feature vector and the discharging feature vector are concatenated to obtain a fused feature vector.

[0059] The charging and discharging feature vectors obtained in step S300 are directly concatenated (seminated) in a fixed order to form a longer, unified fused feature vector. For example, if the charging feature vector is 25-dimensional and the discharging feature vector is 25-dimensional, then the fused feature vector is 50-dimensional. This fused vector comprehensively represents the complete behavioral pattern of recent charging and discharging.

[0060] A single charging or discharging characteristic may not fully reflect the risks. For example, a rapid discharge may not be dangerous in itself, but if it is followed by a high-current charge, the cumulative thermal effects may pose a risk. Integrating charging and discharging characteristics provides the model with a global perspective, enabling it to assess the combined risks of continuous use across processes, significantly improving the comprehensiveness and accuracy of risk identification.

[0061] S500, the fused feature vector is input into a preset risk prediction model to obtain an initial risk probability value that characterizes the safety risk level of the current battery state and charging / discharging mode.

[0062] Furthermore, the preset risk prediction model is a gradient boosting decision tree model or a random forest model.

[0063] The preset risk prediction model can be trained through the following steps: In a laboratory or specific testing environment, a large number of power bank samples were collected under different operating conditions (such as different ambient temperatures, charge / discharge rates, and battery health status). For each charge / discharge process, segments were divided exactly according to the same rules as in the above embodiments, and a massive number of "segment data recording unit" samples were simulated and generated in an overwrite update manner. Each sample contains a sequence of power, voltage, current, temperature, and timestamp.

[0064] For each collected sample, after extracting the fused feature vector, it is labeled with a continuous true risk probability or an ordered multi-level risk label. The label can be determined by the following two methods: Method 1 (Direct Probability Truth Value): In a strictly controlled experimental environment, a quantifiable "risk score" can be defined. For example, core safety indicators such as the highest temperature and maximum temperature rise rate monitored during a single charge-discharge cycle can be mapped to a continuous value between 0 and 1 using an expert formula or normalization function. This value then serves as the direct target for model learning.

[0065] Method 2 (Ordered Ranking and Probabilistic Approach): Experts assess each charge / discharge process as having multiple ordered risk levels (e.g., 1-Safe, 2-Low Risk, 3-Medium Risk, 4-High Risk) based on multidimensional safety data (such as temperature profiles, voltage stability, etc.). During training, these level labels are treated as ordered classification targets. GBDT and Random Forest algorithms can handle this type of task well, outputting the probability distribution of each level during prediction. Normalizing the probability or expected value of the "high risk" level yields the required continuous risk probability value.

[0066] The model is trained using datasets with either continuous or ordered multi-class labels. The training objective is to minimize the error (such as mean squared error or cross-entropy loss) between the model's predicted probability values / ranks and the true labels. Cross-validation is used to ensure that the probabilities predicted by the model are well-calibrated (i.e., a sample with a predicted probability of 0.7 should have a real probability of occurring close to 70%).

[0067] The trained model undergoes lightweight processing. For gradient boosting decision trees or random forests, this can be simplified by limiting the total number of trees, their maximum depth, and by filtering based on feature importance. Then, the simplified model parameters (such as tree split points and leaf node values) and structure are solidified into a lightweight lookup table or decision rule set. This solidified model file is extremely small, requiring only simple judgments and additions for computation, and is ultimately pre-programmed (burned) into the power bank's microprocessor memory, becoming the aforementioned "preset risk prediction model."

[0068] This method completes complex model training and optimization on a resource-rich offline device, and finally transforms the results into an extremely lightweight firmware module suitable for real-time operation on embedded devices, thereby realizing advanced intelligent risk prediction functions on power banks without requiring them to have powerful local computing capabilities.

[0069] S600, the initial risk probability value is corrected according to the time interval between the timestamps corresponding to each segment to obtain the target risk probability value.

[0070] Furthermore, step S600 includes the following steps: S610, for the p-th subsegment ZD p Get ZD p The corresponding latest timestamp T p .

[0071] The system iterates through all preset charging and discharging sub-segments (collectively referred to as sub-segments ZD). p (p=1,2,…,q). For each sub-segment ZD p Read the timestamp T from its corresponding data record unit. p This timestamp records the moment when the data in this segment was most recently updated (overwritten). For example, a T-segment with electron emission "80%-60%"... p It was probably around 3 PM yesterday, and a charging sub-segment was at "20%-40%" T. p It was probably a week ago.

[0072] A timestamp is the only objective record of data timeliness. This step systematically collects the time source of all feature data, providing a basis for subsequent calculation of the "age" of each data unit and serving as a prerequisite for timeliness compensation.

[0073] S620, according to T p ZD confirmed p Data freshness decay factor F p =exp(-λ×ΔT); ΔT=T now -T p ΔT represents the current time T. now With T p The time interval between them, where λ is the preset attenuation coefficient.

[0074] For ZD p First, calculate its data age ΔT=T now -T p , among which, T now This represents the current system time. The unit for ΔT must be consistent (e.g., seconds). Then, substitute this into the exponential decay formula to calculate the data freshness decay factor F for this segment. p =exp(-λ×ΔT); where λ is a preset decay coefficient greater than 0, used to control the rate at which freshness decays over time. The larger the value of λ, the faster the freshness decreases over time.

[0075] The physical meaning of λ is "the rate at which the value of information decays per unit time". Its reciprocal 1 / λ has the dimension of time and can be approximately understood as the "half-life" or characteristic time constant of data validity.

[0076] You can define how long data is considered "basically obsolete" (e.g., F).p The freshness factor decays to approximately 0.37 (i.e., 1 / e). For example, if you want the freshness factor to decay to 0.37 after 24 hours (86,400 seconds), you can use formula F... p =exp(-λ×ΔT)=0.37. By reverse calculation, we get λ≈1 / ΔT=1 / 86400 (per second).

[0077] Once the initial range is determined, the specific optimal value of λ needs to be fine-tuned through historical data simulation and performance evaluation: Data preparation: Collect a batch of charging and discharging datasets with time-series markers and historical real-time risk labels. These datasets should contain mixed segments of data from different time intervals.

[0078] Simulation and Evaluation: On this dataset, the entire S100-S700 process was simulated, but with different candidate values ​​for λ (e.g., multiple values ​​were selected between 1.0e-6 and 1.0e-4). For each λ, its corrected target risk probability P_target was calculated and compared with the true risk label.

[0079] Optimization objective: To select the value of λ that optimizes the prediction performance of P_target. Common optimization objectives include: Prediction accuracy: The overall prediction accuracy or F1 score is the highest across different scenarios with fresh and mixed data.

[0080] Calibration: The best consistency is between the predicted probability and the actual frequency of risk occurrence (e.g., assessed using the Brier score).

[0081] Security bias: There is a preference for selecting λ, which can produce a slight conservative correction (i.e., not excessively reducing the probability of risk) when the data is outdated, in order to prioritize security.

[0082] This process is similar to hyperparameter tuning in machine learning, and can use methods such as grid search or Bayesian optimization to find the optimal λ on the validation set.

[0083] After determining a stable and robust λ value through the above optimization process, this value is used as a fixed parameter of the algorithm and compiled and embedded into the power bank's firmware along with the risk prediction model. During the productization stage, this value can also be fine-tuned for different battery chemistry systems (such as ternary lithium and lithium iron phosphate) or product designs to form a series of parameters.

[0084] This method of determining the λ value based on physical constraints and data-driven approaches ensures that the data freshness correction mechanism has a sound physical basis while maximizing the accuracy of actual risk prediction and system robustness. It avoids the arbitrariness of parameter settings, making the innovative step of "quantifying data timeliness" highly repeatable and feasible, allowing those skilled in the art to determine appropriate attenuation coefficients for their specific products.

[0085] This step maps the time age ΔT of the data to a continuous decay factor F between 0 and 1 using an exponential function. p F p =1 indicates that the data is brand new and completely reliable; F p The closer the value is to 0, the older the data and the more severe the decay of its informational value. The exponential function can well simulate the characteristic of information value being forgotten over time, providing a more reasonable model than linear decay.

[0086] S630, Based on the data freshness decay factor corresponding to each segment, determine the comprehensive confidence coefficient C_total, which reflects the overall data freshness. Among them, w p For ZD p The corresponding weighting coefficient, w p According to ZD p This pertains to the charging or discharging process and ZD p The range of the current power level is set; q is the number of sub-segments.

[0087] Not all data segments are equally important for risk assessment. Therefore, it is necessary to perform ZD analysis for each segment. p Assign a preset weighting coefficient w p This is to reflect its importance. The rules for setting the weighting coefficients are based on: Charge / discharge type: Data from the discharge process may be more sensitive to thermal risks than data from the charging process, so the weight of the discharge segment can be slightly higher.

[0088] Battery behavior in high-charge ranges (e.g., 80%-100%) and low-charge ranges (e.g., 0%-20%) is typically more critical, and its weight can be set higher than that of the intermediate-charge range. Next, calculate the overall confidence coefficient C_total, which is the weighted average freshness of all sub-segments.

[0089] For example: Assume the system has 5 charging stages and 5 discharging stages (q=10). Assign a higher weight w to the high-charge discharging stages (e.g., 100%-80%). p =2.0, assigning a lower weight to the intermediate charging segment w. p =0.8. The F value calculated for each segment... p Rather than wp Multiply the results, sum them, and then divide by the sum of all weights to get C_total. If the data has been updated frequently recently, C_total may be close to 0.9; if the data is generally outdated, C_total may be lower than 0.3.

[0090] In this embodiment, the discharge process, especially high-rate discharge, typically exhibits more direct and intense current and internal resistance heat generation effects. Therefore, the basic weight of the discharge stage can generally be higher than that of the charging stage. Energy range: High charge range (e.g., 80%-100%): This is a sensitive area for overcharge risk. In this range, the battery's internal resistance is usually low, its high current capacity is high, and it releases a large amount of energy if there is an abnormality.

[0091] Low charge range (e.g., 0%-20%): This is a sensitive area for over-discharge risk, where the battery's polarization resistance increases, resulting in more significant heat generation at the same current. Therefore, the weight of the high and low charge ranges should be higher than that of the intermediate charge range.

[0092] Based on the above theory, w p This can be determined through the following steps: Step 1: Basic Weight Allocation. Based on the dimensions mentioned above, establish a basic weight matrix. For example, initially set the baseline weight for the discharge process to 1.2 and the charging process to 1.0. Then, within each process, assign higher coefficients (e.g., 1.5) to the high and low charge ranges, and lower coefficients (e.g., 0.8) to the intermediate ranges. Sub-segment weight w p It is the product of the baseline weight of the process to which the sub-segment belongs and its electrical range coefficient.

[0093] Step Two: Experimental Data Validation and Fine-tuning. In the laboratory, by deliberately applying stress (such as high temperature, overcharge, over-discharge, and short-circuit tests) to different charge ranges and monitoring the triggering conditions and evolution path of thermal runaway, the dependence of risk events on data from each charge range can be analyzed. Based on this experimental data, the basic weight matrix is ​​fine-tuned, increasing the weights of sub-segments that are strongly correlated with early risk signs.

[0094] Step 3: Normalization. Normalize all weights (e.g., make their sum equal to 1 or the total number of segments q) so that they can be used when calculating the overall confidence coefficient C_total.

[0095] This step aggregates the scattered freshness assessments from various sub-segments into a single metric, C_total, reflecting the overall data reliability. By introducing weights wp, the system can more intelligently determine that even if some minor data is outdated, as long as the core key data is fresh, the overall confidence level can still remain high. This avoids the one-sidedness of a "one-vote veto" and makes the assessment more in line with engineering practice.

[0096] S640, the initial risk probability value P_initial is corrected according to C_total to obtain the target risk probability value P_target=P_initial×[C_min+(1-C_min)×C_total]; where C_min is the preset minimum confidence threshold.

[0097] When the data is very fresh (C_total≈1), P_target≈P_initial. The system fully trusts the model's predictions. When the data is very old (C_total≈0), P_target=P_initial×C_min. The system's predictions are significantly discounted, retaining only a portion of C_min. When the data freshness is moderate, P_target is a linear interpolation between P_initial and P_initial×C_min.

[0098] For example: Suppose the model outputs P_initial=0.8 (high risk), and C_min=0.3. If the overall data freshness is extremely high, C_total=0.95, then P_target=0.8×[0.3+0.7×0.95]≈0.8×0.965≈0.77. The risk assessment remains high.

[0099] If the data as a whole is very outdated, C_total=0.1, then P_target=0.8×[0.3+0.7×0.1]=0.8×0.37=0.296. The risk probability is significantly reduced, and the system becomes more conservative.

[0100] When data is reliable, the system boldly adopts the judgment of the intelligent model; when data is unreliable, it automatically reverts to a conservative strategy. This effectively prevents false alarms or missed alarms that might result from blindly trusting the model even with poor data quality. By setting C_min, even if the data is completely outdated, the system retains a basic risk perception capability (P_initial × C_min), avoiding the extreme situation where the early warning function is completely lost due to data expiration, thus improving the robustness of the system.

[0101] S700, dynamically adjust the safety monitoring parameters of the power bank according to the target risk probability value; the safety monitoring parameters include a temperature change rate threshold and a temperature threshold.

[0102] Furthermore, step S700 includes the following steps: S710, obtain the current application's temperature threshold T_base, temperature change rate threshold ΔT_base, and battery health state coefficient SOH of the lithium battery power bank; where 0≤SOH≤1, and SOH is positively correlated with the battery state of the lithium battery power bank.

[0103] Read the preset temperature threshold T_base (e.g., 50°C) and temperature change rate threshold ΔT_base (e.g., 1°C / s) from the system memory. These values ​​are the safety starting points determined during product design based on the battery chemistry.

[0104] State of Harm (SOH) is a coefficient reflecting the degree of battery capacity decay or aging, where 1 represents brand new and 0 represents complete failure. SOH can be determined using the following method: The coulomb counting method combined with the nominal capacity is: SOH ≈ (current full charge capacity / battery nominal capacity). The current full charge capacity can be estimated by the cumulative discharge capacity of the most recent complete discharges / (1 - initial SOC)

[0105] Internal resistance variation method: Battery internal resistance increases with aging. Under specific operating conditions (such as a low-current constant current range), the DC internal resistance R = |ΔV / I| can be estimated by measuring the voltage change ΔV and current I, and then mapped using the internal resistance decay model SOH = f(R). The function f(R) is provided by the battery manufacturer or calibrated experimentally. Finally, the estimated result is limited to between 0 and 1.

[0106] SOH enables the system to recognize that the safety boundaries of new batteries differ from those of old batteries, providing a crucial basis for subsequent differentiated threshold adjustments that match the aging state.

[0107] S720, determine the main adjustment coefficient K_main=1-η×tanh(γ×P_target) based on P_target; where η is the preset maximum adjustment amplitude coefficient, 0<η<1; γ is the preset sensitivity adjustment coefficient, γ>0; tanh() is the hyperbolic tangent function.

[0108] η: Maximum adjustment coefficient. For example, setting η=0.3 means the threshold can be tightened by a maximum of 30% (K_main can be tightened by a minimum of 0.7). γ: Sensitivity adjustment coefficient. Controls the steepness of the function's response to changes in P_target. The larger γ is, the more drastic the function changes in the middle range. tanh(): Hyperbolic tangent function, its output range is between (-1, 1). Therefore, the range of η×tanh() is (-η, η), ensuring that K_main is within (1-η, 1+η). However, we usually only care about the tightening direction when the risk increases, so the actual range of K_main is (1-η, 1).

[0109] Example: Let η = 0.25, γ = 3.0. When P_target = 0.1 (low risk), tanh(3 × 0.1) = tanh(0.3) ≈ 0.29, K_main = 1 - 0.25 × 0.29 ≈ 0.93. The threshold tightens slightly by 7%.

[0110] When P_target = 0.5 (medium risk), tanh(1.5) ≈ 0.91, K_main = 1 - 0.25 × 0.91 ≈ 0.77. The threshold tightens by 23%.

[0111] When P_target = 0.9 (high risk), tanh(2.7) ≈ 0.99, K_main = 1 - 0.25 × 0.99 ≈ 0.75. The threshold tightens by 25%, approaching the lower limit.

[0112] A nonlinear risk response was implemented using the tanh function. In the low-risk region (where P_target is small), K_main changes gradually, avoiding excessive interference with normal use; as the risk enters the medium-to-high region, K_main decreases rapidly (threshold tightening), resulting in a sensitive response; in the high-risk region, it tends to saturate, preventing overshoot. This is more in line with safety engineering practices than a linear response.

[0113] S730, obtain the inertial smoothing coefficient K_inertia related to the historical adjustment trend; K_inertia is determined based on the moving average of the main adjustment coefficient K_main calculated in the most recent consecutive calculations.

[0114] The system maintains a fixed-length queue (e.g., length = M) to store the main adjustment coefficient K_main calculated most recently (M times). The inertial smoothing coefficient K_inertia is the simple moving average of all values ​​in this queue. Each time a new K_main is calculated, it is added to the queue, the oldest value is discarded, and the average is recalculated.

[0115] For example, if the historical values ​​of K_main for the last 5 times are [0.90, 0.88, 0.85, 0.82, 0.80], then the current K_inertia = (0.90+0.88+0.85+0.82+0.80) / 5 = 0.85.

[0116] This step introduces time inertia to filter out drastic threshold fluctuations caused by occasional variations in the results of a single risk assessment (which could be noise or transient anomalies). This makes changes to security policies smoother and more stable, improving user experience and system reliability.

[0117] S740, based on SOH, K_main and K_inertia, determine the comprehensive adjustment coefficient K_final=[SOH×K_main+(1-SOH)×K_inertia] / [SOH+(1-SOH)].

[0118] When the battery is new (SOH≈1), K_final≈K_main. The system trusts the current real-time risk assessment and responds quickly.

[0119] When the battery ages significantly (SOH≈0), K_final≈K_inertia. The system relies more on the smoothed historical adjustment trend, and the strategy tends to be conservative and stable, avoiding radical adjustments due to the potential uncertainties of the aging battery model.

[0120] This step achieves dynamic coupling between the adjustment strategy and the battery aging state. New batteries are "daring to act," while old batteries prioritize stability, enabling the entire adaptive safety system to intelligently accompany the battery throughout its entire life cycle, always maintaining a reasonable and reliable level of safety alert.

[0121] S750, the base threshold is scaled using the comprehensive adjustment coefficient K_final to obtain the real-time application temperature threshold T_alarm and temperature change rate threshold ΔT_alarm: where T_alarm=T_base×K_final; ΔT_alarm=ΔT_base×K_final.

[0122] The results of all the aforementioned intelligent analyses are ultimately transformed into executable security rules. By dynamically lowering thresholds, the system can "issue early warnings and intervene in advance" under high-risk conditions, buying valuable time for security protection.

[0123] Furthermore, after step S750, the method further includes the following steps: S760, preset safety lower limit values ​​T_min and ΔT_min and safety upper limit values ​​T_max and ΔT_max for T_alarm and ΔT_alarm respectively, so that the adjusted threshold always satisfies: T_min≤T_alarm≤T_max and ΔT_min≤ΔT_alarm≤ΔT_max.

[0124] This is the ultimate safety valve for the entire dynamic adjustment system. It ensures that the algorithm's flexibility never comes at the expense of absolute safety, preventing the threshold from being adjusted to a dangerous (too high) or completely unusable (too low) value due to model errors, computational anomalies, or other unknown factors. This guarantees the absolute reliability and robustness of the system's behavior.

[0125] Furthermore, although the steps of the method in this disclosure are described in a specific order in the accompanying drawings, this does not require or imply that the steps must be performed in that specific order, or that all the steps shown must be performed to achieve the desired result. Additional or alternative steps may be omitted, multiple steps may be combined into one step, and / or a step may be broken down into multiple steps.

[0126] Embodiments of the present invention also provide a lithium battery power bank temperature alarm device, the device comprising: a main control MCU, a temperature sensor connected to the main control MCU for monitoring the temperature of different parts of the lithium battery power bank, a battery management chip, and a multimodal alarm module controlled by the main control MCU; the multimodal alarm module includes a light alarm unit, an sound alarm unit, and a tactile alarm unit; the main control MCU is used to determine whether to drive the multimodal alarm module to trigger an alarm based on the data from the temperature sensor, and to execute the lithium battery power bank temperature alarm method described in the above embodiments.

[0127] The above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for temperature alarm of a lithium battery power bank, characterized in that, Includes the following steps: S100, a number of sub-segments divided by power range are preset for the charging and discharging processes of the lithium battery power bank; wherein, each sub-segment corresponds to a data recording unit, which is used to store the data collected in the most recent charging and discharging process of the corresponding sub-segment; the data includes the power value sequence, voltage value sequence, current value sequence, temperature value sequence and corresponding timestamp of the lithium battery power bank in the sub-segment. S200, in response to the current charging and discharging operation of the lithium battery power bank covering the power range of any sub-segment, updates the data recording unit corresponding to the sub-segment, and overwrites the original data with the data collected in the corresponding sub-segment during the current charging and discharging operation; S300 extracts charging feature vectors based on the latest data of each charging segment and discharge feature vectors based on the latest data of each discharging segment. S400, the charging feature vector and the discharging feature vector are concatenated to obtain a fused feature vector; S500, the fused feature vector is input into a preset risk prediction model to obtain an initial risk probability value that characterizes the safety risk level of the current battery state and charging / discharging mode; S600, the initial risk probability value is corrected according to the time interval between the timestamps corresponding to each segment to obtain the target risk probability value; S700, dynamically adjust the safety monitoring parameters of the power bank according to the target risk probability value; the safety monitoring parameters include a temperature change rate threshold and a temperature threshold.

2. The lithium battery power bank temperature alarm method according to claim 1, characterized in that, Step S300 includes the following steps: S310, obtain each charging sub-segment and each discharging sub-segment, and obtain a list A of charging sub-segments and a list B of discharging sub-segments; A = (A1, A2, ..., A...). i A n ), i=1,2,…,n; A i Let n be the number of charging sub-segments, where B is the i-th charging sub-segment; B = (B1, B2, ..., B...). j B m ), j=1,2,…,m; B j Let m be the j-th electron-emitting segment, and m be the number of electron-emitting segments. S320, from A i From the stored sequences of electrical quantity, voltage, current, and temperature values, a set of k preset feature values ​​is extracted to form A. i eigenvectors XA i =(XA i，1 XA i，2 , ..., XA i，r , ..., XA i，k ), r=1,2,…,k;XA i，r For A i The r-th eigenvalue; S330, from B i From the stored sequences of electrical quantity, voltage, current, and temperature values, a set of k preset feature values ​​is extracted to form B. i eigenvectors XB i =(XB i，1 XB i，2 , ..., XB i，r , ..., XB i，k ); XB i，r For B i The r-th eigenvalue; S340, Based on the feature sub-vectors corresponding to each charging segment in A, generate the charging feature vector Charge_Vec = (XA1, XA2, ..., XA...). i , ..., XA n Based on the characteristic sub-vectors corresponding to each electron-discharging segment in B, a discharge characteristic vector Discharge_Vec = (XB1, XB2, ..., XB) is generated. j , ..., XB m ).

3. The lithium battery power bank temperature alarm method according to claim 2, characterized in that, The k characteristic values ​​include: duration, average rate of change of charge, average voltage, voltage change, average current, current standard deviation, temperature change, and temperature change rate.

4. The lithium battery power bank temperature alarm method according to claim 1, characterized in that, Step S600 includes the following steps: S610, for the p-th subsegment ZD p Get ZD p The corresponding latest timestamp T p ; S620, according to T p ZD confirmed p Data freshness decay factor F p =exp(-λ×ΔT); ΔT=T now -T p ΔT represents the current time T. now With T p The time interval between them, where λ is a preset attenuation coefficient; S630, Based on the data freshness decay factor corresponding to each segment, determine the comprehensive confidence coefficient C_total, which reflects the overall data freshness. Among them, w p For ZD p The corresponding weighting coefficient, w p According to ZD p This pertains to the charging or discharging process and ZD p The range of the current power level is set; q is the number of sub-segments; S640, the initial risk probability value P_initial is corrected according to C_total to obtain the target risk probability value P_target=P_initial×[C_min+(1-C_min)×C_total]; where C_min is the preset minimum confidence threshold.

5. A method for temperature alarm of a lithium battery power bank according to claim 1, characterized in that, Step S700 includes the following steps: S710, obtain the current application's temperature threshold T_base, temperature change rate threshold ΔT_base, and the battery health state coefficient SOH of the lithium battery power bank; where 0≤SOH≤1, and SOH is positively correlated with the battery state of the lithium battery power bank; S720, determine the main adjustment coefficient K_main=1-η×tanh(γ×P_target) based on the target risk probability value P_target; where η is the preset maximum adjustment amplitude coefficient, 0<η<1; γ is the preset sensitivity adjustment coefficient, γ>0; tanh() is the hyperbolic tangent function; S730, obtain the inertial smoothing coefficient K_inertia related to the historical adjustment trend; K_inertia is determined based on the moving average of the main adjustment coefficient K_main obtained from the most recent consecutive calculations; S740, based on SOH, K_main and K_inertia, determine the comprehensive adjustment coefficient K_final=[SOH×K_main+(1-SOH)×K_inertia] / [SOH+(1-SOH)]; S750, the base threshold is scaled using the comprehensive adjustment coefficient K_final to obtain the real-time application temperature threshold T_alarm and temperature change rate threshold ΔT_alarm: where T_alarm=T_base×K_final; ΔT_alarm=ΔT_base×K_final.

6. The lithium battery power bank temperature alarm method according to claim 1, characterized in that, Following step S750, the method further includes the following steps: S760, preset safety lower limit values ​​T_min and ΔT_min and safety upper limit values ​​T_max and ΔT_max for T_alarm and ΔT_alarm respectively, so that the adjusted threshold always satisfies: T_min≤T_alarm≤T_max and ΔT_min≤ΔT_alarm≤ΔT_max.

7. A method for temperature alarm of a lithium battery power bank according to claim 1, characterized in that, The preset risk prediction model is either a gradient boosting decision tree model or a random forest model.

8. A temperature alarm device for a lithium battery power bank, characterized in that, The device includes: a main control MCU, a temperature sensor connected to the main control MCU for monitoring the temperature of different parts of the lithium battery power bank, a battery management chip, and a multimodal alarm module controlled by the main control MCU; the multimodal alarm module includes a light alarm unit, an sound alarm unit, and a tactile alarm unit; the main control MCU is used to determine whether to drive the multimodal alarm module to trigger an alarm based on the data from the temperature sensor, and to execute the lithium battery power bank temperature alarm method as described in any one of claims 1-7.

9. A readable storage medium, characterized in that, It stores a computer program, which, when executed, implements the lithium battery power bank temperature alarm method according to any one of claims 1-7.

10. A computer program product, characterized in that, It includes a computer program / instructions that, when executed by a processor, implement the lithium battery power bank temperature alarm method according to any one of claims 1-7.

Images