Monitoring and early warning system and monitoring and early warning method for power lithium battery transport case

Abstract

The invention discloses a thermal runaway monitoring and early warning system and method for a power lithium battery in a transport case. The system comprises a monitoring unit, a data processing unit, a data transmission unit, a power supply unit, a display unit and the like. The monitoring unit comprises an infrared array temperature sensor, a gas sensor, a smoke sensor and a pressure sensor, and is arranged in the monitoring transportation box by adopting a distributed scheme; and the data transmission unit sends the early warning information to the terminal. According to the multi-information monitoring thermal runaway early warning method under different transportation working conditions, data of different transportation working conditions and data of batteries in a transportation box are fused, a clustering algorithm and an integrated learning algorithm are involved, the lithium battery thermal runaway risk level is quantified, and the early warning accuracy is improved. And according to the evaluation level, presetting two-level early warning indexes, judging whether the lithium battery in the transport case has a thermal runaway risk or not, and deciding whether to send out early warning information or not. The system is high in reliability, low in power consumption, flexible and convenient, can provide the safety state of the power lithium battery in the transportation process in real time and perform early warning on thermal runaway, and guarantees safe and efficient transportation of the power lithium battery.

Description

Technical Field

[0001] The present invention relates to the technical field of power lithium battery transportation safety, and particularly to a monitoring and early warning system and method for a power lithium battery transportation box. Background Art

[0002] The monitoring and early warning system and method for a power lithium battery transportation box are developed for the potential dangers such as combustion explosion and release of toxic gases during the transportation of power lithium batteries (including various forms such as battery cells, modules, and battery packs). Power lithium batteries belong to Class 9 dangerous goods and require transportation packaging with high protection performance to control safety risks. However, the currently mainly used traditional transportation packaging forms such as cardboard boxes, wooden boxes, and metal boxes have obvious limitations in protection performance, lack real-time monitoring and early warning functions, and cannot effectively cope with the risk of thermal runaway that may occur in power lithium batteries.

[0003] CN119471450A Lithium battery thermal runaway early warning system and method based on multi-source parameter monitoring discloses a lithium battery thermal runaway early warning system and method based on multi-source parameter monitoring; the method includes: collecting multi-source battery parameters in real time and marking them as real-time multi-source battery parameters; processing the real-time multi-source battery parameters to extract battery characteristic parameters; performing fusion analysis on the real-time multi-source battery parameters and the extracted battery characteristic parameters to predict the thermal runaway probability of the lithium battery; judging whether to generate an evaluation instruction according to the thermal runaway probability; if an evaluation instruction is generated, evaluating the thermal runaway type of the lithium battery and starting an active prevention and control measure.

[0004] CN119357838A Multi-parameter detection method for energy storage battery fire based on XGBoost discloses Step 1, constructing an XGBoost model; Step 2, expressing the objective function as a function of the number of samples and the number of trees, and approximately optimizing the objective function by the second-order expansion of the Taylor formula to approximate the loss function; Step 3, for each sample, updating the predicted value in an iterative manner; Step 4, adjusting the parameters of the XGBoost model according to the actual data of the energy storage battery fire, including the learning rate, the depth of the tree, and the regularization coefficient, and training the model to achieve accurate detection of multi-parameters of the energy storage battery fire.

[0005] CN119502701A New energy vehicle lithium battery thermal runaway monitoring system and method discloses including in-vehicle lithium batteries, multi-source data acquisition module, multi-source data fusion module, hidden Markov model monitoring module, state recognition module, data analysis and processing module, early warning module, emergency processing module, wireless communication module, cloud platform, and battery management system. The multi-source data acquisition module is responsible for collecting multi-dimensional data such as the voltage, temperature, and internal resistance of the lithium battery, and the data fusion module processes the data and transmits it to the hidden Markov model monitoring module for state monitoring and prediction.

[0006] CN119097873A A comprehensive system for early warning and fire extinguishing of lithium battery thermal runaway discloses an early warning subsystem, a fire extinguishing subsystem, an oxygen consumption detection subsystem, and a signal analysis, processing, and control subsystem. The early warning subsystem includes a temperature early warning unit, a current voltage - ultrasonic wave early warning unit, and a gas - sound signal early warning unit, which are respectively used to monitor the temperature, current voltage, ultrasonic wave signal, gas composition, and sound signal of the lithium battery in real time; the oxygen consumption detection subsystem includes an oxygen consumption detection and analysis unit to conduct statistics on the heat release during thermal runaway, perform data statistics for the thermal runaway early warning parameters, find the critical value of the heat released during thermal runaway, and prepare for fire extinguishing; the fire extinguishing subsystem includes a fire extinguishing unit, which is used to quickly start the fire extinguishing device to extinguish the fire after confirming the thermal runaway of the lithium battery; the signal analysis and control subsystem includes a signal analysis, processing, and control unit, which is used to receive and analyze the early warning signal and control the fire extinguishing subsystem to extinguish the fire according to the early warning signal.

[0007] CN119538129A An intelligent data analysis technology for a lithium battery fire hazard detector discloses predicting the occurrence of thermal runaway by detecting extremely trace characteristic gases released from lithium batteries. Through a large number of experimental tests on different brand battery cells, this technology has established a characteristic model database for the first stage of lithium battery thermal runaway. It adopts an intelligent data analysis algorithm that can analyze the detected gas data in real time and compare it with the characteristic model database, so as to quickly determine whether the lithium battery is in the early stage of thermal runaway. This intelligent data analysis algorithm needs to consider various complex situations and factors, such as ambient temperature, battery aging degree, charge - discharge status, etc. It is integrated with the battery management system BMS. Once a potential hazard of lithium battery thermal runaway is detected, it will immediately send an alarm to the BMS and recommend corresponding treatment measures.

[0008] However, the above - mentioned technologies have insufficient precision and are not suitable for the environment of power lithium - battery transportation boxes. Summary of the Invention

[0009] The purpose of the present invention is to provide a new type of monitoring and early warning system and method for power lithium - battery transportation boxes in view of the deficiencies of the prior art. Through monitoring parameters such as battery temperature, gas, smoke, pressure, etc., early warning of thermal runaway during the transportation of power lithium batteries is carried out to accurately determine the early warning of battery thermal runaway.

[0010] To solve the above - mentioned technical problems, the present invention adopts the following technical solutions:

[0011] The present invention provides a monitoring and early warning system for a power lithium - battery transportation box, and the system includes a monitoring unit, a data processing unit, a data transmission unit, a power supply unit, a display unit, and a response unit;

[0012] The monitoring unit includes temperature, gas, pressure, and smoke sensors;

[0013] The monitoring unit is deployed on the lower surface inside the box cover. The sensing device adopts a distributed design and is equipped with temperature, gas, pressure, and smoke sensors.

[0014] The data processing unit is arranged on the inner side of the box cover of the transport box.

[0015] The display unit includes a display panel and signal lights.

[0016] The display unit is arranged on the outer side of the transport box cover. The display panel can display the battery temperature, gas concentration, smoke concentration, and the pressure inside the box inside the transport box. The display light will give an optoelectronic alarm when it is judged that the battery has a thermal runaway. The power supply unit is a lithium battery pack, and the power supply unit is arranged on the inner side of the transport box cover to supply power to the entire early warning system. The data transmission unit includes a 4G module and WIFI. The data transmission unit and the data processing unit are integrated in a housing. The response unit includes a release box and platform-sent alarm information. When the early warning system determines that a thermal runaway occurs inside the transport box, the release box is opened through an electrical signal, and the fire extinguishing material placed inside the box will fall to suppress the battery flame. The early warning system sends information to the terminal.

[0017] The temperature sensor includes a non-contact infrared temperature sensor and a thermocouple. The infrared temperature sensor is fixed on a small fixed plate. The probe of the thermocouple extends outside the sensor housing. The wires of the infrared temperature sensor and the thermocouple are both connected from the center of the back of the housing to the collector connection wire. A number of infrared temperature sensors are connected through the CAN bus. A number of thermocouples are connected through ADC wires. The sensor housing is fixed on the inner box cover of the transport box by screws and vertically irradiates the battery below.

[0018] Preferably, the gas sensor includes a number of hydrogen sensors and a number of carbon monoxide sensors. The sensors are connected by wires and connected to the RS485 interface on the data collector. The sensor housing is fixed on the inner box cover of the transport box by screws.

[0019] Preferably, the smoke sensor includes a number of smoke sensors. The sensors are directly connected by wires and connected to the RS485 interface on the data collector. The sensor housing is fixed on the inner box cover of the transport box by screws.

[0020] Preferably, the pressure sensor includes a number of pressure sensors. The sensors are connected through the CAN bus and connected to the data collector. The sensor housing is fixed on the inner box cover of the transport box by screws.

[0021] Preferably, the data collector includes a main control PCB board, a main control chip, a data interface, and a housing; the main control PCB board is fixed in the center of the housing, the main control chip is arranged in the lower middle part of the data collector, and the data interfaces are distributed on the left side of the data collector.

[0022] Preferably, the data collector includes a communication interface for connecting a sensing device; specifically, the infrared temperature sensor communicates with the data collector in a CAN manner; the thermocouple is connected through the ADC interface for data acquisition; the gas sensor communicates in an RS485 manner; an RS232 interface is reserved on the data collector for future function expansion; the data collector also needs to be able to transmit signals of multiple sensors of the same type on one data transmission line.

[0023] Further, the CAN interface adopts 3 high-speed communication interfaces, the baud rate can reach 500Kbps, and a single bus supports simultaneous access of multiple sensors.

[0024] Further, the ADC interface adopts 1 communication interface and uses a 24-bit high-precision AD sampling chip.

[0025] Further, the RS485 interface adopts 3 communication interfaces, the baud rate can reach 115200Kbps, and a single bus supports simultaneous access of multiple sensors.

[0026] Further, 1 RS232 interface also needs to be reserved for future system expansion.

[0027] Further, the data collector is also provided with 3 UART interfaces, one of which is connected to a 4G module; one is connected to WIFI, and one is connected to an RS485 adapter.

[0028] Further, the data collector requires 16MB of byte size for storing data.

[0029] Further, the data collector can store 1 week's worth of collected data and upload continuous signals to the monitoring platform at the same time.

[0030] Further, the sampling frequency of the data collector is 2HZ.

[0031] Further, the serial port transmission speed of the data collector is not less than 9600bps.

[0032] Further, the length of the main control PCB board in the data collector is 200mm, the width is 100mm, and the thickness is 10mm.

[0033] Further, the length of the encapsulation shell of the data collector is 205.6 mm, the width is 105.6 mm, and the thickness is 11.6 mm.

[0034] Preferably, the data transmission unit includes a 4G module, WIFI, and an antenna; specifically, the 4G module and ZI GBEE are installed inside the collector; the antenna needs to be pulled out to the outside of the transport box for patch installation to ensure the communication signal strength.

[0035] Preferably, the display unit includes an LCD display panel, an alarm lamp, and control buttons; specifically, the display panel is turned on and off through the control buttons; further, the display content on the display panel is adjusted through the touch buttons; further, the safety status of the lithium battery inside the transport box is displayed by LED flashing.

[0036] Further, the display panel displays the battery temperature, H2 concentration, CO concentration, and environmental pressure inside the transport box.

[0037] Further, the display panel can display the early warning information at the first or second level inside the box.

[0038] The power supply unit includes a power supply battery and an encapsulation. Specifically, the battery encapsulation size is 224 mm × 152 mm × 73.5 mm.

[0039] The response unit includes a release box and alarm information sent by the platform; when the early warning system determines that a thermal runaway occurs inside the transport box, the release box is opened through an electrical signal, and the fire extinguishing material placed inside the box will fall to suppress the battery flame; the early warning system sends information to the terminal.

[0040] Preferably, the data transmission unit includes a 4G module and WIFI. Specifically, the data collector uploads data to the monitoring platform through the 4G module; when transported to an area with unstable signals, WIFI is used to report data to the gateway and then upload it to the monitoring platform.

[0041] The present invention also provides a monitoring and early warning method based on the above-mentioned lithium battery transport thermal runaway monitoring and early warning system, and the specific steps are as follows:

[0042] S1: Place the lithium battery in the transport box, and then perform temperature, vibration, shock, and humidity treatments on the transport box;

[0043] S2: Monitor the temperature, gas, smoke, and pressure inside the transport box through different types of sensors of the monitoring unit;

[0044] S3: Obtain various characteristic parameters of the lithium battery risk, and perform normalization processing on the acquired data;

[0045] S4: Classify risks by monitoring the multi-information status data of lithium batteries, and calculate the threshold values of lithium battery thermal runaway risk characteristic parameters under different transportation conditions;

[0046] S5: Input the quantified characteristic values into the clustering algorithm to quantify the thermal runaway risk level of lithium batteries;

[0047] S6: To improve the risk recognition accuracy and prevent false alarms, establish a data balancing - ensemble learning algorithm to construct a thermal runaway risk recognition model for power lithium batteries in the transportation box.

[0048] S7. Through the thermal runaway risk recognition model of power lithium batteries under transportation conditions, set dynamic thresholds and a secondary warning process. If it is recognized that the lithium battery has a thermal runaway risk, report a warning signal to the background.

[0049] The different transportation condition states X of the battery in the transportation box y (t) (temperature Tenv(t) T , vibration V(t), humidity S(t), shock H(t)) are,

[0050] X y (t) = [Tenv(t), V(t), S(t), H(t)].

[0051] Monitor the thermal runaway risk state X of the battery through the transportation box monitoring and warning system d (battery temperature T bat , H2 concentration H2 in the transportation box, CO concentration CO in the transportation box, smoke concentration SM in the transportation box, pressure P) and different states R of the battery monitored under transportation conditions y (t) are as follows

[0052] X d (t) = [T bat (t), H2(t), CO(t), SM(t), P(t)] R y (t) = [X y (t), X d (t)].

[0053] Extract the lithium battery thermal runaway risk characteristic parameters under different transportation conditions, including the battery temperature rise rate, H2 concentration change rate, CO concentration change rate, smoke concentration change rate, pressure change rate, to describe the change trend of the battery

[0054] ΔT = dT bat / dt ΔH2 = dH2 / dt ΔCO = dCO / dt ΔSM = dSM / dt ΔP = dP / dt.

[0055] Comprehensively determine the dynamically changing threshold R through the calculated lithium battery thermal runaway risk parameters *i to further determine its risk level. When the characteristic parameter is lower than R * i it indicates that the battery is in a normal state; when the characteristic parameter is greater than R * i it indicates that the battery is in a risk state. W T 、 W CO 、W SM 、W p are the weights of battery temperature, H2 concentration, CO concentration, smoke concentration and pressure respectively, and ΔT n 、ΔH 2n 、ΔCO n 、ΔSM n 、ΔP n are the characteristic values after normalization of battery temperature, H2 concentration, CO concentration, smoke concentration and pressure respectively. Therefore, the comprehensive characteristic threshold R * i is

[0056]

[0057] Use an algorithm for clustering to divide the risk probability levels of thermal runaway of lithium batteries when different parameters are monitored during transportation; through n samples, each sample has three dimensions, set K objects, and compare the object X i with the distance d of each clustering centroid C j , the jth dimension value of X i and C j is X it 、C jt

[0058]

[0059] CL is the Lth clustering centroid; N L is the number of samples in the Lth clustering centroid, and recalculate the centroid point with the mean value of all objects in this class,

[0060]

[0061] Repeat the above steps until the clustering centroid no longer changes, and finally complete the clustering. At the same time, use the elbow method to determine the number of clustering categories K accordingly.

[0062] The elbow method evaluates the change of the sum of squared errors (SSE) within the cluster with the number of clusters. Among them, x is the data sample point, and C i is the ith cluster where the sample point x is located, and μ is the centroid of C i

[0063] ​

[0064] Under the transportation condition, the sample size of thermal runaway of the battery is small and the distribution of the entire data set is uneven. Therefore, it is necessary to balance the data set so that the high-risk sample size is equivalent to the low-risk sample size.

[0065] According to each minority class sample x i , using the Euclidean distance as the calculation method to calculate the distance from x i to all the remaining minority class samples, so as to obtain the size of its neighborhood range

[0066] Secondly, by evaluating the imbalance degree between classes in the data set, a sampling rate M is set, and this sampling rate is adjusted based on the current imbalance ratio. Then, according to the sampling rate M, several data samples are randomly selected from the neighborhood of x i , and the number is determined by M. These selected neighboring data samples are denoted as

[0067] Finally, using the selected to generate new minority class samples, where λ is a random number between 0 and 1, and the newly generated data sample is x n ;

[0068]

[0069] Adopt a learning algorithm. First, initialize the estimated values of all samples. After determining the loss function, obtain the predicted values of the probabilities of each thermal runaway occurrence and the derivative of this function; secondly, based on the aforementioned values, create a new decision tree and add its prediction results to the previous estimated values; finally, on the basis of the second step, obtain the derivative of the loss function again,

[0070] (1) The objective function combines the loss function S and the regularization term Ω: The loss function measures the difference between the predicted value and the actual value, and the regularization term penalizes the complexity of the model to prevent overfitting. y i is the actual value, is the predicted value of the i-th instance, f k is the k-th tree, K is the total number of trees, and the objective function for a given step is defined as:

[0071]

[0072] (2) Gradient information: It is improved based on gradient boosting. It not only uses the first-order gradient information but also uses the second-order gradient to more accurately approximate the curvature of the loss function. For a given loss function S, the gradient gi and hi of each instance are calculated as follows:

[0073]

[0074] (3) Decision tree construction: For each decision tree, the best split point is found by enumerating all possible split points for all features. This process is based on the calculation of the structure score, which uses the gradient statistics of the data points falling into each split region. Here, L and R represent the left and right sub-regions after splitting, T represents the entire region before splitting, λ and γ are regularization parameters respectively, and the gain obtained by splitting is given by the following formula:

[0075]

[0076] For an imbalanced data multi-class recognition model, it is usually necessary to perform macro-average calculation and micro-average calculation on accuracy, precision, recall, F-score, and AUC value as evaluation indicators. Calculate the indicators based on the following confusion matrix

[0077] Table 1 Confusion matrix calculation indicators

[0078]

[0079] (1) Accuracy: It represents the proportion of the number of correctly identified risk levels of a certain type in the total sample size (i.e., the sum of all elements in the confusion matrix). Taking the low-risk category as an example for calculation, and so on for other categories.

[0080]

[0081] (2) Precision: It represents the proportion of the number of true risk levels caused by a certain type of working condition in the number of identified risk levels caused by a certain type of working condition. Taking the low-risk category as an example for calculation, and so on for other categories.

[0082]

[0083] (3) Recall, also known as the duplication rate (recall): It represents the proportion of the number of identified risk levels of a certain type in the number of true risk levels of a certain type. Taking the low-risk category as an example for calculation, and so on for other categories.

[0084]

[0085] (4) Comprehensive evaluation index (F-score): Perform a weighted harmonic average of Accuracy and Precision.

[0086] Specifically, when the parameter, it is the most common F1-score:

[0087]

[0088] (5) Macro Average calculates the evaluation index for each category independently and then takes the average. Therefore, the macro average of precision, recall, and F1-score can be calculated.

[0089]

[0090] Based on the thermal runaway identification model of lithium batteries under transportation conditions, a thermal runaway monitoring and early warning method for lithium batteries under different transportation conditions is obtained, including dynamic early warning thresholds and early warning processes.

[0091] Beneficial effects: (1) This system is based on intelligent monitoring and early warning technology. Various sensors (temperature, gas, pressure, smoke sensors, etc.) are installed in the power lithium battery transport box to collect transportation environment and battery status data, and upload the data to the monitoring platform through wireless communication technology. When an abnormal situation is detected, the system can issue an early warning signal in time. The early warning system can not only issue an early warning signal, but also link the response unit to release fire extinguishing materials. And start emergency response measures. Through the fusion processing of multi-sensor data and various transportation operating parameters, the probability of thermal runaway of lithium batteries can be accurately evaluated, the false alarm rate can be reduced, and the timeliness of early warning can be improved.

[0092] (2) The model combines multiple transportation conditions (temperature, vibration, shock, humidity) and multi-dimensional perception parameters (temperature, gas, smoke, pressure), divides the risk level through clustering algorithm and ensemble learning algorithm, quantifies the degree of thermal runaway by feature normalization and weighted comprehensive indicators, introduces data balancing processing to alleviate small sample bias, and constructs an ensemble learning model to classify and identify thermal runaway, thereby improving the accuracy of thermal runaway warning. Through this system, the transportation status of power lithium batteries can be clearly and comprehensively monitored, and efficient and accurate warning and risk prevention and control can be achieved, which can buy precious time for emergency rescue, minimize accident losses and risks, and ensure the safe and stable operation of the transportation process. The development and application of this method and system are of great significance to ensuring the safety of power lithium battery transportation.

[0093] (3) Due to the different transportation environments and types of batteries transported, the early warning method is based on the identification model of power lithium batteries in the transport box, using the trend of each feature change instead of simply setting a static threshold.

[0094] (4) In addition, the built-in emergency response unit can respond more quickly to lithium battery fires in transport boxes, thereby ensuring the safety of personnel and property. BRIEF DESCRIPTION OF THE DRAWINGS

[0095] Figure 1 This is a diagram of a monitoring and early warning device for a power lithium battery transport box of the present invention;

[0096] Figure 2It is the layout diagram of the warning system on the transportation box cover;

[0097] Figure 3 It is the warning flow chart. Specific implementation manner

[0098] For the convenience of those skilled in the art to understand, the present invention will be further described below in conjunction with embodiments. The content mentioned in the implementation manner does not limit the present invention.

[0099] A monitoring and warning system for a power lithium battery transportation box includes a sensing device 1, a data acquisition and transmission unit 4. The display unit 2 is connected to the data acquisition and transmission unit 4 through a wire. The data acquisition and transmission unit 4 sends an electrical signal to the response unit 5, and the power supply unit 3 supplies power to the entire system. The sensing device 1 includes an H2 sensor 23, a CO sensor 23, an infrared temperature sensor 21, and a pressure sensor 22.

[0100] A monitoring and warning method based on the above-mentioned monitoring and warning system for thermal runaway of lithium battery transportation is as follows:

[0101] Place the lithium battery in the transportation box, and then perform temperature, vibration, shock, and humidity treatments on the transportation box. Each working condition is divided into three dimensions: low, normal, and high; obtain different transportation working condition states X where the battery is located in the transportation box y (t) (temperature Tenv(t) T , vibration V(t), humidity S(t), shock H(t)) is

[0102] X y (t) = [Tenv(t), V(t), S(t), H(t)]

[0103] Monitor the thermal runaway risk state X of the battery through the transportation box monitoring and warning system d (battery temperature T bat , H2 concentration H2 in the transportation box, CO concentration CO in the transportation box, smoke concentration SM in the transportation box, pressure P) and different states R of the battery under the transportation working condition y (t) are as follows

[0104] X d (t) = [T bat (t), H2(t), CO(t), SM(t), P(t)] Monitor the temperature, gas, smoke, and pressure data in the transportation box through different types of sensors of the monitoring unit as shown in the following table;

[0105] Table 2 Monitoring data of the monitoring and warning system

[0106]

[0107] Obtain the characteristic parameters of the risks of lithium batteries, and normalize the acquired data by calculating scores.

[0108] Classify the risks by monitoring the multi-information status data of lithium batteries, including battery temperature and temperature rise rate, H2 concentration and change rate, CO concentration and change rate, smoke concentration and change rate, pressure value and change rate, and calculate the threshold values of the characteristic parameters of the thermal runaway risks of lithium batteries under different transportation conditions.

[0109] Table 3 Threshold values of thermal runaway characteristic parameters

[0110]

[0111] 5. Obtain the thermal runaway risk scores of 5 characteristic indicators under 4 different working conditions by using the interquartile range method and normalization calculation, and quantify the thermal runaway level of lithium batteries according to the scores.

[0112] Table 4 Thermal runaway risk scores

[0113]

[0114] Comprehensively determine the dynamically changing threshold R based on the calculated thermal runaway risk parameters of lithium batteries. * i To further determine its risk level. When the characteristic parameter is lower than R * i it indicates that the battery is in a normal state; when the characteristic parameter is greater than R * i it indicates that the battery is in a risk state.

[0115] First, calculate the weights of battery temperature, H2 concentration, CO concentration, smoke concentration, and pressure.

[0116] W T = 0.287, W CO = 0.196, W SM = 0.182, W p = 0.139

[0117] ΔT n 、ΔH 2n 、ΔCO n 、ΔSM n 、ΔP n are the characteristic values after normalization of battery temperature, H2 concentration, CO concentration, smoke concentration, and pressure respectively. Therefore, the comprehensive characteristic threshold R * i is

[0118]

[0119] Use an algorithm for clustering to divide the risk probability levels of thermal runaway of lithium batteries when different parameters are monitored during transportation; through n samples, each sample has three dimensions, set K objects, and compare object X in turn i with each clustering centroid C j for the distance d, where the j-th dimension value of X i and C j is X it and C jt

[0120]

[0121] CL is the L-th clustering centroid; N L is the number of samples in the L-th clustering centroid, and recalculate the centroid point with the mean of all objects in this class

[0122]

[0123] Repeat the above steps until the clustering centroid no longer changes, and finally complete the clustering. At the same time, use the elbow method to determine the number of clustering categories K accordingly

[0124] The elbow method evaluates the change of the sum of squared errors (SSE) within the cluster with the number of clusters. Among them, x is the data sample point, and C i is the i-th cluster where the sample point x is located, and μ is the centroid of C i

[0125]

[0126] The clustering results are shown in the following table. It can be seen from the table results that for category 1, the proportion is the lowest, and for category 3, the proportion is the highest. Therefore, it is divided into normal state, low risk of thermal runaway, and high risk of thermal runaway

[0127] Table 5 K-means clustering results

[0128]

[0129] The sample size of thermal runaway of the battery under the transportation working condition is small, and the distribution of the entire dataset is uneven. Therefore, it is necessary to balance the dataset so that the sample sizes of high-risk and low-risk are equivalent

[0130] According to each minority-class sample x i , use the Euclidean distance as the calculation method to calculate the distance from x i to all the remaining minority-class samples, so as to obtain the range size of its neighbor interval

[0131] ​Secondly, by evaluating the degree of imbalance between categories in the dataset, a sampling rate M is set, which is adjusted based on the current imbalance ratio. Then, according to the sampling rate M, several data samples are randomly selected from the neighborhood of x i The number is determined by M, and these selected neighboring data samples are denoted as

[0132] Finally, using the selected Generate new minority-class samples, where λ is a random number between 0 and 1, and the newly generated data sample is x n .

[0133]

[0134] Using a learning algorithm, first initialize the estimated values of all samples. After determining the loss function, calculate the predicted values of the probability of each thermal runaway occurrence and the derivative of this function. Secondly, based on the aforementioned values, create a new decision tree and add its prediction result to the previous estimated values. Finally, based on the second step, obtain the derivative of the loss function again,

[0135] (1) The objective function combines the loss function S and the regularization term Ω: The loss function measures the difference between the predicted value and the actual value, and the regularization term penalizes the complexity of the model to prevent overfitting. y i is the actual value, is the predicted value of the i-th instance, f k is the k-th tree, K is the total number of trees, and the objective function for a given step is defined as:

[0136]

[0137] (2) Gradient information: An improvement is made on the basis of gradient boosting. Not only the first-order gradient information is used, but also the second-order gradient is used to more accurately approximate the curvature of the loss function. For a given loss function S, the gradient gi and hi of each instance are calculated as follows:

[0138]

[0139] (3) Decision tree construction: For each decision tree, the best split point is found by enumerating all possible split points of all features. This process is based on the calculation of the structure score, which uses the gradient statistics of the data points falling into each split region. Among them, L and R represent the left and right sub-regions after splitting, T represents the entire region before splitting, λ and γ are regularization parameters respectively, and the gain obtained by splitting is given by the following formula:

[0140]

[0141] For an imbalanced data multi-class recognition model, it is usually necessary to perform macro-average calculation and micro-average calculation on evaluation indicators such as accuracy, precision, recall, F-score, and AUC value. The indicators are calculated based on the following confusion matrix.

[0142] (1) Accuracy: It represents the proportion of the number of correctly recognized risk levels of a certain class in the total number of samples (i.e., the sum of all elements in the confusion matrix). Taking the low-risk category as an example for calculation, and the same for other categories.

[0143]

[0144] (2) Precision: It represents the proportion of the number of truly caused thermal runaway risk levels of a certain class in the number of recognized thermal runaway risk levels caused by a certain class of working conditions. Taking the low-risk category as an example for calculation, and the same for other categories.

[0145] Recall, also known as the duplicate check rate (recall): It represents the proportion of the number of recognized risk levels of a certain class in the number of truly risk levels of a certain class. Taking the low-risk category as an example for calculation, and the same for other categories.

[0146]

[0147] (4) Comprehensive evaluation index (F-score): It performs a weighted harmonic average of Accuracy and Precision.

[0148] Specifically, when the parameter is, it is the most common F1-score:

[0149]

[0150] (5) Macro Average calculates the evaluation index independently for each category and then takes the average. Therefore, the macro-averages of precision, recall, and F1-score can be calculated.

[0151]

[0152] 12. Label these risk samples as three categories: 0, 1, and 2. Then, use the Python programming language to perform data modeling in the Spyder environment. First, import the vehicle trajectory dataset by calling the numpy and pandas libraries, and read the feature metrics of all samples and their corresponding risk labels. Second, introduce three algorithms, namely XGBoost, LGBM, and LCE, in the Sklearn machine learning library to directly build the model. Common dataset division ratios include 70% training set and 30% test set, 80% training set and 20% test set, etc. Establish a risk driving behavior recognition model and evaluate the performance of the model. The model performance evaluation metrics are shown in Table 6.

[0153] Table 6 Performance Evaluation of Ensemble Learning Algorithms (Macro)

[0154]

[0155] Based on the lithium battery thermal runaway recognition model under transportation conditions, a lithium battery thermal runaway monitoring and early warning method under different transportation conditions is obtained, including dynamic warning thresholds and a two-level warning process.

[0156] The specific steps are as follows:

[0157] Step S1: According to the volume of the transportation box, set several monitoring segments, including environmental parameters such as the H2 concentration, CO concentration, monitoring point temperature, and pressure inside the transportation equipment; deploy sensors inside the transportation equipment for data collection and set warning thresholds.

[0158] Step S2: Start the non-contact temperature sensor to monitor the battery surface temperature, and collect the H2 concentration, CO concentration, monitoring point temperature, and environmental pressure data in real time. Compare and analyze the real-time collected data to determine whether the thermal runaway determination conditions are met.

[0159] Step S3: When the battery temperature reaches the set threshold, it is judged that a low-risk thermal runaway has occurred, and a first-level warning is triggered. At this time, the temperature value monitored by the non-contact temperature sensor rises abnormally, and the system will automatically turn on the H2 sensor, CO sensor, smoke sensor, and pressure sensor, and upload the abnormal information to the remote monitoring platform at the same time.

[0160] Step S4: After the first-level warning, start the second-level warning mechanism. Use the H2 sensor and CO sensor to monitor the gas concentration generated during the battery thermal runaway process, use the pressure sensor to monitor the environmental pressure inside the transportation box, and use the contact temperature sensor to monitor the monitoring point temperature. When any one of the monitoring data exceeds the set threshold, the system will immediately send a warning message to the remote monitoring platform and start the corresponding response process.

[0161] Step S5: When the secondary warning is triggered and the high risk of thermal runaway occurs, the system automatically selects an emergency plan and sends the disposal instructions to the driver through vehicle-mounted and ship-mounted terminals (applicable to road, railway, and waterway transportation). For air transportation, the warning information is transmitted to the cockpit through the aircraft cargo hold Wi-Fi and responds according to the predetermined emergency handling process.

[0162] Step S6: At the same time, the system activates the top cover release box and starts the fire extinguishing process through the automatic fire extinguisher with internal fire extinguishing materials to minimize the accident losses caused by thermal runaway.

[0163] Those skilled in the art can understand that, unless otherwise defined, all terms used herein (including technical terms and scientific terms) have the same meaning as the general understanding of those of ordinary skill in the field to which the present invention belongs. It should also be understood that terms such as those defined in a general dictionary should be understood to have a meaning consistent with the meaning in the context of the prior art, and will not be interpreted in an idealized or overly formal sense unless specifically defined as here.

[0164] It should be understood that the detailed description of the technical solution of the present invention by means of the preferred embodiments above is illustrative rather than restrictive. Those of ordinary skill in the art can modify the technical solutions recorded in each embodiment on the basis of reading the specification of the present invention, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of each embodiment of the present invention.

Claims

1. A monitoring and early warning system for a power lithium battery transport box, installed inside the lithium battery transport box to monitor the battery status during transportation; characterized in that: The system includes a monitoring unit, a data processing unit, a data transmission unit, a power supply unit, a display unit and a response unit; The monitoring unit includes temperature, gas, pressure and smoke sensors; The monitoring unit is deployed on the lower surface inside the box cover. The sensing device adopts a distributed design and is equipped with temperature, gas (H2, CO), smoke and pressure sensors; The data processing unit is arranged on the inner side of the transport box cover; The display unit includes a display panel and signal lights; The display unit is arranged on the outside of the transport box cover. The display panel can display the battery temperature, gas concentration, smoke concentration and the pressure inside the box inside the transport box; The display light will give an optoelectronic alarm when it is judged that the battery has a thermal runaway; The power supply unit is a lithium battery pack, The power supply unit is arranged on the inner side of the transport box cover to supply power to the entire early warning system; The data transmission unit includes a 4G module and WIFI; The data transmission unit and the data processing unit are integrated in one housing; The response unit includes a release box and platform sends alarm information; When the early warning system determines that a thermal runaway occurs inside the transport box, the release box is opened through an electrical signal, and the fire extinguishing material placed inside the box will fall to suppress the battery flame; The early warning system sends information to the terminal.

2. The system according to claim 1, characterized in that By arranging multiple temperature, gas, pressure and smoke sensors inside the transport box, the battery surface data is collected; The temperature, gas, smoke and pressure sensors are all non-contact sensors; The temperature, gas, smoke and pressure sensors are respectively encapsulated inside their respective housings.

3. The system according to claim 1, wherein The response unit installs a fire extinguishing release box under the transport box cover. When the data processing unit judges that the battery inside the transport box has a thermal runaway, an electrical signal is sent to open the release box, and the fire extinguishing material inside the release box falls to suppress the flame emitted by the lithium battery.

4. The system according to claim 1, wherein The display panel can display the battery temperature, gas concentration, smoke concentration and the pressure inside the box inside the transport box; The display light is an LED alarm light. When the data processing unit judges that the battery inside the transport box has a thermal runaway, the LED shows red and gives an optoelectronic alarm.

5. A multi-information monitoring and transportation battery thermal runaway warning method for the power lithium battery transportation box monitoring and warning system according to claim 1, characterized in that, Including the following steps: S1: Place the lithium battery in the transport box, and then perform temperature, vibration, shock and humidity treatments on the transport box; S2: Monitor the temperature, gas, smoke and pressure inside the transport box through different types of sensors of the monitoring unit; S3: Obtain various characteristic parameters of the lithium battery risk, and perform normalization processing on the acquired data; S4: Classify the risks by monitoring the multi-information status data of the lithium battery, and calculate the threshold of the lithium battery thermal runaway risk characteristic parameters under different transportation conditions; S5: Input the quantified characteristic values into the clustering algorithm to quantify the lithium battery thermal runaway risk level; S6: To improve the risk recognition accuracy and prevent false alarms, establish a data balancing - integrated learning algorithm to construct a risk recognition model for thermal runaway of power lithium batteries inside the transport box; S7. Through the dynamic threshold and secondary warning process set by the power lithium battery thermal runaway risk identification model under transportation conditions, if a thermal runaway risk of the lithium battery is identified, a warning signal is reported to the background.

6. The method according to claim 5, characterized in that, Different transportation condition states X of the battery in the shipping box in step S1 y (t) (temperature Tenv(t) T , vibration V(t), humidity S(t), shock H(t)) are X y (t) = [Tenv(t), V(t), S(t), H(t)].

7. The method according to claim 6, characterized in that, Battery thermal runaway risk state X in step S2 d (Battery temperature T bat , H2 concentration H2 in the transport box, CO concentration CO in the transport box, smoke concentration SM in the transport box, pressure P) and different states R y (t) measured by the battery under the transport conditions are as follows X d (t) = [T bat (t), H2(t), CO(t), SM(t), P(t)], R y (t) = [X y (t), X d (t)]; Extract the risk characteristic parameters of lithium battery thermal runaway under different transportation conditions, including battery temperature and temperature rise rate, H2 concentration and change rate, CO concentration and change rate, smoke concentration and change rate, pressure value and change rate, to describe the change trend of the battery ΔT = dT bat / dt; ΔH2 = dH2 / dt; ΔCO = dCO / dt; ΔSM = dSM / dt; ΔP = dP / dt; Comprehensively determine the threshold R of the dynamic change of the risk parameters of the lithium battery thermal runaway * i to further determine its risk level. When the characteristic parameter is lower than R * i it indicates that the battery is in a normal state; When the characteristic parameter is greater than R * i it indicates that the battery is in a risk state; Comprehensive feature threshold R * i is W T 、 W CO 、W SM 、W p are the weights of battery temperature, H2 concentration, CO concentration, smoke concentration and pressure respectively, and ΔT n 、ΔH 2n 、ΔCO n 、ΔSM n 、ΔP n are the characteristic values after normalization of battery temperature, H2 concentration, CO concentration, smoke concentration and pressure respectively.

8. The method according to claim 5, characterized in that In step S5, an algorithm is used for clustering to divide the risk probability levels of thermal runaway of lithium batteries when different parameters are monitored during transportation; with n samples, each sample having three dimensions, K objects are set, and the object X is compared in turn i with each cluster centroid C j for the distance d, where the j-th dimension value of X i and C j is X it and C jt CL is the L-th cluster centroid; N L is the number of samples in the L-th cluster centroid, and the centroid point is recalculated with the mean value of all objects in this class Repeat the above steps until the cluster centroids no longer change, and finally complete the clustering; at the same time, determine the number of K of the clustering accordingly; Determine the variation of the sum of squared errors within clusters with the number of clusters, Among them, SSE(k) is the sum of squared errors within the cluster, x is the data sample point, and C i is the i-th cluster where the sample point x is located, and μ is the centroid of C i of.

9. The method according to claim 5, characterized in that, Balance the data set so that the number of high-risk samples is equivalent to the number of low-risk samples; According to each minority class sample x i , using the Euclidean distance as the calculation method to calculate the distance from x i to all the remaining minority class samples, so as to obtain the size of its nearest neighbor interval range; By evaluating the degree of imbalance between categories in the dataset, a sampling rate M is set, and according to the sampling rate M, several data samples are randomly selected from the neighborhood of x i The number is determined by M, and the selected neighboring data samples are denoted as Using the selected generate new minority class samples, where λ is a random number between 0 and 1, and the newly generated data sample is x n ; and / or In step S6, a learning algorithm is adopted. First, initialize the estimated values of all samples. After determining the loss function, obtain the predicted values of the thermal runaway occurrence probabilities and the derivatives of the function. Secondly, based on the foregoing values, create a new decision tree and add its prediction results to the previous estimated values. Finally, obtain the derivative of the loss function again on the basis of the second step. (1) The objective function combines the loss function S and the regularization term Ω: The loss function measures the difference between the predicted value and the actual value, and the regularization term penalizes the complexity of the model to prevent overfitting. y i is the actual value, is the predicted value for the i-th instance, f k is the k-th tree, K is the total number of trees, and the objective function for a given step is defined as: (2) Gradient information: An improvement is made on the basis of gradient boosting. Not only the first-order gradient information is used, but also the second-order gradient is used to more accurately approximate the curvature of the loss function; (3) Decision tree construction: For each decision tree, the best split point is found by enumerating all possible split points of all features. This process is based on the calculation of the structure score, which uses the gradient statistics of the data points falling into each split region. Among them, L and R represent the left and right sub-regions after splitting, T represents the entire region before splitting, λ and γ are regularization parameters respectively, and the gain obtained by splitting is given by the following formula:

10. A method for monitoring and early warning using a monitoring and early warning system for a power lithium battery transport box, characterized in that, The lithium battery thermal runaway identification model based on transportation conditions described in any one of claims 5-9 is input into the transportation box monitoring and warning system described in any one of claims 1-4 to monitor and warn of the thermal runaway of the lithium battery during transportation.

Images