Detection system for automatically identifying thermal runaway of lithium battery

Abstract

The invention relates to the technical field of lithium battery detection, in particular to a detection system for automatically identifying thermal runaway of a lithium battery, which comprises a server side and a user side, the server comprises a data acquisition module which is used for carrying out real-time acquisition on lithium battery image data corresponding to a lithium battery thermal runaway test; the AI image recognition module is used for recognizing the state of the lithium battery at the current moment based on a preset AI image thermal runaway recognition model according to the collected image data of the lithium battery and generating a corresponding recognition result; and the recording module is used for recording the identification result and sending an informing signal to a user side when the identification result is that the state of the lithium battery at the current moment is the thermal runaway state.

Description

Technical Field

[0001] The present invention relates to the technical field of lithium battery detection, and in particular to a detection system for automatically identifying thermal runaway of a lithium battery. Background Art

[0002] With the rapid development of energy storage technology, lithium-ion batteries are widely used in various energy storage systems due to their high energy density and long life. However, lithium-ion batteries are at risk of thermal runaway during use, which may lead to serious safety accidents such as fire and explosion. Therefore, conducting thermal runaway experiments on lithium-ion batteries and evaluating their safety performance have become key links in the design and application of energy storage systems.

[0003] At present, the thermal runaway test of energy storage batteries mainly relies on traditional manual operation. Although this method can realize thermal runaway detection, it still has the following problems:

[0004] The test process before thermal runaway usually lasts about 1 to 2 hours. The tester needs to maintain a frequency of once a minute, observe the feedback value of the data acquisition instrument, adjust the output power of the heating power supply in time according to specific rules, and observe and record the experimental phenomena at any time. This not only increases the labor intensity of the tester, but also easily leads to inaccurate or missing data records due to human factors, and cannot achieve efficient and accurate identification of thermal runaway of lithium batteries.

[0005] After thermal runaway occurs, the battery will usually smoke, catch fire, or even explode. Testers need to take strict protective measures when operating test instruments on site, but there are still certain safety risks and the safety of the test process cannot be ensured. Summary of the invention

[0006] One of the purposes of the present invention is to provide a detection system for automatically identifying thermal runaway of lithium batteries, which can realize automation and accuracy in identifying thermal runaway of lithium batteries and greatly improve the safety and reliability of the testing process.

[0007] In order to achieve the above-mentioned purpose, a detection system for automatically identifying thermal runaway of lithium batteries is provided, including a service end and a user end;

[0008] The server includes:

[0009] A data acquisition module is used to collect the lithium battery image data corresponding to the lithium battery thermal runaway test in real time;

[0010] The AI ​​image recognition module is used to identify the lithium battery status at the current moment according to the collected lithium battery image data and based on the preset AI image thermal runaway recognition model, and generate the corresponding recognition result;

[0011] The recording module is used to record the recognition result and send a notification signal to the user end when the recognition result shows that the state of the lithium battery at the current moment is a thermal runaway state.

[0012] Technical principles and effects of this solution: In this solution, first, the lithium battery image data corresponding to the lithium battery thermal runaway test is collected in real time. This step realizes the full-scale collection of the lithium battery status during the lithium battery thermal runaway test. Then, the collected lithium battery image data is input into the preset AI image thermal runaway recognition model, so as to realize the recognition of the lithium battery status at the current moment and form the corresponding recognition result. By judging the recognition result, it is determined whether the corresponding recognition result is that the status of the lithium battery at the current moment is a thermal runaway state. If so, the corresponding recognition result will be recorded and fed back to the user end.

[0013] Through the cooperation of the data acquisition module, AI image recognition module and recording module, automatic identification and detection of thermal runaway of lithium batteries are realized. Compared with the existing thermal runaway detection of lithium batteries, which requires testers to collect data and observe phenomena at a preset frequency, this solution reduces the dependence on testers and improves the reliability of the testing process through the automation of data collection and thermal runaway identification.

[0014] At the same time, after the thermal runaway state is identified, the identification result will be recorded and sent to the user end, so that the tester can remotely control the test process, avoiding the tester's arrival at the test site for field test data collection, improving the safety of the test process, and there is no need for testers to be on duty. The thermal runaway phenomenon will be automatically identified and recorded, greatly shortening the tester's participation time at the relatively dangerous test site.

[0015] Furthermore, the server also includes:

[0016] A historical data acquisition module is used to acquire the corresponding historical lithium battery image data set from the historical database;

[0017] A data labeling module, used to label each historical lithium battery image data in the acquired historical lithium battery image data set, wherein the labeling content is fireworks and non-fireworks;

[0018] Model building module, used to build the corresponding AI image thermal runaway recognition model;

[0019] A data partitioning module is used to divide the data in the historical lithium battery image data set with annotated content into a training set and a validation set according to a preset ratio;

[0020] The model training module is used to iteratively train the AI ​​image thermal runaway recognition model based on the historical lithium battery image data with annotated content in the training set and the preset model training strategy until the training is completed and output the trained AI image thermal runaway recognition model;

[0021] The model verification module is used to input the historical lithium battery image data with annotated content in the verification set into the AI ​​image thermal runaway recognition model after training, output the corresponding historical verification prediction results, and based on the historical verification prediction results, judge whether the AI ​​image thermal runaway recognition model after training is feasible. If not, repeat the model training module, otherwise, output the corresponding AI image thermal runaway recognition model.

[0022] Beneficial effects: In this solution, through the historical data acquisition module, the system can obtain a large amount of high-quality training and verification data, which improves the generalization ability and robustness of the model.

[0023] The data labeling module ensures the accuracy and consistency of the labeled data, providing a solid foundation for model training. The data partitioning module ensures that the data distribution of the training set and the validation set is consistent, avoiding overfitting and underfitting problems.

[0024] The model verification module ensures that the performance of the model meets the expected standards through testing of an independent verification set, thereby improving the reliability and credibility of the model.

[0025] Furthermore, the preset model training strategy is:

[0026] S10, randomly generating N processing flows according to the corresponding preprocessing operation set in the historical database, wherein the processing flow is a preset number of preprocessing operations randomly selected from the preprocessing operation set;

[0027] S20, pre-processing the historical lithium battery image data with annotated content in the training set according to the randomly generated N processing flows, to obtain N training sets that have undergone different processing flows;

[0028] S30, randomly generating a number of parameter groups corresponding to AI image thermal runaway recognition models;

[0029] S40, inputting the training sets that have undergone different processing flows into the AI ​​image thermal runaway recognition model corresponding to each parameter group, and outputting the recognition prediction results corresponding to each parameter group corresponding to the training sets of each different processing flow;

[0030] S50, according to the recognition prediction results corresponding to each parameter group corresponding to the training set of each different processing flow, and the corresponding annotation content, calculate the loss value of the result output by the AI ​​image thermal runaway recognition model corresponding to each participating group corresponding to the training set of each different processing flow, and sort the loss values ​​of each participating group under each different processing flow in order from small to large, to form a loss value arrangement table corresponding to each different processing flow;

[0031] S60, according to the loss value arrangement table corresponding to each different processing flow, determine the proportion of each loss value in the loss value arrangement table that is less than the preset loss threshold value. If the proportion exceeds the preset proportion value, select the parameter groups ranked before the pre-ranking of each parameter group under the corresponding processing flow, and randomly select one of the parameter groups to form the AI ​​image thermal runaway recognition model corresponding to the participating group. Otherwise, according to each parameter group under the corresponding processing flow, based on the preset parameter adjustment strategy, adjust each participating group under the corresponding processing flow. After the adjustment is completed, re-execute S40 for each participating group under the corresponding processing flow;

[0032] S70. After the parameter groups under all processing flows form corresponding AI image thermal runaway recognition models, a corresponding AI image thermal runaway recognition model set is formed, and each AI image thermal runaway recognition model in the AI ​​image thermal runaway recognition model set is respectively associated with a corresponding processing flow.

[0033] Beneficial effects: In this solution, N processing flows are randomly generated from the preprocessing operation set, each processing flow contains a preset number of preprocessing operations, and the generated N processing flows are used to preprocess the historical lithium battery image data with annotated content in the training set, respectively, to obtain N training sets that have undergone different processing flows. By randomly generating multiple preprocessing flows and parameter groups, different preprocessing methods and model parameter configurations can be fully explored to avoid falling into a local optimal solution.

[0034] Different preprocessing methods can enhance the model's adaptability to different image changes and improve the model's robustness and generalization ability.

[0035] By sorting loss values ​​and adjusting parameters, the model with the best performance is automatically selected and optimized, reducing the workload of manual parameter adjustment.

[0036] Further, the model verification module includes:

[0037] A data processing module, used to process the historical lithium battery image data with annotated content in the verification set according to the processing flow associated with each AI image thermal runaway recognition model in the AI ​​image thermal runaway recognition model set;

[0038] An input module is used to input the processed historical lithium battery image data with annotated content in the verification set into the AI ​​image thermal runaway recognition model associated with the processing flow, and output the model output results corresponding to each processing flow;

[0039] A data fusion module is used to fuse the output results of each model corresponding to the same lithium battery image data according to the model output results corresponding to each processing flow and based on the preset model weight values ​​corresponding to each processing flow to form the corresponding model final output result;

[0040] The statistical module is used to calculate the accuracy of the final output of the model and the annotation content according to the final output of the model corresponding to each lithium battery image data and the corresponding annotation content, and judge the relationship between the accuracy and the preset accuracy threshold. If the accuracy is greater than or equal to the preset accuracy threshold, the corresponding AI image thermal runaway recognition model set is a feasible model. Otherwise, the model training module is re-executed. In this embodiment, in order to better

[0041] Beneficial effect: In this scheme, when verifying each AI image thermal runaway recognition model in the trained AI image thermal runaway recognition model set, the first step is to perform corresponding processing on the historical lithium battery image data in the verification set according to the processing flow associated with each model. At this time, the verification set data after multiple processing flows will be generated, and then these data will be input into the AI ​​image thermal runaway recognition models associated with their respective processing flows to obtain the corresponding model output results. Then, according to the preset model weight value corresponding to the fusion of the output results corresponding to the models associated with each processing flow, the model output results corresponding to the same data after different processing flows are fused to form the final output result data of the model. The accuracy of the data and the annotation content is calculated to determine the feasibility of the corresponding entire AI image thermal runaway recognition model set.

[0042] Through the data fusion module, the output results of multiple models can be combined to reduce the error of a single model and improve the accuracy of the final output results. The entire verification process is automated, reducing manual intervention and improving verification efficiency and consistency. Through a variety of preprocessing methods and model combinations, the model's adaptability to different image changes is enhanced, making it more stable and effective in practical applications.

[0043] Furthermore, the preset parameter adjustment strategy is:

[0044] According to each participating group under the corresponding processing flow, based on the loss value arrangement table corresponding to the processing flow, the participating groups corresponding to the loss values ​​in the loss value arrangement table are numbered positively and negatively; the positive and negative numbers are to sort the participating groups corresponding to the loss values ​​according to the sorting of the loss values, and to number each participating group in the order from front to back and from back to front at the same time; at this time, each participating group corresponds to two numbers, namely the first number and the second number;

[0045] Pair each participating group in pairs according to the order of the first number and the second number from small to large, forming multiple parameter pairing groups. The pairing principle is that the first number of one participating group is the same as the second number of another participating group. If the first number and the second number of a participating group cannot be matched with other participating groups, a participating group will be randomly selected for pairing.

[0046] According to the multiple parameter pairing groups formed, the two participating groups in each parameter pairing group randomly generate a random number P from 0 to n. 1i and a P from 0 to m 2i , based on the update probability calculation formula, calculate the update probability corresponding to each parameter pairing group;

[0047] The update probability calculation formula is:

[0048]

[0049] Where U i Is the update of the two competing groups in the i-th parameter pairing group, 1 means update, 0 means no update, P 1i is the random number corresponding to the first number of the participating group in the i-th parameter pairing group, P 2i is the random number corresponding to the second number of the participating group in the i-th parameter pairing group that is the same as the first number, P s is the random probability; m>n;

[0050] In a certain participating pairing group corresponding to the U i =1, select update segments X1 and X2 with the same parameter type from the two participating groups, X1 is the update segment corresponding to the participating group with a smaller first number, and X2 is the update segment corresponding to the participating group with a larger first number;

[0051] In P 1i <P 2i When the update segment X2 is updated to the update segment X1, otherwise, the segments of the preset proportion in the update segment X2 are updated to the corresponding segments in the update segment X1.

[0052] Beneficial effect: By positively numbering the groups with lower loss values ​​and pairing them with other groups, it can ensure that the parameters with better performance have the opportunity to be adopted or partially adopted by other groups. This can accelerate the spread of excellent parameters throughout the system. 1i , P 2i , and combined with the random probability P s To decide whether to update the parameters, it can avoid premature convergence to the local optimal solution to a certain extent. This mechanism helps to explore more solution spaces and increase the diversity of parameter configurations.

[0053] The update rule allows 1i <P 2i When X1 is directly used to replace X2, and in P 1i ≥P 2i This allows for flexibility in adjusting parameters in different situations, by replacing only part of the solution when the uncertainty is high. This design allows for the rapid adoption of significantly better solutions, while also allowing for a more cautious approach when uncertainty is high.

[0054] Although there is a random element involved in the process, the loss-based sorting and pairing rules ensure that even in the case of random selection, the parameter settings with better performance are retained or optimized. This reduces the negative effects that may be caused by completely random selection. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 This is a logic block diagram of a detection system for automatically identifying thermal runaway of a lithium battery in Embodiment 1 of the present invention.

[0056] Figure 2 This is a training, verification and evaluation diagram of the AI ​​image thermal runaway recognition model corresponding to another embodiment of the present invention. DETAILED DESCRIPTION

[0057] The following is further described in detail through specific implementation methods:

[0058] Embodiment 1

[0059] A detection system for automatically identifying thermal runaway of lithium batteries, basically as follows Figure 1 As shown, it includes a server and a user;

[0060] The server includes:

[0061] The data acquisition module is used to collect the lithium battery image data corresponding to the lithium battery thermal runaway test in real time; in this embodiment, a high-resolution camera or other image acquisition equipment is used to record the lithium battery in real time.

[0062] The AI ​​image recognition module is used to identify the lithium battery status at the current moment according to the collected lithium battery image data and based on the preset AI image thermal runaway recognition model, and generate the corresponding recognition result;

[0063] The recording module is used to record the recognition result and send a notification signal to the user end when the recognition result shows that the state of the lithium battery at the current moment is a thermal runaway state.

[0064] The server also includes:

[0065] A historical data acquisition module is used to acquire the corresponding historical lithium battery image data set from the historical database;

[0066] A data labeling module, used to label each historical lithium battery image data in the acquired historical lithium battery image data set, wherein the labeling content is fireworks and non-fireworks;

[0067] The model building module is used to build the corresponding AI image thermal runaway recognition model; the model can be built using BP neural network technology, convolutional neural network or YOLOv8 network. This embodiment uses YOLOv8 network construction.

[0068] A data partitioning module is used to divide the data in the historical lithium battery image data set with annotated content into a training set and a validation set according to a preset ratio;

[0069] The model training module is used to iteratively train the AI ​​image thermal runaway recognition model based on the historical lithium battery image data with annotated content in the training set and the preset model training strategy until the training is completed and output the trained AI image thermal runaway recognition model;

[0070] The preset model training strategy is:

[0071] S10, randomly generating N processing flows according to the corresponding preprocessing operation set in the historical database, wherein the processing flow is a preset number of preprocessing operations randomly selected from the preprocessing operation set;

[0072] S20, pre-processing the historical lithium battery image data with annotated content in the training set according to the randomly generated N processing flows, to obtain N training sets that have undergone different processing flows;

[0073] S30, randomly generating a number of parameter groups corresponding to AI image thermal runaway recognition models;

[0074] S40, inputting the training sets that have undergone different processing flows into the AI ​​image thermal runaway recognition model corresponding to each parameter group, and outputting the recognition prediction results corresponding to each parameter group corresponding to the training sets of each different processing flow;

[0075] S50, according to the recognition prediction results corresponding to each parameter group corresponding to the training set of each different processing flow, and the corresponding annotation content, calculate the loss value of the result output by the AI ​​image thermal runaway recognition model corresponding to each participating group corresponding to the training set of each different processing flow, and sort the loss values ​​of each participating group under each different processing flow in order from small to large, to form a loss value arrangement table corresponding to each different processing flow;

[0076] S60, according to the loss value arrangement table corresponding to each different processing flow, determine the proportion of each loss value in the loss value arrangement table that is less than the preset loss threshold value. If the proportion exceeds the preset proportion value, select the parameter groups ranked before the pre-ranking of each parameter group under the corresponding processing flow, and randomly select one of the parameter groups to form the AI ​​image thermal runaway recognition model corresponding to the participating group. Otherwise, according to each parameter group under the corresponding processing flow, based on the preset parameter adjustment strategy, adjust each participating group under the corresponding processing flow. After the adjustment is completed, re-execute S40 for each participating group under the corresponding processing flow;

[0077] The preset parameter adjustment strategy is:

[0078] According to each participating group under the corresponding processing flow, based on the loss value arrangement table corresponding to the processing flow, the participating groups corresponding to the loss values ​​in the loss value arrangement table are labeled positively and negatively; the positive and negative labels are based on the sorting of the loss values, and the participating groups corresponding to the loss values ​​are also sorted, and each participating group is labeled at the same time in the order from front to back and from back to front; at this time, each participating group corresponds to two numbers, namely the first number and the second number; in this embodiment, for example, there are 6 participating groups, then when marking the first number and the second number, the first number corresponding to a certain participating group is 1, then the second number is 6.

[0079] According to the order of the first number and the second number from small to large, each participating group is paired in pairs to form multiple parameter pairing groups. The pairing principle is that the first number of one participating group is the same as the second number of another participating group; if the first number and the second number of a participating group cannot be matched with other participating groups, a participating group is randomly selected for pairing; in this embodiment, for example, the first number of participating group A is 2 and the second number is 5, and the first number of participating group B is 5 and the second number is 2. At this time, the first number of participating group A and the second number of participating group B are both 2, then participating group A and participating group B are matched into a parameter pairing group. When the number of the entire participating group is an even number, each participating group can be matched with the corresponding participating group, but when the number of participating groups is an odd number, the first number and the second number of the middle parameter group are the same and cannot be matched with other participating groups. At this time, a random selection is made, which greatly improves the randomness of the matching.

[0080] According to the multiple parameter pairing groups formed, the two participating groups in each parameter pairing group randomly generate a random number P from 0 to n. 1i and a P from 0 to m 2i , based on the update probability calculation formula, calculate the update probability corresponding to each parameter pairing group;

[0081] The update probability calculation formula is:

[0082]

[0083] Where U i Is the update of the two competing groups in the i-th parameter pairing group, 1 means update, 0 means no update, P 1i is the random number corresponding to the first number of the participating group in the i-th parameter pairing group, P 2i is the random number corresponding to the second number of the participating group in the i-th parameter pairing group that is the same as the first number, P s is the random probability; m>n;

[0084] In a certain participating pairing group corresponding to the U i =1, select update segments X1 and X2 with the same parameter type from the two participating groups, X1 is the update segment corresponding to the participating group with a smaller first number, and X2 is the update segment corresponding to the participating group with a larger first number;

[0085] In P 1i <P 2i , then the update segment X2 is updated to the update segment X1, otherwise, the preset proportion of segments in the update segment X2 is updated to the corresponding segments in the update segment X1. In this embodiment, for example, in a certain matching group, the update segments corresponding to the competition group A and the competition group B are respectively (x11 , x 12 , x 13 , x 14 , x 15 , x 16 ) and (x 21 , x 22 , x 23 , x 24 , x 25 , x 26 );

[0086] When updating:

[0087] If P 1i <P 2i , then (x 21 , x 22 , x 23 , x 24 , x 25 , x 26 )All updated to (x 11 , x 12 , x 13 , x 14 , x 15 , x 16 );

[0088] If P 1i ≥P 2i , then partial updates are performed based on the preset ratio;

[0089] For example, if the preset ratio is one-third, one-third of the segments are randomly selected, for example (x 22 , x 24 ), then directly convert (x 21 , x 22 , x 23 , x 24 , x 25 , x 26 ) is updated to (x 21 , x 12 , x 23 , x 14 , x 25 , x 26 ).

[0090] S70. After the parameter groups under all processing flows form corresponding AI image thermal runaway recognition models, a corresponding AI image thermal runaway recognition model set is formed, and each AI image thermal runaway recognition model in the AI ​​image thermal runaway recognition model set is respectively associated with a corresponding processing flow.

[0091] In another embodiment, the corresponding preset model training strategy is to perform full-process preprocessing on the data in the training set according to the preprocessing operation set, that is, to perform full-process preprocessing operations on the historical lithium battery image data in the training set;

[0092] Then, the pre-processed training set data of the entire process is input into the corresponding AI image thermal runaway recognition model, and the model is iteratively trained;

[0093] After each iteration of training is completed, the corresponding AI image thermal runaway recognition model is output, and the historical battery image data with annotated content in the verification set is input into the corresponding model to verify the training effect of the trained AI image thermal runaway recognition model to determine whether the training effect meets the standard. If not, continue training. If so, stop training, and retrieve other historical lithium battery image data with annotated content from the database and input them into the AI ​​image thermal runaway recognition model. The feasibility of the model is determined based on the output test results. If it is not feasible, continue training, otherwise output the corresponding model. In this embodiment, in order to better describe the training and verification effects, as shown in FIG. Figure 2 As shown in the figure, the bounding box loss, category loss and distance focus loss in the training phase and the verification phase are plotted respectively. At the same time, the precision and recall rate in the training phase and the average precision in the verification phase are also plotted. Through these data, we can know that the corresponding model is constantly converging.

[0094] The model verification module is used to input the historical lithium battery image data with annotated content in the verification set into the AI ​​image thermal runaway recognition model after training, output the corresponding historical verification prediction results, and based on the historical verification prediction results, judge whether the AI ​​image thermal runaway recognition model after training is feasible. If not, repeat the model training module, otherwise, output the corresponding AI image thermal runaway recognition model.

[0095] The model verification module includes:

[0096] A data processing module, used to process the historical lithium battery image data with annotated content in the verification set according to the processing flow associated with each AI image thermal runaway recognition model in the AI ​​image thermal runaway recognition model set;

[0097] An input module is used to input the processed historical lithium battery image data with annotated content in the verification set into the AI ​​image thermal runaway recognition model associated with the processing flow, and output the model output results corresponding to each processing flow;

[0098] A data fusion module is used to fuse the output results of each model corresponding to the same lithium battery image data according to the model output results corresponding to each processing flow and based on the preset model weight values ​​corresponding to each processing flow to form the corresponding model final output result;

[0099] The statistical module is used to calculate the accuracy of the final output results of the model and the annotation content corresponding to the final output results of the model corresponding to each lithium battery image data and the corresponding annotation content, and to determine the relationship between the accuracy and the preset accuracy threshold. If the accuracy is greater than or equal to the preset accuracy threshold, the corresponding AI image thermal runaway recognition model set is a feasible model. Otherwise, the model training module is re-executed.

[0100] The above is only an embodiment of the present invention. The common sense such as the known specific structure and characteristics in the scheme is described too much here. The ordinary technicians in the relevant field know all the common technical knowledge in the technical field of the invention before the application date or priority date, can know all the existing technologies in the field, and have the ability to apply the conventional experimental means before that date. The ordinary technicians in the relevant field can improve and implement this scheme in combination with their own abilities under the enlightenment given by this application. Some typical known structures or known methods should not become obstacles for ordinary technicians in the relevant field to implement this application. It should be pointed out that for those skilled in the art, without departing from the structure of the present invention, several deformations and improvements can be made, which should also be regarded as the protection scope of the present invention, which will not affect the effect of the implementation of the present invention and the practicality of the patent. The protection scope required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.

Claims

1. A detection system for automatically identifying thermal runaway of lithium batteries, characterized by: Including server and user side; The server includes: A data acquisition module is used to collect the lithium battery image data corresponding to the lithium battery thermal runaway test in real time; The AI ​​image recognition module is used to identify the lithium battery status at the current moment according to the collected lithium battery image data and based on the preset AI image thermal runaway recognition model, and generate the corresponding recognition result; The recording module is used to record the recognition result and send a notification signal to the user end when the recognition result shows that the state of the lithium battery at the current moment is a thermal runaway state.

2. The automatic detection system for identifying thermal runaway of a lithium battery according to claim 1, characterized in that: The server also includes: A historical data acquisition module is used to acquire the corresponding historical lithium battery image data set from the historical database; A data labeling module, used to label each historical lithium battery image data in the acquired historical lithium battery image data set, wherein the labeling content is fireworks and non-fireworks; Model building module, used to build the corresponding AI image thermal runaway recognition model; A data partitioning module is used to divide the data in the historical lithium battery image data set with annotated content into a training set and a validation set according to a preset ratio; The model training module is used to iteratively train the AI ​​image thermal runaway recognition model based on the historical lithium battery image data with annotated content in the training set and the preset model training strategy until the training is completed and output the trained AI image thermal runaway recognition model; The model verification module is used to input the historical lithium battery image data with annotated content in the verification set into the AI ​​image thermal runaway recognition model after training, output the corresponding historical verification prediction results, and based on the historical verification prediction results, judge whether the AI ​​image thermal runaway recognition model after training is feasible. If not, repeat the model training module, otherwise, output the corresponding AI image thermal runaway recognition model.

3. The automatic detection system for identifying thermal runaway of a lithium battery according to claim 2, characterized in that: The preset model training strategy is: S10, randomly generating N processing flows according to the corresponding preprocessing operation set in the historical database, wherein the processing flow is a preset number of preprocessing operations randomly selected from the preprocessing operation set; S20, pre-processing the historical lithium battery image data with annotated content in the training set according to the randomly generated N processing flows, to obtain N training sets that have undergone different processing flows; S30, randomly generating a number of parameter groups corresponding to AI image thermal runaway recognition models; S40, inputting the training sets that have undergone different processing flows into the AI ​​image thermal runaway recognition model corresponding to each parameter group, and outputting the recognition prediction results corresponding to each parameter group corresponding to the training sets of each different processing flow; S50, according to the recognition prediction results corresponding to each parameter group corresponding to the training set of each different processing flow, and the corresponding annotation content, calculate the loss value of the result output by the AI ​​image thermal runaway recognition model corresponding to each participating group corresponding to the training set of each different processing flow, and sort the loss values ​​of each participating group under each different processing flow in order from small to large, to form a loss value arrangement table corresponding to each different processing flow; S60, according to the loss value arrangement table corresponding to each different processing flow, determine the proportion of each loss value in the loss value arrangement table that is less than the preset loss threshold value. If the proportion exceeds the preset proportion value, select the parameter groups ranked before the pre-ranking of each parameter group under the corresponding processing flow, and randomly select one of the parameter groups to form the AI ​​image thermal runaway recognition model corresponding to the participating group. Otherwise, according to each parameter group under the corresponding processing flow, based on the preset parameter adjustment strategy, adjust each participating group under the corresponding processing flow. After the adjustment is completed, re-execute S40 for each participating group under the corresponding processing flow; S70. After the parameter groups under all processing flows form corresponding AI image thermal runaway recognition models, a corresponding AI image thermal runaway recognition model set is formed, and each AI image thermal runaway recognition model in the AI ​​image thermal runaway recognition model set is respectively associated with a corresponding processing flow.

4. The automatic detection system for identifying thermal runaway of a lithium battery according to claim 3, characterized in that: The model verification module includes: A data processing module, used to process the historical lithium battery image data with annotated content in the verification set according to the processing flow associated with each AI image thermal runaway recognition model in the AI ​​image thermal runaway recognition model set; An input module is used to input the processed historical lithium battery image data with annotated content in the verification set into the AI ​​image thermal runaway recognition model associated with the processing flow, and output the model output results corresponding to each processing flow; A data fusion module is used to fuse the output results of each model corresponding to the same lithium battery image data according to the model output results corresponding to each processing flow and based on the preset model weight values ​​corresponding to each processing flow to form the corresponding model final output result; The statistical module is used to calculate the accuracy of the final output results of the model and the annotation content corresponding to the final output results of the model corresponding to each lithium battery image data and the corresponding annotation content, and to determine the relationship between the accuracy and the preset accuracy threshold. If the accuracy is greater than or equal to the preset accuracy threshold, the corresponding AI image thermal runaway recognition model set is a feasible model. Otherwise, the model training module is re-executed.

5. The automatic detection system for identifying thermal runaway of a lithium battery according to claim 4, characterized in that: The preset parameter adjustment strategy is: According to each participating group under the corresponding processing flow, based on the loss value arrangement table corresponding to the processing flow, the participating groups corresponding to the loss values ​​in the loss value arrangement table are numbered positively and negatively; the positive and negative numbers are to sort the participating groups corresponding to the loss values ​​according to the sorting of the loss values, and to number each participating group in the order from front to back and from back to front at the same time; at this time, each participating group corresponds to two numbers, namely the first number and the second number; Pair each participating group in pairs according to the order of the first number and the second number from small to large, forming multiple parameter pairing groups. The pairing principle is that the first number of one participating group is the same as the second number of another participating group. If the first number and the second number of a participating group cannot be matched with other participating groups, a participating group will be randomly selected for pairing. According to the multiple parameter pairing groups formed, the two participating groups in each parameter pairing group randomly generate a random number P from 0 to n. 1i and a P from 0 to m 2i , based on the update probability calculation formula, calculate the update probability corresponding to each parameter pairing group; The update probability calculation formula is: Where U i Is the update of the two competing groups in the i-th parameter pairing group, 1 means update, 0 means no update, P 1i is the random number corresponding to the first number of the participating group in the i-th parameter pairing group, P 2i is the random number corresponding to the second number of the participating group in the i-th parameter pairing group that is the same as the first number, P s is the random probability; m>n; In a certain participating pairing group corresponding to the U i =1, select update segments X1 and X2 with the same parameter type from the two participating groups, X1 is the update segment corresponding to the participating group with a smaller first number, and X2 is the update segment corresponding to the participating group with a larger first number; In P 1i <P 2i When the update segment X2 is updated to the update segment X1, otherwise, the segments of the preset proportion in the update segment X2 are updated to the corresponding segments in the update segment X1.

Images