Battery pole full-life cycle safety management method based on multi-modal fusion

Abstract

The application provides a battery pole charging full life cycle safety management method based on multi-modal fusion, and relates to the technical field of new energy pole charging. The method comprises collecting multi-modal data and environmental parameters of the battery pole charging full life cycle, integrating and processing, and correcting in combination with the environmental parameters. A three-dimensional safety state model is generated, and a safety region is divided. A battery pole charging safety risk prediction model is constructed, the independent safety sub-region is taken as the input, and the pole charging safety risk level of each region of the battery is output. The battery pole charging full life cycle is classified and safety controlled. The application reflects the battery state through multiple dimensions, completely eliminates the safety monitoring blind area, generates a three-dimensional safety state model through battery state inversion and divides the region, can accurately identify the safety risk level of each region of the battery, realizes the transformation from passive monitoring to active prediction, and provides all-round and full-cycle protection for new energy battery pole charging safety.

Description

Technical Field

[0001] This invention relates to the field of new energy charging technology, and in particular to a method for full life-cycle safety management of battery charging based on multimodal fusion. Background Technology

[0002] With the increasing popularity of new energy vehicles in the transportation sector, users' demand for charging efficiency has driven the rapid development of ultra-fast charging technology. However, the safety management issues throughout the entire lifecycle of batteries using ultra-fast charging are becoming increasingly prominent. Current safety management solutions for new energy batteries have significant limitations. Most rely on single types of monitoring data, focusing only on local information such as battery electrical performance parameters or surface temperature. They cannot comprehensively capture the correlation between changes in the battery's internal structure, external environmental influences, and performance degradation during ultra-fast charging, resulting in blind spots in safety monitoring.

[0003] Traditional management methods fail to consider the dynamic characteristics of batteries throughout their entire lifecycle. At different extreme charging stages—pre-charging, constant current, and constant voltage—batteries exhibit varying sensitivities to environmental temperature, humidity, and air pressure, yet a uniform monitoring and control strategy is employed, making it difficult to address safety risks in complex scenarios. Furthermore, existing solutions lack effective multi-source data integration and environmental impact correction mechanisms. When environmental conditions fluctuate, monitoring data is prone to bias, leading to misjudgments or missed risks. In addition, most safety management methods focus only on real-time monitoring during charging, failing to identify potential risks in advance. Moreover, tiered control measures lack specificity, either excessively consuming resources or struggling to address high-risk scenarios, thus failing to meet the safety management needs of new energy batteries throughout their entire extreme charging lifecycle. Summary of the Invention

[0004] This invention provides a battery charging full life cycle safety management method based on multimodal fusion to address the deficiencies in existing technologies.

[0005] This invention provides a battery charging lifecycle safety management method based on multimodal fusion, including:

[0006] Collect multimodal data and environmental parameters throughout the entire life cycle of the battery. The multimodal data includes battery electrical performance parameters, thermal characteristic data, and structural state information, while the environmental parameters include ambient temperature, humidity, and air pressure.

[0007] Multimodal data is preprocessed, and an adaptive weighted fusion algorithm is used to integrate the multimodal data. The integration result is then corrected based on environmental parameters to obtain fused secure data.

[0008] Based on the fused safety data, a three-dimensional safety state model of the battery's entire life cycle is generated using the battery state inversion method.

[0009] Based on the consistency of the battery performance parameters and the stability of the structure state in the three-dimensional safety state model, the three-dimensional safety state model is divided into a safety region to obtain M independent safety sub-regions.

[0010] Historical battery pole charging safety event data is collected, a battery pole charging safety risk prediction model based on a deep belief network is constructed, and an improved bald eagle search algorithm is used to optimize the hyperparameters of the battery pole charging safety risk prediction model, the independent safety sub-regions are taken as inputs, and the pole charging safety risk levels of the battery regions are output, the pole charging safety risk levels include low risk, medium risk and high risk.

[0011] According to the pole charging safety risk level, the battery pole charging full life cycle is classified and safety controlled.

[0012] According to the battery pole charging full life cycle safety management method based on multi-modal fusion provided by the application, the battery performance parameters include charging voltage, charging current, internal resistance, capacity attenuation rate and battery state of charge. The thermal characteristic data includes battery surface temperature distribution, temperature rise rate, thermal diffusion coefficient and heat dissipation efficiency. The structure state information includes the deformation amount of the pole piece, the integrity of the diaphragm, the electrolyte loss rate and the sealing performance of the shell.

[0013] According to the battery pole charging full life cycle safety management method based on multi-modal fusion provided by the application, the process of integrating and processing the multi-modal data by using the adaptive weighted fusion algorithm includes:

[0014] The abnormal values in the multi-modal data are eliminated by the Grubbs criterion.

[0015] The data integration is carried out by using the adaptive weighted fusion algorithm, and a data fusion weight distribution model is established.

[0016] The fusion result is corrected through the weight iterative optimization process.

[0017] According to the battery pole charging full life cycle safety management method based on multi-modal fusion provided by the application, the integrated result is corrected in combination with the environmental parameters to obtain the fusion safety data, and the process includes:

[0018] An environmental parameter influence factor matrix is constructed, which is used to quantify the comprehensive influence degree of environmental temperature, humidity and air pressure on the multi-modal data fusion result.

[0019] An environmental correction function based on fuzzy reasoning is established, and the environmental parameter influence factor is mapped to a correction coefficient.

[0020] The environmental correction function is coupled with the multi-modal data fusion result to obtain the corrected fusion safety data.

[0021] Smooth filtering processing is performed on the modified fusion safety data to eliminate noise caused by environmental parameter mutation, and the fusion safety data is output.

[0022] According to the battery full life cycle safety management method based on multi-modal fusion provided by the application, the process of generating the three-dimensional safety state model of the battery full life cycle by using the battery state inversion method includes:

[0023] The battery state inversion method based on Bayesian regularization is used to construct an inversion objective function with the fusion safety data as a constraint condition.

[0024] The posterior probability distribution of the model parameters is obtained by solving the inversion objective function through the improved particle swarm-simulated annealing hybrid algorithm.

[0025] Based on the maximum a posteriori estimation result of the posterior probability distribution, a three-dimensional grid safety model of the battery full life cycle is constructed to generate a three-dimensional safety state model.

[0026] According to the battery full life cycle safety management method based on multi-modal fusion provided by the application, the process of dividing the three-dimensional safety state model into safety regions includes:

[0027] The capacity attenuation rate in the battery electrical performance parameter and the temperature rise rate in the thermal characteristic data are taken as core division indexes to calculate the safety characteristic difference coefficient of any two grid units in the three-dimensional safety state model.

[0028] The difference coefficient threshold is set, and when the difference coefficient of the two grid units is not greater than the difference coefficient threshold, the two grid units are determined to be divided into the same safety region.

[0029] The region where the separator integrity and the shell sealing performance in the structure state information are lower than the preset threshold is divided into different safety sub-regions.

[0030] After the initial division is completed by the density clustering algorithm, the region boundary is subjected to smoothing optimization processing to obtain M independent safety sub-regions.

[0031] According to the battery full life cycle safety management method based on multi-modal fusion provided by the application, the historical battery full life cycle safety event data includes battery characteristic data corresponding to the event occurrence region, including the electrical performance parameter, the thermal characteristic data and the structure state information of the region, and simultaneously including the occurrence time, the full life cycle stage, the safety accident type and the corresponding risk probability value of the event.

[0032] According to the battery full life cycle safety management method based on multi-modal fusion provided by the application, the process of constructing the battery full life cycle safety risk prediction model based on the deep belief network includes:

[0033] The deep belief network base model is designed, including an input layer, a plurality of restricted Boltzmann machine hidden layers and an output layer. The input layer is used for receiving independent safety sub-region feature data, and the independent safety sub-region feature data includes electrical performance parameters, thermal characteristic data and structure state information of the independent safety sub-region. The plurality of hidden layers are used for feature extraction and nonlinear mapping of the data received by the input layer. The output layer is used for outputting the extreme charge safety risk probability value of the independent safety sub-region.

[0034] The region feature data corresponding to the event occurrence region in the historical battery extreme charge safety event data is taken as input, and the risk probability value corresponding to the event is taken as output, and the deep belief network base model is pre-trained, and the model parameters meeting the preset accuracy are retained, to obtain a battery extreme charge safety risk prediction model based on a deep belief network.

[0035] According to the battery extreme charge full life cycle safety management method based on multi-modal fusion provided by the application, the process of optimizing the hyperparameters of the battery extreme charge safety risk prediction model by using the improved bald eagle search algorithm includes:

[0036] The learning rate, the number of hidden layer neurons and the regularization coefficient of the battery extreme charge safety risk prediction model form a hyperparameter combination.

[0037] Initialize the bald eagle population, and the position of each bald eagle individual represents a set of hyperparameters.

[0038] The F1 score of the model is used as the fitness value.

[0039] According to the fitness value, the optimal position of each bald eagle individual and the global optimal position of the entire population are updated.

[0040] The process of initializing the population and updating the position is repeated until the preset number of iterations is reached, and the optimal hyperparameter combination is obtained.

[0041] According to the battery extreme charge full life cycle safety management method based on multi-modal fusion provided by the application, the process of grading safety control for each stage of the battery extreme charge full life cycle includes:

[0042] Periodic safety inspection is implemented for the low-risk region. Real-time monitoring is adopted for the medium-risk region, and the monitoring frequency is dynamically adjusted. A full-dimensional stereoscopic monitoring system is deployed for the high-risk region, and an emergency response mechanism is linked, and power-off protection and forced cooling measures are started immediately after triggering the high-risk early warning.

[0043] The application provides a battery pole charging full life cycle safety management method based on multi-modal fusion, which realizes comprehensive coverage of multi-modal data by collecting battery electrical performance parameters, thermal characteristic data, structural state information and environmental parameters, can reflect the battery state from the dimensions of electricity, heat, structure and environment, completely eliminates the safety monitoring blind area, and provides complete data support for subsequent safety management. The adaptive weighted fusion algorithm and the double-factor dynamic correction model are adopted, which not only can efficiently integrate multi-modal data, but also can accurately correct the fusion results combined with environmental parameters and pole charging stage characteristics, and further improve the data accuracy through the deviation feedback calibration mechanism, avoid the monitoring deviation caused by environmental fluctuations, and ensure the reliability of subsequent model construction and risk prediction. A three-dimensional safety state model is generated by battery state inversion and region division, a risk prediction model is constructed combined with a deep belief network and an improved bald eagle search algorithm, which can accurately identify the safety risk level of each region of the battery, realizes the change from passive monitoring to active prediction, and avoids potential safety hazards in advance. The hierarchical safety control strategy implemented based on the risk level adopts differentiated control measures for different risk regions, which not only ensures the safety redundancy of high-risk regions, but also avoids the waste of resources in low-risk regions, simultaneously covers the full life cycle of battery pole charging, ensures the pertinence and effectiveness of safety management in each stage, provides all-round and full-cycle protection for new energy battery pole charging safety, and promotes the safe application and popularization of pole charging technology in the field of short-distance travel. BRIEF DESCRIPTION OF DRAWINGS

[0044] In order to more clearly illustrate the technical solutions in the present application or prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0045] Fig. 1 is a flowchart of the battery pole charging full life cycle safety management method based on multi-modal fusion provided by the embodiment of the present application;

[0046] Fig. 2 is a flowchart of the method for obtaining fused safety data in the embodiment of the present application;

[0047] Fig. 3 is a flowchart of the method for generating a three-dimensional safety state model of the battery full life cycle by using the battery state inversion method in the embodiment of the present application. DETAILED DESCRIPTION

[0048] In order to make the objects, technical solutions and advantages of the present application clearer, the technical solutions in the present application will be described clearly and completely below in conjunction with the drawings in the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the protection scope of the present application.

[0049] The present application is described below in conjunction with Figs. 1-3 The present application is described below in conjunction with

[0050] The present application is described below in conjunction with

[0051] The present application is described below in conjunction with

[0052] In the present embodiment, the data is collected by constructing a multi-source perception network, which is composed of a flexible temperature sensing unit, a micro voltage and current sensing unit, a strain sensing unit and an environmental sensing unit. The flexible temperature sensing unit is pasted on the surface of the battery and embedded in the interstice of the pole piece at 12 point positions with an interval of 2 cm; the micro voltage and current sensing unit is connected in series in the charging circuit and in parallel to the positive and negative poles of the battery; the strain sensing unit is pasted on the surface of the pole piece and the key stress points of the shell; and the environmental sensing unit is fixed at the ventilation position inside the battery pack.

[0053] According to the dynamic adjustment of the pole charging stage, the sampling frequency is set to 20 Hz in the pre-charging stage (0-10% SOC) and the trickle stage (90%-100% SOC), and the sampling frequency is set to 10 Hz in the constant current stage (10%-80% SOC) and the constant voltage stage (80%-90% SOC) (SOC is the state of charge of the battery).

[0054] The battery electrical performance parameters include charging voltage, charging current, internal resistance, capacity decay rate and battery state of charge. The thermal characteristic data include battery surface temperature distribution, temperature rise rate, thermal diffusion coefficient and heat dissipation efficiency. The structure state information includes pole piece deformation, diaphragm integrity, electrolyte loss rate and shell sealing performance.

[0055] The multi-modal data is pre-processed, the adaptive weighted fusion algorithm is used to integrate and process the multi-modal data, and the integrated results are corrected in combination with the environmental parameters to obtain the fusion safety data.

[0056] When preprocessing the multi-modal data, the collected original data is synchronized and aligned, and the sampling data of different sensing units is unified to the same time dimension based on the time stamp of the CAN bus.

[0057] The outliers in the single group of modal data are identified and removed by using the Grubbs criterion.

[0058] The Z-score standardization method is used for standardization processing, and each modal data is converted to the interval with mean value of 0 and standard deviation of 1.

[0059] The process of integrating the multi-modal data by using the adaptive weighted fusion algorithm includes:

[0060] The outliers in the multi-modal data are removed by using the Grubbs criterion.

[0061] The data integration is performed by using the adaptive weighted fusion algorithm, and a data fusion weight distribution model is established, which is expressed by the formula:

[0062]

[0063] In the formula, is the fusion weight of the qth modal data, is the variance matrix of the qth modal data, is the inverse matrix of the variance matrix, T is the total number of modal data types, is the standardized data of the qth modal data after preprocessing, F is the multi-modal data fusion result, N is the sample number of the qth modal data, is the nth sample value of the qth modal data, is the sample mean value of the qth modal data.

[0064] The fusion result is corrected by the weight iterative optimization process, and the iterative optimization formula is expressed as:

[0065]

[0066]

[0067] In the formula, is the weight of the qth modal data after the k+1th iteration, is the weight after the kth iteration, is the weight update step (value range 0.01-0.1), L is the fusion error loss function, is the fusion result of the nth sample, is the true reference value of the nth sample, is the partial derivative of the loss function to the kth iteration weight.

[0068] The process of correcting the integration result combined with the environmental parameters to obtain the fusion safety data includes:

[0069] An environmental parameter influence factor matrix E is constructed, which is used to quantify the environmental temperature humidity and air pressure The comprehensive influence degree of the multi-modal data fusion result F is expressed by the matrix:

[0070]

[0071] In the formula, are the normalized environmental temperature, humidity and air pressure values, , , are the corresponding environmental sensitivity coefficients, which are obtained by battery environmental adaptability experiments.

[0072] An environmental correction function based on fuzzy reasoning is established The environmental parameter influence factor matrix E is mapped to the correction coefficient, and the function expression is:

[0073]

[0074] In the formula, is the correction intensity factor (value range 0.05-0.15), is the steepness coefficient of the influence curve (value range 0.5-2.0), is the environmental reference influence factor.

[0075] The environmental correction function is coupled with the multi-modal data fusion result F to obtain the final corrected fusion safety data F', and the calculation formula is:

[0076]

[0077] In the formula, is the environmental disturbance compensation factor (value range 0.01-0.05), is the first-order difference of the multi-modal data fusion result F in the time domain, which is used to introduce the compensation effect of the dynamic change of the environment.

[0078] The corrected fusion safety data is subjected to smoothing filter processing to eliminate the noise caused by the sudden change of the environmental parameters, and the fusion safety data is output.

[0079] According to the fusion safety data, a battery full-life cycle three-dimensional safety state model is generated by using the battery state inversion method, and the process includes:

[0080] The full life cycle of the battery pole is divided into three stages: new battery stage (cycle times 0-200), middle stage (201-800), and end stage (801-1000). At least 30 sets of fusion safety data are collected as inversion basis data for each stage to ensure the performance change characteristics of each stage.

[0081] With fusion safety data as the constraint condition, a battery state inversion method based on Bayesian regularization is used to construct the inversion objective function, which is expressed as:

[0082]

[0083] In the formula, is the battery state model parameter vector, including electrode reaction rate constant and ion diffusion coefficient, Y is the fusion safety data observation vector, is the battery state forward model operator, is the regularization parameter (value range 0.001-0.01), is the model parameter regularization term, which adopts regularization form .

[0084] The improved particle swarm-simulated annealing hybrid algorithm is used to solve the inversion objective function to obtain the posterior probability distribution of the model parameters.

[0085] Based on the maximum a posteriori estimation result of the posterior probability distribution, a three-dimensional grid safety model of the battery full life cycle is constructed, the model parameters are assigned to the corresponding grid elements, and a three-dimensional safety state model is generated.

[0086] Tetrahedral mesh division is adopted, and each grid element corresponds to a small area inside the battery. The model parameters of this area obtained by inversion are assigned to the grid element. In the time dimension, each life cycle stage corresponds to a time slice (new battery stage t=0, middle stage t=0.5, and end stage t=1). Through interpolation, a three-dimensional model under different cycle times is obtained.

[0087] Based on the consistency of battery performance parameters and the stability of structural state in the three-dimensional safety state model, the three-dimensional safety state model is divided into M independent safety sub-regions.

[0088] The process of dividing the three-dimensional safety state model into safety regions includes:

[0089] Taking the capacity decay rate in the battery performance parameters and the temperature rise rate in the thermal characteristics data as the core division indexes, the safety characteristic difference coefficient of any two grid elements in the three-dimensional safety state model is calculated, which is expressed as:

[0090]

[0091] wherein, and are the capacity fade rate of two grid cells, respectively, and are the temperature rise rate of two grid cells, respectively, is the maximum value of the capacity fade rate in the whole life cycle of the battery, is the maximum value of the temperature rise rate in the whole life cycle of the battery.

[0092] Set the difference coefficient threshold (value range 0.15-0.3), when the difference coefficient of two grid cells , it is judged that the two grid cells can be divided into the same safety area.

[0093] In combination with the diaphragm integrity and the shell sealing performance in the structure state information, the area with diaphragm damage exceeding the standard or the shell sealing performance lower than the preset threshold is divided into different safety sub-areas.

[0094] After the initial division is completed by the density clustering algorithm, the region boundary is smoothed and optimized to obtain M independent safety sub-areas.

[0095] Collect historical battery extreme charging safety event data, construct a battery extreme charging safety risk prediction model based on a deep belief network, and use an improved bald eagle search algorithm to optimize the hyperparameters of the battery extreme charging safety risk prediction model. The independent safety sub-areas are taken as the input, and the extreme charging safety risk level of each area of the battery is output. The extreme charging safety risk level includes low risk, medium risk and high risk.

[0096] The historical battery extreme charging safety event data includes battery characteristic data and environmental data corresponding to the event occurrence area, including the electrical performance parameters, thermal characteristic data, structure state information of the area, and the environmental temperature, humidity and air pressure at the time of the event, as well as the occurrence time, extreme charging stage, safety accident type (such as thermal runaway, liquid leakage, bulging, etc.) and corresponding risk probability value (through the severity assessment of the accident consequences, thermal runaway is 1.0, liquid leakage is 0.7, and bulging is 0.5).

[0097] Data collection and preprocessing: 1000 groups of historical data are collected, including 200 groups of thermal runaway event data, 300 groups of liquid leakage, 200 groups of bulging, and 300 groups of normal data. The data is normalized (same as the Z-score method in the previous text), and divided into training set (700 groups), validation set (200 groups) and test set (100 groups) according to the ratio of 7:2:1. To solve the problem of data imbalance, the SMOTE algorithm is used for oversampling of the minority class (thermal runaway) to make the data amount of each class reach 300 groups.

[0098] The process of constructing the battery pole charging safety risk prediction model based on the deep belief network comprises:

[0099] Designing a deep belief network base model, comprising an input layer, a plurality of restricted Boltzmann machines hidden layers and an output layer. The input layer is used to receive independent safety sub-region feature data, and the independent safety sub-region feature data comprises electrical performance parameters, thermal characteristic data and structural state information of the independent safety sub-region. The plurality of hidden layers are used to perform feature extraction and nonlinear mapping on the data received by the input layer. The output layer is used to output a pole charging safety risk probability value of the independent safety sub-region. The activation function of the RBM layer adopts a Sigmoid function, and the connection weight between the visible layer and the hidden layer is initialized as a random value in an interval, and the bias term is initialized as 0.

[0100] Taking the region feature data corresponding to the event occurrence region in the historical battery pole charging safety event data as input and taking the risk probability value corresponding to the event as output, the deep belief network base model is pre-trained. In the pre-training stage, a greedy layer-by-layer training method is adopted, the first RBM (input layer-first hidden layer) is trained first, and then the second (first hidden layer-second hidden layer) is trained. The number of training iterations of each is 100, and the learning rate is 0.01. In the fine-tuning stage, the deep belief network is taken as a deep neural network, the output layer is connected with the second hidden layer, the Adam optimizer is adopted, the learning rate is 0.001, the batch size is 32, the number of iterations is 200, and the loss function adopts a cross-entropy loss function. In the training process, an early stopping strategy is adopted, the training is stopped when the loss of the validation set does not decrease for 10 consecutive iterations, the model parameters meeting the preset accuracy are reserved, and a battery pole charging safety risk prediction model based on the deep belief network is obtained.

[0101] The process of optimizing the hyperparameters of the battery pole charging safety risk prediction model by using the improved bald eagle search algorithm comprises:

[0102] The hyperparameter combination is composed of the learning rate of the battery pole charging safety risk prediction model, the number of hidden layer neurons and the regularization coefficient.

[0103] The bald eagle population is initialized, and the position of each bald eagle individual represents a set of hyperparameters.

[0104] The F1 score of the model evaluated by using the validation set data is used as the fitness value, and the formula is represented as:

[0105]

[0106] In the formula, P is the prediction accuracy of the model, and R is the prediction recall rate of the model.

[0107] ​The optimal position of each vulture individual and the global optimal position of the whole population are updated according to the fitness value, and the vulture individual position update formula is expressed as:

[0108]

[0109] In the formula, is the position of the mth vulture individual at the k+1th iteration, is the global optimal position at the kth iteration, is the position of the mth vulture individual at the kth iteration, and A is a control parameter, is a linearly decreasing coefficient (linearly decreasing from 2 to 0), is a random number in the interval [0, 1].

[0110] The process of initializing the population and updating the position is repeated until the preset number of iterations is reached, and the optimal hyperparameter combination is obtained.

[0111] According to the risk level of extreme charging, the safety control of each stage of the whole life cycle of the battery is classified, and the process includes:

[0112] Periodic safety inspection is implemented in low-risk areas. Real-time monitoring is adopted in medium-risk areas, and the monitoring frequency is dynamically adjusted. A full-dimensional three-dimensional monitoring system is deployed in high-risk areas, and an emergency response mechanism is linked to trigger power-off protection and forced cooling measures immediately after high-risk warning.

[0113] The inspection content of the low-risk area includes appearance inspection for bulge, leakage and simple electrical performance test. The fixed-point monitoring node is deployed in the medium-risk area, and the monitoring frequency is dynamically adjusted according to the environmental parameters. When the environmental temperature is >35℃ or the humidity is >80%, the monitoring frequency is increased to 5Hz, otherwise it is 2Hz, and the monitoring data is uploaded to the cloud platform in real time. When the data exceeds the normal range (such as voltage fluctuation ±5%), an alert message is sent. In the high-risk area, four-in-one monitoring equipment of temperature, voltage, smoke and gas (CO, H2) is deployed. When the temperature is >60℃ or the gas concentration exceeds the standard, the charging circuit is cut off, forced cooling is started, and warning information is sent to the user and the operation and maintenance personnel through SMS and APP.

[0114] The risk warning accuracy and false alarm rate of each area are calculated every month to establish an evaluation mechanism for the control effect. If the false alarm rate of the low-risk area is >10%, the inspection cycle is extended; if the missed alarm rate of the medium-risk area is >5%, the monitoring frequency is increased; if the accident rate after warning in the high-risk area is >95%, the warning threshold is optimized to avoid excessive protection.

[0115] In summary, the embodiment provides a battery pole charging full life cycle safety management method based on multi-modal fusion. By collecting battery electrical performance parameters, thermal characteristic data, structural state information and environmental parameters, comprehensive coverage of multi-modal data is achieved, which can reflect the battery state from the dimensions of electricity, heat, structure and environment, completely eliminating the safety monitoring blind area and providing complete data support for subsequent safety management. The adaptive weighted fusion algorithm and the double-factor dynamic correction model can not only efficiently integrate multi-modal data, but also accurately correct the fusion results by combining environmental parameters and pole charging stage characteristics. At the same time, the data accuracy is further improved through the deviation feedback calibration mechanism, avoiding monitoring deviation caused by environmental fluctuations and ensuring the reliability of subsequent model construction and risk prediction. By generating a three-dimensional safety state model through battery state inversion and dividing the area, a risk prediction model is constructed by combining a deep belief network and an improved bald eagle search algorithm, which can accurately identify the safety risk level of each region of the battery, realize the transition from passive monitoring to active prediction, and avoid potential safety hazards in advance. The hierarchical safety control strategy based on the risk level adopts differentiated control measures for different risk areas, which not only ensures the safety redundancy of high-risk areas, but also avoids resource waste in low-risk areas. At the same time, it covers the entire life cycle of battery pole charging, ensuring the pertinence and effectiveness of safety management at each stage, providing comprehensive and full-cycle protection for new energy battery pole charging safety, and promoting the safe application and popularization of pole charging technology in the short-distance travel field.

[0116] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be realized by means of software and the necessary general hardware platform, and of course, it can also be realized by hardware. Based on such understanding, the above technical solutions essentially or say the part that contributes to the prior art can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a magnetic disk, an optical disk, etc., including a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the method described in each embodiment or some parts of the embodiment.

[0117] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for some 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 the embodiments of the present application.

Claims

1. A battery lifecycle safety management method based on multimodal fusion, characterized in that, include: Collect multimodal data and environmental parameters throughout the entire life cycle of the battery. The multimodal data includes battery electrical performance parameters, thermal characteristic data, and structural state information. The environmental parameters include ambient temperature, humidity, and air pressure. The multimodal data is preprocessed, and an adaptive weighted fusion algorithm is used to integrate the multimodal data. The integration result is then corrected based on environmental parameters to obtain fused secure data. Based on the fused safety data, a three-dimensional safety state model of the battery's entire life cycle is generated using the battery state inversion method. Based on the consistency of battery performance parameters and the stability of structural state in the three-dimensional safety state model, the three-dimensional safety state model is divided into safety regions to obtain M independent safety sub-regions. Historical battery charging safety event data are collected, a battery charging safety risk prediction model based on deep belief network is constructed, and an improved vulture search algorithm is used to optimize the hyperparameters of the battery charging safety risk prediction model. The independent safety sub-regions are used as inputs, and the charging safety risk level of each independent safety sub-region of the battery is output. The charging safety risk level includes low risk, medium risk and high risk. Based on the aforementioned extreme charging safety risk level, graded safety management is implemented for each stage of the battery extreme charging life cycle.

2. The battery charging lifecycle safety management method based on multimodal fusion according to claim 1, characterized in that, The battery electrical performance parameters include charging voltage, charging current, internal resistance, capacity decay rate, and battery state of charge; the thermal characteristic data include battery surface temperature distribution, temperature rise rate, thermal diffusivity, and heat dissipation efficiency; the structural state information includes electrode deformation, separator integrity, electrolyte loss rate, and casing sealing performance.

3. The battery charging lifecycle safety management method based on multimodal fusion according to claim 1, characterized in that, The process of integrating the multimodal data using an adaptive weighted fusion algorithm includes: Outliers in the multimodal data are removed using the Grubbs criterion. An adaptive weighted fusion algorithm is used for data integration, and a data fusion weight allocation model is established. The fusion results are corrected through a weighted iterative optimization process.

4. The battery charging lifecycle safety management method based on multimodal fusion according to claim 1, characterized in that, The process of refining the integration results by incorporating environmental parameters to obtain fused security data includes: An environmental parameter influence factor matrix is ​​constructed to quantify the comprehensive influence of environmental temperature, humidity, and air pressure on the multimodal data fusion results. An environmental correction function based on fuzzy inference is established to map the environmental parameter influencing factors into correction coefficients; The environmental correction function is coupled with the multimodal data fusion result to obtain the corrected fused security data; The corrected fused security data is smoothed and filtered to eliminate noise caused by sudden changes in environmental parameters, and then the fused security data is output.

5. The battery charging lifecycle safety management method based on multimodal fusion according to claim 1, characterized in that, The process of generating a three-dimensional safety state model of a battery throughout its entire life cycle using battery state inversion methods includes: Using fused safety data as a constraint, a battery state inversion method based on Bayesian regularization is adopted to construct an inversion objective function. The inversion objective function is solved by an improved particle swarm optimization-simulated annealing hybrid algorithm to obtain the posterior probability distribution of the model parameters; Based on the maximum a posteriori estimation result of the posterior probability distribution, a three-dimensional mesh safety model of the battery's entire life cycle is constructed, and a three-dimensional safety state model is generated.

6. The battery charging lifecycle safety management method based on multimodal fusion according to claim 1, characterized in that, The process of dividing the three-dimensional safety status model into safety zones includes: Using the capacity decay rate in the battery electrical performance parameters and the temperature rise rate in the thermal characteristic data as the core dividing indicators, the safety characteristic difference coefficient of any two grid cells in the three-dimensional safety state model is calculated. A difference coefficient threshold is set. When the difference coefficient between two grid cells is not greater than the difference coefficient threshold, it is determined that the two grid cells can be divided into the same safe area. By combining the diaphragm integrity and shell sealing performance in the structural status information, areas where the diaphragm damage exceeds the standard or the shell sealing performance is lower than the preset threshold are divided into different safety sub-regions; After initial partitioning using density clustering algorithm, the region boundaries are smoothed and optimized to obtain M independent safe sub-regions.

7. The battery charging lifecycle safety management method based on multimodal fusion according to claim 1, characterized in that, The historical battery extreme charging safety event data includes battery characteristic data corresponding to the area where the event occurred, including the electrical performance parameters, thermal characteristic data and structural state information of the area, as well as the time of occurrence of the event, extreme charging stage, safety accident type and corresponding risk probability value.

8. The battery lifecycle safety management method based on multimodal fusion according to claim 1, characterized in that, The process of constructing a battery fast-charging safety risk prediction model based on deep belief networks includes: The deep belief network basic model is designed, including an input layer, multiple restricted Boltzmann machine hidden layers, and an output layer. The input layer is used to receive feature data of independent safe sub-regions, which includes electrical performance parameters, thermal characteristics, and structural state information of the independent safe sub-regions. The multiple hidden layers are used to extract features and perform nonlinear mapping on the data received by the input layer. The output layer is used to output the extreme charge safety risk probability value of the independent safe sub-regions. Using the regional feature data corresponding to the event occurrence area in the historical battery charging safety event data as input and the risk probability value corresponding to the event as output, the deep belief network basic model is pre-trained, and the model parameters that meet the preset accuracy are retained to obtain the battery charging safety risk prediction model based on the deep belief network.

9. The battery lifecycle safety management method based on multimodal fusion according to claim 1, characterized in that, The process of optimizing the hyperparameters of the battery extreme charging safety risk prediction model using the improved vulture search algorithm includes: The hyperparameters of the battery charging safety risk prediction model are composed of the learning rate, the number of hidden layer neurons, and the regularization coefficient. Initialize the bald eagle population, where the position of each individual bald eagle represents a set of hyperparameters; The F1 score of the model was evaluated using validation set data as the fitness value; The optimal position of each individual bald eagle and the global optimal position of the entire population are updated based on the fitness value. The process of initializing the population and updating positions is repeated iteratively until the preset number of iterations is reached to obtain the optimal combination of hyperparameters.

10. The battery lifecycle safety management method based on multimodal fusion according to claim 1, characterized in that, The process of implementing graded safety management at each stage of the battery's entire lifecycle includes: Periodic safety inspections are conducted in low-risk areas; fixed-point real-time monitoring is adopted in medium-risk areas and the monitoring frequency is dynamically adjusted; a full-dimensional three-dimensional monitoring system is deployed in high-risk areas and linked with the emergency response mechanism, and power outage protection and forced heat dissipation measures are immediately activated after a high-risk warning is triggered.

Images