A method and system for evaluating accident battery damage based on Internet of Vehicles big data

Abstract

The present invention relates to the field of damage assessment technology, specifically to an accident battery damage assessment method and system based on Internet of Vehicles big data. In the present invention, real-time and historical battery operation data are combined with statistical methods to locate and identify abnormal cells in battery accidents and output results. By analyzing battery data before and after the accident, an equivalent circuit model is used to simulate and diagnose the cell failure mode and output the results. The battery capacity attenuation coefficient and actual capacity are calculated based on the collected real-time data. The capacity attenuation coefficient after cell damage in the battery pack is determined using a linear relationship and the results are output. A convolutional neural network algorithm model is used to determine the corresponding indicators and battery damage results based on the output results, and the indicators are compared with the above three different output results. It is determined whether the battery damage is the final result based on the comparison results, and the three methods are optimized. After the optimization, they are compared again until the final battery evaluation result is determined.

Description

Technical Field

[0001] The present invention relates to the technical field of damage assessment, and in particular to an accident battery damage assessment method and system based on Internet of Vehicles big data. Background Art

[0002] Traditional battery damage assessment methods usually rely on simple threshold judgments, regular physical inspections, or monitoring of battery health status based on relatively rough models. In these methods, data collection is often not real-time, and there is a lack of analysis and comprehensive utilization of large amounts of historical data. For example, engineers may rely on regular battery inspections, such as visual inspections, capacity tests, or internal resistance measurements to assess the status of the battery. These tests are often intermittent and cannot provide continuous battery performance monitoring.

[0003] Due to data limitations, traditional methods have difficulty identifying complex or hidden failure modes, and are difficult to effectively identify slow chemical changes occurring inside the battery or minor physical damage in the initial stages. When relying on manual inspection, there is not only a risk of omissions, but also low efficiency and inability to adapt to the needs of real-time monitoring. In addition, due to the lack of support for complex algorithms and data analysis technology, traditional methods are insufficient in identifying subtle performance changes and accurately predicting battery life.

[0004] These limitations may make battery maintenance and servicing plans too conservative or delayed, making it impossible to identify impending problems and take preventive measures in a timely manner, increasing the risk of sudden failures. Moreover, due to the limitations of analytical methods, the available data resources cannot be fully utilized, and the battery management system fails to achieve the optimal maintenance strategy, which is likely to lead to premature retirement of batteries or unnecessary maintenance costs. In the field of new energy such as electric vehicles, these traditional methods are particularly inapplicable and cannot meet the rapidly developing technical standards and growing data processing needs. Summary of the invention

[0005] The purpose of the present invention is to provide a method and system for evaluating accident battery damage based on Internet of Vehicles big data to solve the problems raised in the above-mentioned background technology.

[0006] To achieve the above object, the present invention provides the following technical solution: a method for evaluating accident battery damage based on Internet of Vehicles big data, the method steps comprising:

[0007] S1. Use real-time and historical battery operation data combined with statistical methods to locate and identify abnormal cells in battery accidents. The output results include battery parameters and battery abnormal deviation index;

[0008] S2. By analyzing the battery data before and after the accident, the equivalent circuit model is used to simulate and diagnose the battery cell failure mode, identify the long-term battery performance changes, and the output results include circuit model parameters and fault indicators;

[0009] S3. Calculate the battery capacity attenuation coefficient and actual capacity based on the collected real-time data, determine the capacity attenuation coefficient after the battery cell is damaged in the battery pack using a linear relationship, and calculate the actual capacity of the damaged battery pack based on this, and the output result includes the actual capacity and the capacity attenuation coefficient;

[0010] S4, integrating the battery parameters output by S1, the circuit model parameters output by S2, and the actual capacity output by S3 as the input feature vector, and training the convolutional neural network algorithm model, using the trained convolutional neural network algorithm model to predict the battery abnormal deviation index, fault index, capacity attenuation coefficient and battery damage result according to the input feature vector, and comparing the predicted results with the battery parameters output by S1, the circuit model parameters output by S2, and the actual capacity output results output by S3, respectively. If the comparison results of the three are all consistent, the battery damage result predicted by the convolutional neural network algorithm is determined to be the final result, otherwise the inconsistent method is optimized;

[0011] S5. When the battery abnormal deviation index predicted by the neural network is inconsistent with the battery abnormal deviation index in S1, high-frequency data transmission is performed by shortening the time interval for collecting data;

[0012] When the fault indicators predicted by the neural network are inconsistent with the fault indicators in S2, the data quality is improved by using data cleaning technology;

[0013] When the capacity attenuation coefficient predicted by the neural network is inconsistent with the capacity attenuation coefficient in S3, the output results are normalized to keep the result weight balanced;

[0014] After the optimization process, the comparison is performed again according to the previous comparison logic until the final battery evaluation result is determined.

[0015] As a further improvement of the present technical solution, the real-time and historical battery operation data are combined with statistical methods to locate and identify abnormal cells in battery accidents. The output results include battery parameters and battery abnormal deviation index. By collecting fine-grained real-time battery operation data before and after the accident, the battery type and power system are identified, and voltage deviation analysis and outlier detection methods are used to set thresholds and quantify the degree of battery abnormality. These indicators are used to grade the battery status and generate a data set with a timestamp to record the operating parameters of each battery cell and its abnormal deviation index, so as to achieve rapid identification of abnormal cells in accident vehicles.

[0016] As a further improvement of the technical solution, the battery data before and after the accident is analyzed, and the equivalent circuit model is used to simulate and diagnose the battery cell failure mode, identify the long-term battery performance changes, and the output results include circuit model parameters and fault indicators. The failure mode is identified by viewing historical data, and the battery equivalent circuit model is used to simulate the behavior of the battery under normal and abnormal conditions, diagnose battery failures and identify damage caused by accidents and other long-term performance changes, including using outlier detection technology to analyze cells with atypical behavior, compare performance parameter changes before and after the accident, predict and analyze possible structural damage, and create a fault feature matrix, which includes circuit model parameters and fault indicators of different battery cells.

[0017] As a further improvement of the technical solution, the training of the convolutional neural network algorithm model specifically includes:

[0018] The model consists of an input layer, a hidden layer, a branch layer, and an output layer. The input layer takes the features as input and connects them to the hidden layer.

[0019] There are multiple branch layers, one of which is used to predict a specific branch layer of battery abnormal deviation index, one is used to predict a specific branch layer of fault indicator task, one is used to predict a specific branch layer of capacity attenuation coefficient task, and one is used to predict a specific branch layer of battery damage result task;

[0020] The branch layer receives the output of the hidden layer as input and processes it through its own fully connected neurons and activation functions;

[0021] The neural network passes input data from the input layer to the output layer, where the output is calculated by the activation function and the weights between the layers;

[0022] After the forward propagation, the predictions obtained by the neural network are compared with the corresponding label values, and a loss function is calculated, where the loss function measures the difference between the predictions and the actual labels.

[0023] The error is propagated back to the network using the loss function, and the contribution of each parameter to the loss is calculated. The gradient is calculated backwards from the output layer to the input layer using the chain rule, and the value of each parameter is updated according to the direction of the gradient to minimize the loss function. The parameters in the neural network are updated based on the calculated gradient information.

[0024] As a further improvement of the technical solution, the convolutional neural network algorithm model is used to predict the battery abnormal deviation index, fault index, capacity attenuation coefficient and battery damage result according to the input feature vector, specifically including:

[0025] The trained neural network model is loaded in, which includes the number of network layers, the number of neurons in each layer, the selection of activation functions, and the connection weights and biases between layers for network calculations;

[0026] The input data is fed into the network, starting from the input layer, and is calculated by each neuron layer by layer, passing information along the network layers until it reaches the last layer, which is the output layer, and obtains the prediction result;

[0027] In each neuron, nonlinear features are introduced and feature transformation is performed by applying an activation function to the weighted sum of the intermediate layer outputs, where the activation function nonlinearly maps the weighted sum of the intermediate layer outputs.

[0028] As a further improvement of the present technical solution, when the battery abnormal deviation index predicted by the neural network is inconsistent with the battery abnormal deviation index in S1, high-frequency data transmission is performed by shortening the time interval for collecting data, reducing data collection from once every 30 seconds to once every 10 seconds, so that system resources and bandwidth can handle higher-frequency data transmission.

[0029] As a further improvement of the present technical solution, when the fault indicator predicted by the neural network is inconsistent with the fault indicator in S2, the data quality is improved by using data cleaning technology, missing value processing: check the entire data set, determine the location of the missing data value, and directly delete the row containing the missing value; outlier value processing: use statistical methods to detect outliers in the data set, determine the location of the outliers, and delete the row containing the outliers: duplicate value processing: check whether there are duplicate values ​​in the data, determine the location of the duplicate values, and delete the row containing the duplicate values.

[0030] As a further improvement of the present technical solution, when the capacity attenuation coefficient predicted by the neural network is inconsistent with the capacity attenuation coefficient in S3, the output result is normalized, the mean and standard deviation of the feature column in the output result data are calculated, the original value is subtracted from the mean and then divided by the standard deviation, and the data is converted into a distribution with a mean and a standard deviation, so that the result weight remains balanced.

[0031] The second object of the present invention is to provide a system for an accident battery damage assessment method based on Internet of Vehicles big data, which includes an abnormal battery cell detection and positioning module, a battery cell failure mode diagnosis module, a battery capacity attenuation assessment module, a convolutional neural network prediction and comparison module, and a module assessment result optimization processing module, wherein:

[0032] The abnormal cell detection and positioning module uses real-time and historical battery operation data combined with statistical methods to locate and identify abnormal cells in battery accidents, and the output results include battery parameters and battery abnormal deviation index;

[0033] The battery cell failure mode diagnosis module analyzes battery data before and after the accident, uses an equivalent circuit model to simulate and diagnose the battery cell failure mode, identifies long-term battery performance changes, and outputs results including circuit model parameters and failure indicators;

[0034] The battery capacity attenuation evaluation module calculates the battery capacity attenuation coefficient and the actual capacity according to the collected real-time data, determines the capacity attenuation coefficient after the battery cell is damaged in the battery pack by using a linear relationship, and calculates the actual capacity of the damaged battery pack accordingly, and the output result includes the actual capacity and the capacity attenuation coefficient;

[0035] The convolutional neural network prediction and comparison module integrates the battery parameters output by the abnormal battery cell detection and positioning module, the circuit model parameters output by the battery cell failure mode diagnosis module, and the actual capacity output by the battery capacity attenuation evaluation module as input feature vectors, and trains the convolutional neural network algorithm model. The trained convolutional neural network algorithm model is used to predict the battery abnormal deviation index, fault index, capacity attenuation coefficient and battery damage result according to the input feature vector, and the prediction results are compared with the battery parameters output by the abnormal battery cell detection and positioning module, the circuit model parameters output by the battery cell failure mode diagnosis module and the actual capacity output result output by the battery capacity attenuation evaluation module. If the comparison results of the three are all consistent, the battery damage result predicted by the convolutional neural network algorithm is determined to be the final result, otherwise the inconsistent comparison method is optimized;

[0036] When the battery abnormal deviation index predicted by the neural network of the module evaluation result optimization processing module is inconsistent with the battery abnormal deviation index in the abnormal battery cell detection and positioning module, high-frequency data transmission is performed by shortening the time interval for collecting data;

[0037] When the fault indicators predicted by the neural network are inconsistent with the fault indicators in the battery cell fault mode diagnosis module, the data quality is improved by using data cleaning technology;

[0038] When the capacity attenuation coefficient predicted by the neural network is inconsistent with the capacity attenuation coefficient in the battery capacity attenuation evaluation module, the output results are normalized to keep the result weight balanced;

[0039] After the optimization process, the comparison is performed again according to the previous comparison logic until the final battery evaluation result is determined.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] 1. This accident battery damage assessment method and system based on Internet of Vehicles big data uses real-time and historical battery operation data combined with statistical methods to more comprehensively understand the operating status of the battery, comprehensively analyzes real-time and historical data, and outputs battery parameters and battery abnormal deviation index; uses an equivalent circuit model to simulate and diagnose the battery cell failure mode, accurately judges the long-term performance changes and failure modes of the battery, and outputs circuit model parameters and failure indicators; calculates the battery capacity attenuation coefficient and actual capacity based on the collected real-time data; finally, uses a convolutional neural network algorithm model to perform battery abnormal deviation index, failure indicator, capacity attenuation coefficient and battery damage result on the output results of the three, compares the predicted results with the three respectively, and determines the final battery damage result based on the comparison results, thereby improving the accuracy and reliability of the assessment.

[0042] 2. This accident battery damage assessment method and system based on Internet of Vehicles big data optimizes the three to varying degrees when there is inconsistency in the comparison, including reducing the time interval for collecting data for high-frequency data transmission, improving data quality by using data cleaning technology, and normalizing the output results to keep the result weights balanced. After the optimization, the comparison is performed again according to the previous comparison logic until the final battery assessment result is determined. This dynamic optimization process can gradually improve the accuracy and reliability of the assessment through continuous feedback and adjustment. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 It is a schematic diagram of the method steps of the present invention;

[0044] Figure 2 It is a schematic diagram of the overall module of the present invention.

[0045] In the figure: 100, abnormal battery cell detection and positioning module; 200, battery cell failure mode diagnosis module; 300, battery capacity attenuation evaluation module; 400, convolutional neural network prediction and comparison module; 500, module evaluation result optimization processing module. DETAILED DESCRIPTION

[0046] The following will be combined with the accompanying drawings in the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0047] See also Figure 1-2 The present invention provides a technical solution: a method for evaluating accident battery damage based on Internet of Vehicles big data, comprising the following method steps:

[0048] S1. Use high-frequency real-time and historical battery operation data combined with statistical methods to locate and identify abnormal cells in battery accidents. The output results include battery parameters and battery abnormal deviation index;

[0049] S2. By analyzing the battery data before and after the accident, the equivalent circuit model is used to simulate and diagnose the battery cell failure mode, identify the long-term battery performance changes, and the output results include circuit model parameters and fault indicators;

[0050] S3. Provide a calculation method for determining the capacity attenuation coefficient after the battery cell is damaged in the battery pack, and calculate the actual capacity of the damaged battery pack based on the capacity attenuation coefficient, the output structure includes the actual capacity and the capacity attenuation coefficient;

[0051] S4, integrating the battery parameters output by S1, the circuit model parameters output by S2, and the actual capacity output by S3 as the input feature vector, and training the convolutional neural network algorithm model, using the trained convolutional neural network algorithm model to predict the battery abnormal deviation index, fault index, capacity attenuation coefficient and battery damage result according to the input feature vector, and comparing the predicted results with the battery parameters output by S1, the circuit model parameters output by S2, and the actual capacity output results output by S3, respectively. If the comparison results of the three are all consistent, the battery damage result predicted by the convolutional neural network algorithm is determined to be the final result, otherwise the inconsistent method is optimized;

[0052] S5. When the battery abnormal deviation index predicted by the neural network is inconsistent with the battery abnormal deviation index in S1, high-frequency data transmission is performed by shortening the time interval for collecting data;

[0053] When the fault indicators predicted by the neural network are inconsistent with the fault indicators in S2, the data quality is improved by using data cleaning technology;

[0054] When the capacity attenuation coefficient predicted by the neural network is inconsistent with the capacity attenuation coefficient in S3, the output results are normalized to keep the result weight balanced;

[0055] After the optimization process, the comparison is performed again according to the previous comparison logic until the final battery evaluation result is determined.

[0056] The details are as follows:

[0057] S1. Identify abnormal cells during the accident period, including analysis of real-time and historical data, and quickly locate cells with possible problems through statistical methods such as voltage deviation, including:

[0058] Collect raw battery information data from the public database during the accident period. Specifically, obtain more than 1,000 frames of battery operation data from 3 to 5 minutes before and after the accident. The data collection time interval is within 30 seconds to ensure that the time granularity is fine enough to provide an accurate snapshot of the battery operation status.

[0059] Identify the battery type (such as lithium-ion battery, nickel-metal hydride battery, etc.) and the vehicle power type (such as electric vehicle or plug-in hybrid) from static data; combine dynamic data, including single cell voltage, single cell number, temperature, current and resistance, etc.; introduce the concept of voltage deviation, that is, the deviation of each battery cell from the average voltage value; use the voltage outlier statistical method to set a dynamic threshold for each battery cell and screen out the specific location of the faulty battery cell;

[0060] In the selected short time window, the number of voltage outliers of each battery cell is counted; a battery abnormal deviation index characterization algorithm is established to quantify the degree of battery abnormality. The deviation index is based on factors such as the frequency of voltage outliers, voltage fluctuation range and time continuity; the abnormal state of the battery is evaluated based on the deviation index and dynamic data, and the degree of abnormality is graded, such as mild, moderate and severe, to achieve rapid identification of abnormal cells in accident vehicles;

[0061] Generate a data set containing timestamps, where each record is a battery parameter (voltage, temperature, etc.) at a time point and the corresponding battery abnormal deviation index.

[0062] S2. Conduct in-depth analysis of these identified abnormal cells, including studying historical data to determine failure modes, using battery equivalent circuit models to simulate battery behavior under normal and abnormal conditions, and ultimately diagnosing damage caused by accidents and other long-term battery performance changes, including:

[0063] Extract historical operating data before the accident, including battery voltage, temperature, current, internal resistance, and charge and discharge status;

[0064] Apply outlier detection techniques to identify cells that have historically behaved atypically (e.g., cells whose voltages often fluctuate abnormally) in historical data;

[0065] Use the battery equivalent circuit model to analyze the contribution of each component (such as battery cells, connectors, battery management system, etc.) to the external voltage under normal operating conditions of the battery. According to the model, simulate the external voltage response characteristics of the battery under normal conditions, including the performance under different states of charge and load conditions;

[0066] Compare historical and real-time battery performance parameters before and after the accident, identify cells that meet the fault characteristics of accelerated aging, increased internal resistance, reduced capacity, and analyze the data of SOC deviation (whether too high or too low), because SOC deviation is caused by inaccurate readings of the battery management system, battery imbalance, or degradation of certain single battery functions;

[0067] According to the fault characteristics corresponding to the change of external voltage (such as voltage reduction corresponding to capacity reduction or internal resistance increase), the actual monitored battery performance is compared with the model prediction to determine the fault type;

[0068] Use the battery equivalent circuit model to predict the external voltage characterization differences of faulty batteries and reveal the internal mechanism of battery performance degradation or damage;

[0069] Combined with abnormal battery data identified after the accident, consider the physical damage caused by the collision to the battery; analyze the possible damage caused by collision mechanics to the battery cell, such as battery structure deformation, connector breakage or internal short circuit of the battery cell, and its impact on battery performance; if physical inspection confirms structural damage caused by the collision, combine this information with the dynamic and static data of the battery to further verify the battery performance degradation caused by external stress interference (such as collision);

[0070] Output a fault signature matrix, where rows represent different cells and columns contain various circuit model parameters and fault indicators.

[0071] S3. Calculate the battery capacity attenuation coefficient and actual capacity based on the collected real-time data, including:

[0072] The number of charge and discharge cycles when the battery capacity decays to a certain percentage of the rated capacity at a certain charge and discharge rate. Referring to the actual battery life test data in the database, the battery's charge and discharge capacity each time is a constant linear function of the rated capacity. After n cycles, the battery capacity is expressed as Where C n is the actual capacity of the battery; N is the battery life; P is the percentage of capacity decay after the battery reaches its service life; C0 is the initial rated capacity of the battery, where the capacity decay is considered to be proportional to the number of charge and discharge cycles according to a linear model, and each cycle causes a fixed proportion of capacity loss. For example, if the rated capacity of a battery is C0 (for example, 100%), when it undergoes N cycles (for example, 1000 times), the expected capacity decay is P (for example, 30%). This formula predicts that after each cycle, the battery capacity decays by P / N (that is, 0.3%). After n cycles, the expected capacity of the battery is C n The actual capacity C will be obtained by multiplying the initial capacity C0 by 1-nP / N, which represents the remaining capacity percentage. nIn short, this formula predicts the actual capacity of a battery at any given point in time using the battery's initial rated capacity, the expected total capacity decay percentage, and the expected number of cycles over the battery's life;

[0073] Battery life is quantified in terms of the number of charge and discharge cycles rather than in actual time, because the degradation of batteries is related to the number of charge and discharge cycles they undergo. Therefore, the service life here refers to the number of charge and discharge cycles that the battery undergoes until its capacity drops to a certain percentage of its initial capacity.

[0074] Since inconsistency is bound to exist between each battery cell, some cells in the battery pack have a relatively small capacity and large internal resistance. Under normal charge and discharge conditions, the discharge depth is relatively large, and it is easy to overcharge and over-discharge, which damages some cells and affects the actual capacity of the battery pack. The capacity attenuation coefficient of the battery pack is defined, that is, the capacity of the damaged battery when used for the i-th time is expressed as C i =f i-1 (ΔC i-1 )C i-1 , where: C i is the capacity value of the battery when it is used for the i-th cycle; ΔC i-1 is the capacity difference between the i-1th use and the i-2th use; f i-1 (ΔC i-1 ) is the battery damage coefficient when used for the i-th time, which is ΔC i-1 function; under the influence of inconsistency, the ideal battery capacity of the battery is the attenuation coefficient of the change multiplied by the calibrated capacity, and the actual capacity expression is: Among them, C n Refers to the actual capacity of the battery after the nth charge and discharge cycle, f n (ΔC) is a function that characterizes the degree of influence of the capacity change ΔC from the previous cycle to the current cycle on the battery capacity, that is, the battery damage coefficient. This damage coefficient depends on the capacity change ΔC and other factors and may be a complex function. ΔC represents the capacity change, that is, the difference between the capacity of the previous cycle and the capacity of the current cycle. (1-np / N) describes the theoretical linear capacity attenuation trend of the battery as the number of cycles n increases. p is the percentage of capacity attenuation after reaching the service life, and N is the expected number of cycles of the battery's service life. C0 refers to the rated capacity of the battery, that is, the initial capacity of the battery measured during production;

[0075] S4, integrating the input feature vector of the convolutional neural network algorithm model, including the battery parameters output by S1, the circuit model parameters output by S2 and the actual capacity output by S3. The convolutional neural network algorithm model predicts the battery abnormal deviation index, fault index, capacity attenuation coefficient and battery damage result according to the input feature vector, including:

[0076] Data collection: Use public databases to collect historical data, including data features of input feature vectors and labels of battery abnormal deviation index, fault indicators, capacity attenuation coefficient, and battery damage results, and use them as training sets for the model;

[0077] Model training: The model consists of an input layer, a hidden layer, a branch layer, and an output layer. The input layer takes the features as input and connects them to the hidden layer. There are multiple branch layers, one of which is used to predict a specific branch layer of the battery abnormal deviation index, one branch layer is used to predict a specific branch layer of the fault indicator task, one branch layer is used to predict a specific branch layer of the capacity attenuation coefficient task, and one branch layer is used to predict a specific branch layer of the battery damage result task. The branch layer receives the output of the hidden layer as input and processes it through its own fully connected neurons and activation functions. The neural network passes the input data from the input layer to the output layer, where the output is calculated by the activation function and the weights between the layers. After forward propagation, the prediction results obtained by the neural network are compared with the corresponding label values ​​to calculate the value of a loss function, where the loss function measures the difference between the prediction results and the actual labels. The error is passed back to the network using the loss function to calculate the contribution of each parameter to the loss. The gradient is calculated from the output layer to the input layer by the chain rule, and the value of each parameter is updated according to the direction of the gradient to minimize the loss function. Finally, the parameters in the neural network are updated according to the calculated gradient information.

[0078] Result prediction: First, the trained neural network model is loaded. The model includes the number of network layers, the number of neurons in each layer, the choice of activation function, and the connection weights and biases between layers for network calculations. Next is the forward propagation calculation. The input data is input into the network, starting from the input layer, and is calculated through each neuron layer by layer, and information is transmitted along the network level until it reaches the last layer, that is, the output layer, to obtain the prediction result. In each neuron, the activation function is applied to the weighted sum of the output of the intermediate layer to introduce nonlinear features and perform feature transformation. The activation function nonlinearly maps the weighted sum of the output of the intermediate layer, thereby introducing the nonlinear nature of the network and introducing nonlinear expression capabilities to the network, enabling it to better model and predict complex data patterns.

[0079] The battery abnormal deviation index, fault index and capacity attenuation coefficient results predicted by the neural network are compared with the battery abnormal deviation index in S1, the fault index in S2 and the capacity attenuation coefficient in S3 respectively. If the three are relatively consistent, the battery evaluation result predicted by the neural network is determined as the final result. Otherwise, the methods in S1, S2 and S3 are optimized. The range of consistency is 5%, and the formula is: |ab|<=0.05*(|a|+|b|), where a and b are the values ​​of the comparison objects.

[0080] S5. The optimization process specifically includes:

[0081] When the battery abnormal deviation index predicted by the neural network is inconsistent with the battery abnormal deviation index in S1, the time interval for collecting data is shortened so that system resources and bandwidth can handle higher-frequency data transmission, reducing data collection from once every 30 seconds to once every 10 seconds;

[0082] When the fault indicators predicted by the neural network are inconsistent with the fault indicators in S2, the data quality is improved by using data cleaning technology. Missing value processing: check the entire data set, determine the location of the missing data value, and directly delete the row containing the missing value; outlier processing: use statistical methods to detect outliers in the data set, determine the location of the outliers, and delete the row containing the outliers: duplicate value processing: check whether there are duplicate values ​​in the data, determine the location of the duplicate values, and delete the row containing the duplicate values;

[0083] When the capacity attenuation coefficient predicted by the neural network is inconsistent with the capacity attenuation coefficient in S3, the output results are normalized, the mean and standard deviation of the feature columns in the output result data are calculated, the original value is subtracted from the mean and then divided by the standard deviation, and the data is converted into a distribution with a mean of 0 and a standard deviation of 1 to keep the result weight balanced for optimization.

[0084] After the optimization process, the comparison is performed again according to the previous comparison logic until the final battery evaluation result is determined.

[0085] The second purpose of this embodiment is to provide a system for an accident battery damage assessment method based on Internet of Vehicles big data, including an abnormal battery cell detection and positioning module 100, a battery cell failure mode diagnosis module 200, a battery capacity attenuation assessment module 300, a convolutional neural network prediction and comparison module 400 and a module assessment result optimization processing module 500, as follows:

[0086] The abnormal cell detection and positioning module 100 uses real-time and historical battery operation data combined with statistical methods to locate and identify abnormal cells in battery accidents, and the output results include battery parameters and battery abnormal deviation index;

[0087] The battery cell failure mode diagnosis module 200 analyzes battery data before and after the accident, uses an equivalent circuit model to simulate and diagnose the battery cell failure mode, identifies long-term battery performance changes, and outputs results including circuit model parameters and failure indicators;

[0088] The battery capacity attenuation evaluation module 300 calculates the battery capacity attenuation coefficient and the actual capacity according to the collected real-time data, determines the capacity attenuation coefficient after the battery cell is damaged in the battery pack by using a linear relationship, and calculates the actual capacity of the damaged battery pack accordingly, and the output result includes the actual capacity and the capacity attenuation coefficient;

[0089] The convolutional neural network prediction and comparison module 400 integrates the battery parameters output by the abnormal battery cell detection and positioning module 100, the circuit model parameters output by the battery cell failure mode diagnosis module 200, and the actual capacity output by the battery capacity attenuation evaluation module 300 as input feature vectors, and trains the convolutional neural network algorithm model, and uses the trained convolutional neural network algorithm model to predict the battery abnormal deviation index, fault index, capacity attenuation coefficient and battery damage result according to the input feature vector, and compares the predicted results with the battery parameters output by the abnormal battery cell detection and positioning module 100, the circuit model parameters output by the battery cell failure mode diagnosis module 200, and the actual capacity output result output by the battery capacity attenuation evaluation module 300, respectively. If the comparison results of the three are all consistent, the battery damage result predicted by the convolutional neural network algorithm is determined to be the final result, otherwise the inconsistent comparison method is optimized;

[0090] When the battery abnormal deviation index predicted by the neural network of the module evaluation result optimization processing module 500 is inconsistent with the battery abnormal deviation index in the abnormal battery cell detection and positioning module 100, high-frequency data transmission is performed by shortening the time interval for collecting data;

[0091] When the fault index predicted by the neural network is inconsistent with the fault index in the battery cell fault mode diagnosis module 200, the data quality is improved by using data cleaning technology;

[0092] When the capacity attenuation coefficient predicted by the neural network is inconsistent with the capacity attenuation coefficient in the battery capacity attenuation evaluation module 300, the output result is normalized to keep the result weight balanced;

[0093] After the optimization process, the comparison is performed again according to the previous comparison logic until the final battery evaluation result is determined.

[0094] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and descriptions are only preferred examples of the present invention and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A method for assessing accident battery damage based on Internet of Vehicles big data, characterized in that: The steps are as follows: S1. Use real-time and historical battery operation data combined with statistical methods to locate and identify abnormal cells in battery accidents. The output results include battery parameters and battery abnormal deviation index; specifically include: By collecting fine-grained real-time battery operation data before and after the accident, identifying the battery type and power system, using voltage deviation analysis and outlier detection methods to set thresholds and quantify the degree of battery abnormality, using the voltage outlier number index and voltage fluctuation range index generated during the voltage deviation analysis process to classify the battery status, and generating a data set with a timestamp to record the operating parameters of each battery cell and its abnormal deviation index, to achieve rapid identification of abnormal cells in the accident vehicle; S2. By analyzing the battery data before and after the accident, the equivalent circuit model is used to simulate and diagnose the battery cell failure mode, identify the long-term battery performance changes, and the output results include circuit model parameters and fault indicators; S3. Calculate the battery capacity attenuation coefficient and actual capacity based on the collected real-time data, determine the capacity attenuation coefficient after the battery cell is damaged in the battery pack using a linear relationship, and calculate the actual capacity of the damaged battery pack based on this, and the output result includes the actual capacity and the capacity attenuation coefficient after damage; S4, integrating the battery parameters output by S1, the circuit model parameters output by S2, and the actual capacity output by S3 as the input feature vector, and training the convolutional neural network algorithm model, using the trained convolutional neural network algorithm model to predict the battery abnormal deviation index, fault index and capacity attenuation coefficient according to the input feature vector, and comparing the predicted results with the battery abnormal deviation index output by S1, the fault index output by S2, and the capacity attenuation coefficient output by S3, and comparing them according to the comparison consistency error formula, where the formula is |ab|<=0.05*(|a|+|b|), a and b are the values ​​of the comparison objects, and the comparison results of the three are all consistent, and the battery damage result predicted by the convolutional neural network algorithm is determined as the final result, otherwise the inconsistent comparison method is optimized; S5. When the battery abnormal deviation index predicted by the neural network is inconsistent with the battery abnormal deviation index in S1, high-frequency data transmission is performed by shortening the time interval for collecting data; When the fault indicators predicted by the neural network are inconsistent with the fault indicators in S2, the data quality is improved by using data cleaning technology; When the capacity decay coefficient predicted by the neural network is inconsistent with the capacity decay coefficient in S3, the output results are normalized to keep the result weight balanced; After the optimization process, the comparison is performed again according to the previous comparison logic until the final battery evaluation result is determined.

2. The accident battery damage assessment method based on Internet of Vehicles big data according to claim 1 is characterized by: The process of training the convolutional neural network algorithm model specifically includes: The model consists of an input layer, a hidden layer, a branch layer, and an output layer. The input layer takes the features as input and connects them to the hidden layer. There are multiple branch layers, one of which is used to predict a specific branch layer of battery abnormal deviation index, one is used to predict a specific branch layer of fault indicator task, one is used to predict a specific branch layer of capacity attenuation coefficient task, and one is used to predict a specific branch layer of battery damage result task; The branch layer receives the output of the hidden layer as input and processes it through its own fully connected neurons and activation functions; The neural network passes input data from the input layer to the output layer, where the output is calculated by the activation function and the weights between the layers; After the forward propagation, the predictions obtained by the neural network are compared with the corresponding label values, and a loss function is calculated, where the loss function measures the difference between the predictions and the actual labels. The error is propagated back to the network using the loss function, and the contribution of each parameter to the loss is calculated. The gradient is calculated backwards from the output layer to the input layer using the chain rule, and the value of each parameter is updated according to the direction of the gradient to minimize the loss function. The parameters in the neural network are updated based on the calculated gradient information.

3. The accident battery damage assessment method based on Internet of Vehicles big data according to claim 1 is characterized by: The trained convolutional neural network algorithm model predicts the battery abnormal deviation index, fault index and capacity attenuation coefficient according to the input feature vector, specifically including: The trained neural network model is loaded in, which includes the number of network layers, the number of neurons in each layer, the selection of activation functions, and the connection weights and biases between layers for network calculations; The input data is fed into the network, starting from the input layer, and is calculated by each neuron layer by layer, passing information along the network layers until it reaches the last layer, which is the output layer, and obtains the prediction result; In each neuron, nonlinear features are introduced and feature transformation is performed by applying an activation function to the weighted sum of the intermediate layer outputs, where the activation function nonlinearly maps the weighted sum of the intermediate layer outputs.

4. The accident battery damage assessment method based on Internet of Vehicles big data according to claim 1 is characterized by: When the battery abnormal deviation index predicted by the neural network is inconsistent with the battery abnormal deviation index in S1, high-frequency data transmission is performed by shortening the time interval for collecting data, reducing data collection from once every 30 seconds to once every 10 seconds, so that system resources and bandwidth can handle higher-frequency data transmission.

5. The accident battery damage assessment method based on Internet of Vehicles big data according to claim 1 is characterized by: When the fault indicators predicted by the neural network are inconsistent with the fault indicators in S2, the data quality is improved by using data cleaning technology, missing value processing: check the entire data set, determine the location of the missing data value, and directly delete the row containing the missing value; outlier value processing: use statistical methods to detect outliers in the data set, determine the location of the outliers, and delete the row containing the outliers; duplicate value processing: check whether there are duplicate values ​​in the data, determine the location of the duplicate values, and delete the row containing the duplicate values.

6. The accident battery damage assessment method based on Internet of Vehicles big data according to claim 1 is characterized by: When the capacity attenuation coefficient predicted by the neural network is inconsistent with the capacity attenuation coefficient in S3, the output results are normalized, the mean and standard deviation of the feature columns in the output result data are calculated, the original value is subtracted from the mean and then divided by the standard deviation, and the data is converted into a distribution with a mean of 0 and a standard deviation of 1 to keep the result weight balanced.

7. A system using the accident battery damage assessment method based on Internet of Vehicles big data as described in any one of claims 1 to 6, characterized in that: The system comprises an abnormal battery cell detection and positioning module (100), a battery cell failure mode diagnosis module (200), a battery capacity attenuation evaluation module (300), a convolutional neural network prediction and comparison module (400) and a module evaluation result optimization processing module (500), wherein: The abnormal battery cell detection and positioning module (100) uses real-time and historical battery operation data combined with statistical methods to locate and identify abnormal batteries in battery accidents, and outputs results including battery parameters and battery abnormal deviation index; The battery cell failure mode diagnosis module (200) analyzes battery data before and after the accident, uses an equivalent circuit model to simulate and diagnose the battery cell failure mode, identifies long-term battery performance changes, and outputs results including circuit model parameters and failure indicators; The battery capacity attenuation evaluation module (300) calculates the battery capacity attenuation coefficient and the actual capacity according to the collected real-time data, determines the capacity attenuation coefficient after the battery cell in the battery pack is damaged by using a linear relationship, and calculates the actual capacity of the damaged battery pack accordingly, and the output result includes the actual capacity and the capacity attenuation coefficient; The convolution neural network prediction and comparison module (400) integrates the battery parameters output by the abnormal battery cell detection and positioning module (100), the circuit model parameters output by the battery cell fault mode diagnosis module (200), and the actual capacity output by the battery capacity attenuation evaluation module (300) as input feature vectors, and trains the convolution neural network algorithm model. The trained convolution neural network algorithm model is used to predict the battery abnormal deviation index, the fault index, and the capacity attenuation coefficient according to the input feature vectors. The prediction results are compared with the battery abnormal deviation index and the battery abnormal deviation index output by the abnormal battery cell detection and positioning module (100), the fault index output by the battery cell fault mode diagnosis module (200), and the capacity attenuation coefficient output by the battery capacity attenuation evaluation module (300), and the comparison is performed according to a comparison consistency error formula, wherein the formula is |ab|<=0.05*(|a|+|b|), a and b are the values ​​of the comparison objects, and if the comparison results of the three are all consistent, the battery damage result predicted by the convolution neural network algorithm is determined to be the final result, otherwise the comparison inconsistent method is optimized; When the battery abnormality deviation index predicted by the neural network of the module evaluation result optimization processing module (500) is inconsistent with the battery abnormality deviation index in the abnormal battery cell detection and positioning module (100), high-frequency data transmission is performed by shortening the time interval for collecting data; When the fault index predicted by the neural network is inconsistent with the fault index in the battery cell fault mode diagnosis module (200), improving the data quality by using data cleaning technology; When the capacity attenuation coefficient predicted by the neural network is inconsistent with the capacity attenuation coefficient in the battery capacity attenuation evaluation module (300), the output result is normalized to keep the result weight balanced; After the optimization process, the comparison is performed again according to the previous comparison logic until the final battery evaluation result is determined.

Images