Method for identifying fault mode of battery pack

Abstract

The invention discloses a method for identifying a fault mode of a battery pack, which comprises the following steps of: 1) defining n fault modes according to fault characteristics, and acquiring a fault data set under each fault mode; 2) screening the fault data set by using the box diagram to obtain a fault optimal data set; 3) building a battery pack fault mode diagnosis model by using a self-organizing mapping-feedforward neural network algorithm; 4) training the battery pack fault mode diagnosis model by using the fault optimization data set to obtain a trained battery pack fault mode recognition model; 5) monitoring a vehicle operation signal, and transmitting the vehicle operation signal to the battery pack fault mode recognition model to obtain a battery pack fault recognition result; the problem that the fault mode cannot be recognized and early warning cannot be provided when the battery pack system breaks down in the long-term use process is solved.

Description

Technical Field

[0001] The present invention relates to the field of battery pack fault diagnosis, and in particular to a method for battery pack fault mode identification. Background Art

[0002] With the development of world industry, the earth's oil resources are gradually decreasing. For strategic development, new energy vehicles have been listed as strategic emerging industries by China. In order to support the good development of new energy vehicles, the country has introduced relevant policies.

[0003] Representative new energy vehicles are pure electric, hybrid and fuel cell vehicles. These three new energy vehicles use the electricity provided by batteries as the main or auxiliary power source, which can reduce environmental pollution to varying degrees. Due to its superiority, pure electric vehicles have been developed by various countries with great efforts.

[0004] With the rapid growth of production and sales of new energy vehicles in my country, the quality problems of new energy vehicles have been increasingly prominent, and safety accidents such as fire and power loss control have occurred frequently. The safety and quality level of new energy vehicles urgently needs to be improved. According to statistics, in recent years, the annual average number of new energy vehicle fire accidents in China has exceeded 2,000, of which more than 90% occurred in the operation link, and the main cause of the accident was thermal runaway of the power battery. At the same time, the problem of reduced output power due to abnormal working state of the power battery, resulting in limited power operation or power interruption of the vehicle, also occurs from time to time, causing loss of life and property to users, which urgently needs to be solved. Based on big data analysis methods, by extracting and analyzing the operating characteristic parameters of the power battery, monitoring the safety status of the vehicle battery in real time, and carrying out fault diagnosis and safety warning, it can effectively identify potential faults of the in-service battery, thereby avoiding accidents. At present, scientific research institutions and related enterprises are all researching safety technologies, but the early warning capabilities generally have deficiencies such as "late warning, low accuracy, and less identification", and there are problems such as risk omission and false alarm. There is an urgent need to develop high-precision safety early warning technology. Summary of the invention

[0005] The object of the present invention is to provide a method for identifying a battery pack failure mode, comprising the following steps:

[0006] 1) Based on the fault characteristics, define n fault modes and obtain the fault data set under each fault mode;

[0007] 2) Use the box plot to filter the fault data set and obtain the fault optimal data set;

[0008] 3) Use the self-organizing map-feedforward neural network algorithm to build a battery pack failure mode diagnosis model;

[0009] 4) Using the fault optimization data set to train the battery pack fault mode diagnosis model to obtain a trained battery pack fault mode recognition model;

[0010] 5) Monitor the vehicle operation signal and transmit it to the battery pack fault mode recognition model to obtain the battery pack fault recognition result; the battery pack fault recognition result includes no fault and fault; when the battery pack fault recognition result is faulty, it also includes a warning signal and a fault mode.

[0011] Furthermore, the failure modes include but are not limited to abnormal voltage sampling, abnormal mileage sampling, abnormal temperature sampling, abnormal temperature sensor connection, insulation failure, sudden internal short circuit, inconsistent discharge of different battery cells, and abnormal self-discharge.

[0012] Furthermore, the voltage sampling anomaly refers to a failure in the battery sampling system that causes distortion of voltage data collected by a single or multiple sensors;

[0013] The mileage sampling abnormality refers to the malfunction of the battery sampling system causing distortion of the mileage data collected by one or more sensors;

[0014] The temperature sampling anomaly refers to a failure in the battery sampling system that causes distortion of temperature data collected by a single or multiple sensors;

[0015] The temperature sensor connection abnormality refers to the loose connection between battery cells or battery packs in the high-voltage circuit, causing distortion of battery temperature data;

[0016] The insulation failure refers to the insulation resistance of the battery system exceeding the preset range due to vehicle wading, battery cell leakage, and insulation layer aging;

[0017] The sudden internal short circuit refers to the sudden thermal runaway of the battery; the causes of the sudden thermal runaway of the battery include abnormal conduction of the positive and negative electrodes inside the battery cell;

[0018] The inconsistent discharge of different cells refers to the different discharge depths of different cells during use;

[0019] The self-discharge anomaly refers to the situation that the energy of the battery is not discharged into the external circuit but is lost in other ways, causing a single or multiple cells to be in an obvious outlier state and posing a safety hazard.

[0020] Further, the steps of using the box plot to filter the fault data set include:

[0021] a1) Use the interquartile range (IQR) to construct a box plot;

[0022] a2) Use box plots to divide the fault data set into four subsets to measure the statistical dispersion and variability of the data; the four subsets are separated by the first quartile Q1, the second quartile Q2, and the third quartile Q3;

[0023] a3) Outlier identification: If any element value in the fault data set is less than Q1-1.5IQR, or greater than Q3+1.5IQR, then the element is an outlier;

[0024] a4) Remove outliers in the faulty dataset.

[0025] Furthermore, after obtaining the fault optimization data set, the fault data set is also normalized.

[0026] Furthermore, the battery pack fault mode diagnosis model combines self-organizing mapping and back propagation, including a feedforward SOM network and a feedback BP network;

[0027] The neurons of the feedforward SOM network all have a weight vector;

[0028] After receiving the input data, the feedforward SOM network compares the weight vectors of all neurons and finds the neuron whose weight is closest to the input data as the winning neuron; the weights of the winning neuron and its neighboring neurons are updated based on the Kohonen learning rule, so that the structure of the feedforward SOM network forms an orderly mapping;

[0029] The input data is a vehicle operation signal;

[0030] The input of the feedback BP network is the output of the SOM layer, and the output is the battery pack fault identification result.

[0031] Furthermore, the feedback BP network includes an input layer, a hidden layer and an output layer;

[0032] Among them, the neurons in the input layer receive the output data of the SOM network and forward propagate the received data to the neurons in the hidden layer;

[0033] The hidden layer weights the received data and forward propagates it to the output layer;

[0034] The output layer outputs the battery pack fault identification result;

[0035] During the forward propagation process, the input matrix, weight matrix, and bias matrix of the feedback BP network are as follows:

[0036] Z0=(i1,i2,i3,…i m )(1)

[0037] W n =(W n 1 ,W n 2 ,W n 3 ,…Wn m )(2)

[0038] B n =(B n 1 ,B n 2 ,B n 3 ,…B n m )(3)

[0039] Where m represents the number of sample sets; n represents the total number of hidden layers and output layers; Z0 represents the input matrix, which is the fault data set; W n and B n Represent the weight matrix and bias matrix of the nth layer respectively. m is the i-th sample set; W n m , B n m is the weight matrix and bias matrix of the nth layer corresponding to the mth sample set;

[0040] The forward propagation process of the feedback BP network is as follows:

[0041] A0=Z0(4)

[0042] Z i =W i T A i-1 +B i , i=1…n(5)

[0043] A i =f i (Z i ), i=1…n(6)

[0044] In the formula, Z0 is the input of layer 0; A0 is the output of layer 0, layer 0 is the input layer; Z i is the input of the i-th layer; W i is the weight matrix of the i-th layer; B i is the bias matrix of the i-th layer; A i is the output of the i-th layer; n is the total number of hidden layers and output layers; f i () is the activation function of the i-th layer. A i-1 is the output of the i-1th layer.

[0045] Furthermore, the vehicle operation signal includes but is not limited to voltage data, mileage data, temperature data, insulation resistance, and battery cell discharge depth.

[0046] Furthermore, the steps of training the battery pack failure mode diagnosis model using the fault optimization data set include:

[0047] b1) dividing the fault optimization data set into a training set and a test set;

[0048] b2) Using the training set to train the battery pack fault mode diagnosis model; during the training process, the feedback BP network uses the back propagation algorithm to adjust the weights, and by comparing the difference between the predicted output and the actual target value, the gradient descent method is used to update the weights to reduce the error until the stop condition is met;

[0049] b3) Use the test set to test the battery pack failure mode diagnosis model. If the difference between the output result of the battery pack failure mode diagnosis model and the actual result is greater than a preset value, re-acquire the fault optimization data set and return to step b1).

[0050] Furthermore, the loss function L of the battery pack failure mode diagnosis model is as follows:

[0051]

[0052] Where: M is the total number of categories. ic is an indicator variable. If the true category of sample i is c, then y ic =1, otherwise y ic =0. is the probability that the model predicts that sample i belongs to category c. N is the number of samples.

[0053] The technical effect of the present invention is unquestionable. The present invention defines eight failure modes using different failure characteristics, and screens the fault data set of the battery pack using the box plot method. After normalizing the fault data set, the SOM-BP algorithm is used to build a battery pack failure mode recognition model, which can monitor various signals of the vehicle in real time, send early warning signals when abnormal signals are identified, and warn of failure modes. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 Flow chart of the battery pack failure mode identification method

[0055] Figure 2 It is the neighbor distance graph of SOM model;

[0056] Figure 3 It is the SOM sampling hit count map;

[0057] Figure 4 This is the relevant R value result diagram of the SOM-BP model;

[0058] Figure 5 This is the recognition confusion matrix diagram of the SOM-BP model. DETAILED DESCRIPTION

[0059] The present invention is further described below in conjunction with the embodiments, but it should not be understood that the above subject matter of the present invention is limited to the following embodiments. Without departing from the above technical ideas of the present invention, various substitutions and changes are made according to the common technical knowledge and customary means in the art, which should all be included in the protection scope of the present invention.

[0060] Embodiment 1:

[0061] See also Figures 1 to 5 , a method for battery pack failure mode identification, comprising the following steps:

[0062] 1) Based on the fault characteristics, define n fault modes and obtain the fault data set under each fault mode;

[0063] 2) Use the box plot to filter the fault data set and obtain the fault optimal data set;

[0064] 3) Use the self-organizing map-feedforward neural network algorithm to build a battery pack failure mode diagnosis model;

[0065] 4) Using the fault optimization data set to train the battery pack fault mode diagnosis model to obtain a trained battery pack fault mode recognition model;

[0066] 5) Monitor the vehicle operation signal and transmit it to the battery pack fault mode recognition model to obtain the battery pack fault recognition result; the battery pack fault recognition result includes no fault and fault; when the battery pack fault recognition result is faulty, it also includes a warning signal and a fault mode.

[0067] The failure modes include but are not limited to abnormal voltage sampling, abnormal mileage sampling, abnormal temperature sampling, abnormal temperature sensor connection, insulation failure, sudden internal short circuit, inconsistent discharge of different battery cells, and abnormal self-discharge.

[0068] The voltage sampling anomaly refers to a failure in the battery sampling system that causes distortion of voltage data collected by a single or multiple sensors;

[0069] The mileage sampling abnormality refers to the malfunction of the battery sampling system causing distortion of the mileage data collected by one or more sensors;

[0070] The temperature sampling anomaly refers to a failure in the battery sampling system that causes distortion of temperature data collected by a single or multiple sensors;

[0071] The temperature sensor connection abnormality refers to the loose connection between battery cells or battery packs in the high-voltage circuit, causing distortion of battery temperature data;

[0072] The insulation failure refers to the insulation resistance of the battery system exceeding the preset range due to vehicle wading, battery cell leakage, and insulation layer aging;

[0073] The sudden internal short circuit refers to the sudden thermal runaway of the battery; the causes of the sudden thermal runaway of the battery include abnormal conduction of the positive and negative electrodes inside the battery cell;

[0074] The inconsistent discharge of different cells refers to the different discharge depths of different cells during use;

[0075] The self-discharge anomaly refers to the situation that the energy of the battery is not discharged into the external circuit but is lost in other ways, causing a single or multiple cells to be in an obvious outlier state and posing a safety hazard.

[0076] The steps to use the box plot to filter the fault data set include:

[0077] a1) Use the interquartile range (IQR) to construct a box plot;

[0078] a2) Use box plots to divide the fault data set into four subsets to measure the statistical dispersion and variability of the data; the four subsets are separated by the first quartile Q1, the second quartile Q2, and the third quartile Q3;

[0079] a3) Outlier identification: If any element value in the fault data set is less than Q1-1.5IQR, or greater than Q3+1.5IQR, then the element is an outlier;

[0080] a4) Remove outliers in the faulty dataset.

[0081] After obtaining the fault optimization data set, the fault data set is also normalized.

[0082] The battery pack failure mode diagnosis model combines self-organizing mapping and back propagation, including a feedforward SOM network and a feedback BP network;

[0083] The neurons of the feedforward SOM network all have a weight vector;

[0084] After receiving the input data, the feedforward SOM network compares the weight vectors of all neurons and finds the neuron whose weight is closest to the input data as the winning neuron; the weights of the winning neuron and its neighboring neurons are updated based on the Kohonen learning rule, so that the structure of the feedforward SOM network forms an orderly mapping;

[0085] The input data is a vehicle operation signal;

[0086] The input of the feedback BP network is the output of the SOM layer, and the output is the battery pack fault identification result.

[0087] The feedback BP network includes an input layer, a hidden layer and an output layer;

[0088] Among them, the neurons in the input layer receive the output data of the SOM network and propagate the received data forward to the neurons in the hidden layer;

[0089] The hidden layer weights the received data and forward propagates it to the output layer;

[0090] The output layer outputs the battery pack fault identification result;

[0091] During the forward propagation process, the input matrix, weight matrix, and bias matrix of the feedback BP network are as follows:

[0092] Z0=(i1,i2,i3,…i m )(1)

[0093] W n =(W n 1 ,W n 2 ,W n 3 ,…Wn n m )(2)

[0094] B n =(B n 1 ,B n 2 ,B n 3 ,…B n m )(3)

[0095] Where m represents the number of sample sets; n represents the total number of hidden layers and output layers; Z0 represents the input matrix, which is the fault data set; W n and B n Represent the weight matrix and bias matrix of the nth layer respectively. m is the i-th sample set; W n m , B n m is the weight matrix and bias matrix of the nth layer corresponding to the mth sample set;

[0096] The forward propagation process of the feedback BP network is as follows:

[0097] A0=Z0(4)

[0098] Z i =W i T Ai-1 +B i , i=1…n(5)

[0099] A i =f i (Z i ), i=1…n(6)

[0100] In the formula, Z0 is the input of layer 0; A0 is the output of layer 0, layer 0 is the input layer; Z i is the input of the i-th layer; W i is the weight matrix of the i-th layer; B i is the bias matrix of the i-th layer; A i is the output of the i-th layer; n is the total number of hidden layers and output layers; f i () is the activation function of the i-th layer. A i-1 is the output of the i-1th layer.

[0101] The vehicle operation signal includes but is not limited to voltage data, mileage data, temperature data, insulation resistance, and battery cell discharge depth.

[0102] The steps of training the battery pack failure mode diagnosis model using the fault optimization dataset include:

[0103] b1) dividing the fault optimization data set into a training set and a test set;

[0104] b2) Using the training set to train the battery pack fault mode diagnosis model; during the training process, the feedback BP network uses the back propagation algorithm to adjust the weights, and by comparing the difference between the predicted output and the actual target value, the gradient descent method is used to update the weights to reduce the error until the stop condition is met;

[0105] b3) Use the test set to test the battery pack failure mode diagnosis model. If the difference between the output result of the battery pack failure mode diagnosis model and the actual result is greater than a preset value, re-acquire the fault optimization data set and return to step b1).

[0106] The loss function L of the battery pack failure mode diagnosis model is as follows:

[0107]

[0108] Where: M is the total number of categories. ic is an indicator variable. If the true category of sample i is c, then y ic =1, otherwise y ic =0. is the probability that the model predicts that sample i belongs to category c. N is the number of samples.

[0109] Embodiment 2:

[0110] A method for battery pack failure mode identification comprises the following steps:

[0111] 1) Based on the fault characteristics, define n fault modes and obtain the fault data set under each fault mode;

[0112] 2) Use the box plot to filter the fault data set and obtain the fault optimal data set;

[0113] 3) Use the self-organizing map-feedforward neural network algorithm to build a battery pack failure mode diagnosis model;

[0114] 4) Using the fault optimization data set to train the battery pack fault mode diagnosis model to obtain a trained battery pack fault mode recognition model;

[0115] 5) Monitor the vehicle operation signal and transmit it to the battery pack fault mode recognition model to obtain the battery pack fault recognition result; the battery pack fault recognition result includes no fault and fault; when the battery pack fault recognition result is faulty, it also includes a warning signal and a fault mode.

[0116] Embodiment 3:

[0117] A method for identifying battery pack failure modes, the technical content of which is the same as that of Example 2. Furthermore, the failure modes include but are not limited to abnormal voltage sampling, abnormal mileage sampling, abnormal temperature sampling, abnormal temperature sensor connection, insulation failure, sudden internal short circuit, inconsistent discharge of different battery cells, and abnormal self-discharge.

[0118] Embodiment 4:

[0119] A method for identifying a battery pack failure mode, the technical content of which is the same as any one of Embodiments 2-3, and further, the voltage sampling abnormality refers to a failure of a battery sampling system causing distortion of voltage data collected by a single or multiple sensors;

[0120] The mileage sampling abnormality refers to the malfunction of the battery sampling system causing distortion of the mileage data collected by one or more sensors;

[0121] The temperature sampling anomaly refers to a failure in the battery sampling system that causes distortion of temperature data collected by a single or multiple sensors;

[0122] The temperature sensor connection abnormality refers to the loose connection between battery cells or battery packs in the high-voltage circuit, causing distortion of battery temperature data;

[0123] The insulation failure refers to the insulation resistance of the battery system exceeding the preset range due to vehicle wading, battery cell leakage, and insulation layer aging;

[0124] The sudden internal short circuit refers to the sudden thermal runaway of the battery; the causes of the sudden thermal runaway of the battery include abnormal conduction of the positive and negative electrodes inside the battery cell;

[0125] The inconsistent discharge of different cells refers to the different discharge depths of different cells during use;

[0126] The self-discharge anomaly refers to the situation that the energy of the battery is not discharged into the external circuit but is lost in other ways, causing a single or multiple cells to be in an obvious outlier state and posing a safety hazard.

[0127] Embodiment 5:

[0128] A method for identifying a battery pack failure mode, the technical content of which is the same as any one of Embodiments 2-4, and further, the step of screening the failure data set using a box plot comprises:

[0129] a1) Use the interquartile range (IQR) to construct a box plot;

[0130] a2) Use box plots to divide the fault data set into four subsets to measure the statistical dispersion and variability of the data; the four subsets are separated by the first quartile Q1, the second quartile Q2, and the third quartile Q3;

[0131] a3) Outlier identification: If any element value in the fault data set is less than Q1-1.5IQR, or greater than Q3+1.5IQR, then the element is an outlier;

[0132] a4) Remove outliers in the faulty dataset.

[0133] Embodiment 6:

[0134] A method for identifying battery pack fault patterns, the technical content of which is the same as any one of Embodiments 2-5. Further, after obtaining the fault optimization data set, the fault data set is also normalized.

[0135] Embodiment 7:

[0136] A method for identifying a battery pack failure mode, the technical content of which is the same as any one of embodiments 2-6, and further, the battery pack failure mode diagnosis model combines self-organizing mapping and back propagation, including a feedforward SOM network and a feedback BP network;

[0137] The neurons of the feedforward SOM network all have a weight vector;

[0138] After receiving the input data, the feedforward SOM network compares the weight vectors of all neurons and finds the neuron whose weight is closest to the input data as the winning neuron; the weights of the winning neuron and its neighboring neurons are updated based on the Kohonen learning rule, so that the structure of the feedforward SOM network forms an orderly mapping;

[0139] The input data is a vehicle operation signal;

[0140] The input of the feedback BP network is the output of the SOM layer, and the output is the battery pack fault identification result.

[0141] Embodiment 8:

[0142] A method for identifying a battery pack failure mode, the technical content of which is the same as any one of Embodiments 2-7, and further, the feedback BP network includes an input layer, a hidden layer, and an output layer;

[0143] Among them, the neurons in the input layer receive the output data of the SOM network and propagate the received data forward to the neurons in the hidden layer;

[0144] The hidden layer weights the received data and forward propagates it to the output layer;

[0145] The output layer outputs the battery pack fault identification result;

[0146] During the forward propagation process, the input matrix, weight matrix, and bias matrix of the feedback BP network are as follows:

[0147] Z0=(i1,i2,i3,…i m )(1)

[0148] W n =(W n 1 ,W n 2 ,W n 3 ,…W n m )(2)

[0149] B n =(B n 1 ,B n 2 ,B n 3 ,…B n m )(3)

[0150] Where m represents the number of sample sets; n represents the total number of hidden layers and output layers; Z0 represents the input matrix, which is the fault data set; Wn and B n Represent the weight matrix and bias matrix of the nth layer respectively.

[0151] The forward propagation process of the feedback BP network is as follows:

[0152] A0=Z0(4)

[0153] Z i =W i T A i-1 +B i , i=1…n(5)

[0154] A i =f i (Z i ), i=1…n(6)

[0155] In the formula, Z0 is the input of layer 0; A0 is the output of layer 0, layer 0 is the input layer; Z i is the input of the i-th layer; W i is the weight matrix of the i-th layer; B i is the bias matrix of the i-th layer; A i is the output of the i-th layer; n is the total number of hidden layers and output layers; f i () is the activation function of the i-th layer.

[0156] Embodiment 9:

[0157] A method for identifying battery pack fault modes, the technical content of which is the same as any one of Examples 2-8, and further, the vehicle operation signal includes but is not limited to voltage data, mileage data, temperature data, insulation resistance, and battery cell discharge depth.

[0158] Embodiment 10:

[0159] A method for identifying a battery pack failure mode, the technical content of which is the same as any one of Embodiments 2-9, and further, the step of training a battery pack failure mode diagnosis model using a fault optimization data set comprises:

[0160] b1) dividing the fault optimization data set into a training set and a test set;

[0161] b2) Using the training set to train the battery pack fault mode diagnosis model; during the training process, the feedback BP network uses the back propagation algorithm to adjust the weights, and by comparing the difference between the predicted output and the actual target value, the gradient descent method is used to update the weights to reduce the error until the stop condition is met;

[0162] b3) Use the test set to test the battery pack failure mode diagnosis model. If the difference between the output result of the battery pack failure mode diagnosis model and the actual result is greater than a preset value, re-acquire the fault optimization data set and return to step b1).

[0163] Embodiment 11:

[0164] A method for identifying a battery pack failure mode, the technical content of which is the same as any one of Embodiments 2-10, and further, the loss function of the battery pack failure mode diagnosis model is as follows:

[0165]

[0166] Where: M is the total number of categories. ic is an indicator variable. If the true category of sample i is c, then y ic =1, otherwise y ic =0. is the probability that the model predicts that sample i belongs to category c.

[0167] Embodiment 12:

[0168] A method for battery pack failure mode identification comprises the following steps:

[0169] 1) Define eight fault modes based on different fault characteristics;

[0170] The eight failure modes defined based on different failure characteristics include:

[0171] 1.1) Voltage sampling anomaly is defined based on the fact that a battery sampling system fails, causing distortion of voltage data collected by a single or multiple sensors;

[0172] 1.2) Mileage sampling anomaly is defined based on the malfunction of the battery sampling system causing distortion of mileage data collected by one or more sensors;

[0173] 1.3) Define temperature sampling anomaly based on the failure of the battery sampling system causing distortion of temperature data collected by one or more sensors;

[0174] 1.4) The temperature sensor connection is abnormal according to the loose connection between the battery cells or battery packs in the high-voltage circuit, which causes the battery temperature data to be distorted;

[0175] 1.5) Insulation failure is defined based on the insulation resistance of the battery system due to vehicle wading, battery cell leakage, insulation layer aging or other reasons;

[0176] 1.6) The sudden internal short circuit is defined as the sudden thermal runaway of the battery due to abnormal conduction of the positive and negative electrodes inside the battery cell;

[0177] 1.7) During use, different discharge depths of different cells have different definitions of discharge inconsistency;

[0178] 1.8) Self-discharge anomaly is defined as the battery energy not being discharged into the external circuit but being lost in other ways, causing one or more cells to be in an obvious outlier state and posing a safety hazard.

[0179] 2) Use the box plot method to filter the fault data set;

[0180] The box plot is constructed using the concept of interquartile range (IQR) to measure the statistical dispersion and variability of the data by dividing the data set into four parts. The data set is divided into four defined intervals based on its value and how it compares to the entire data set. Among them, Q1 represents the first quartile and Q3 represents the third quartile, which are equal to the 25% and 75% corresponding values ​​of all values ​​in the sample arranged from small to large. Outliers are defined as values ​​less than the first quartile minus 1.5 times the interquartile range (IQR, that is, the difference between the third quartile and the first quartile), or greater than the third quartile plus 1.5 times the IQR.

[0181] 3) Divide the fault data set into training set and test set in a ratio of 0.7:0.3;

[0182] 4) Use the self-organizing map-feedforward neural network (SOM-BP) algorithm to build a battery pack failure mode diagnosis model;

[0183] The steps to build the SOM-BP model are:

[0184] 4.1) When building a neural network model, normalizing the fault data set is an important preprocessing step. Normalization is used to scale the data features to a small specific interval [0, 1].

[0185] 4.2) The SOM-BP model usually consists of two parts: a feedforward SOM layer and a feedback BP layer. The input data is first passed to the SOM layer. Each neuron has a weight vector. When a new input arrives, it is compared with the weight vectors of all neurons to find the closest neuron (i.e., the winning neuron). The weights of the winning neuron and its neighboring neurons are updated according to a certain learning rule (such as the Kohonen learning rule) so that the structure of the entire network gradually forms an ordered mapping. The output of the SOM layer serves as the input of the BP layer. The BP network uses the back-propagation algorithm to adjust the weights. During the training process, the BP network updates the weights using the gradient descent method by comparing the difference between the predicted output and the actual target value (i.e., the error) to reduce the error until the stopping condition is met.

[0186] The BP neural network consists of three parts, namely the input layer, hidden layer and output layer. The neurons in the input layer are responsible for receiving data values ​​and forward propagating them to the neurons in the middle layer of the neural network, namely the hidden layer. The weighted sum of the hidden layer is finally forward propagated to the output layer, which will display the output results of the neural network. The forward propagation and back propagation mentioned here are the processes of training deep learning models. The BP neural network model alternately performs forward propagation and back propagation, and iterates the model parameters based on the gradient calculated by back propagation.

[0187] 4.3) Cross entropy is used as the loss function. Cross entropy loss is a widely used loss function in classification problems, especially when dealing with multi-classification problems. It measures the difference between the probability distribution predicted by the model and the probability distribution of the true label. In machine learning and deep learning, the cross entropy loss function is often used to optimize the parameters of the classification model.

[0188] For multi-classification problems, the cross entropy loss usually uses the softmax function to convert the output of the model into a probability distribution and then calculates the cross entropy. The formula is:

[0189]

[0190] Where: M is the total number of categories. ic is an indicator variable. If the true category of sample i is c, then y ic =1, otherwise y ic =0. is the probability that the model predicts that sample i belongs to category c.

[0191] 5) Adjust the learning rate, number of trainings, target training error and other parameters to optimize the recognition accuracy of the model until the highest accuracy battery pack fault pattern recognition model is obtained;

[0192] 6) Use the battery pack fault pattern recognition model to monitor various vehicle signals in real time, send warning signals when abnormal signals are identified, and warn of fault patterns.

[0193] Embodiment 13:

[0194] like Figure 1 As shown, a method for battery pack failure mode identification includes the following steps:

[0195] S1. Define eight failure modes based on different failure characteristics;

[0196] In this embodiment, the fault characteristics include signals such as total voltage, total mileage, total internal resistance, cell voltage, cell temperature, etc.

[0197] Step S1 includes the following sub-steps:

[0198] S11, defining voltage sampling abnormality according to a failure of a battery sampling system causing distortion of voltage data collected by a single or multiple sensors;

[0199] In this embodiment, the specific operation of step S11 is: observing whether there is distortion in the single cell voltage of the battery pack system.

[0200] S12, defining mileage sampling abnormality according to a failure of a battery sampling system causing distortion of mileage data collected by a single or multiple sensors;

[0201] In this embodiment, the specific operation of step S12 is: observing whether there is distortion in the mileage of the battery pack system.

[0202] S13, defining temperature sampling abnormality according to a failure of a battery sampling system causing distortion of temperature data collected by a single or multiple sensors;

[0203] In this embodiment, the specific operation of step S13 is: observing whether there is distortion in the highest and lowest temperatures of the battery pack system.

[0204] S14, defining abnormal connection of the temperature sensor according to loose connection between battery cells or battery packs in the high-voltage circuit, resulting in distortion of battery temperature data;

[0205] In this embodiment, the specific operation of step S14 is: observing whether the temperature of each temperature probe of the battery pack system is distorted.

[0206] S15. Insulation failure is defined based on the insulation resistance of the battery system due to vehicle wading, battery cell leakage, insulation layer aging or other reasons;

[0207] In this embodiment, the specific operation of step S15 is: observing whether the total resistance of the battery pack system is too low.

[0208] S16. Define sudden internal short circuit based on the fact that the battery suddenly experiences thermal runaway due to abnormal conduction of the positive and negative electrodes inside the battery cell;

[0209] In this embodiment, the specific operation of step S16 is: observing whether there is a timing mutation in the maximum temperature of the battery pack system.

[0210] S17. During use, different discharge depths of different cells are defined as inconsistent discharges;

[0211] In this embodiment, the specific operation of step S17 is: observing whether there is a long-term significant difference in the voltages of different cells of the battery pack system.

[0212] S18. Self-discharge anomaly is defined as the situation where the energy of the battery is not discharged into the external circuit but is lost in other ways, causing a single or multiple cells to be in an obvious outlier state and posing a safety hazard.

[0213] In this embodiment, the specific operation of step S18 is: observing whether the voltage of the battery pack system decreases sequentially when there is no power demand in the external circuit.

[0214] S2, using the box plot method to filter the fault data set;

[0215] In this embodiment, step S2 is specifically: using the box plot method to draw the distribution of each characteristic signal, and screening out values ​​that are less than the first quartile minus 1.5 times the interquartile range (IQR, i.e., the difference between the third quartile and the first quartile), or greater than the third quartile plus 1.5 times the IQR.

[0216] S3, divide the fault data set into a training set and a test set in a ratio of 0.7:0.3;

[0217] The training set data in step S3 is used to train and build a battery pack system fault mode recognition model, and the test set is used to test the recognition accuracy of the built model.

[0218] S4. Use the self-organizing map-feedforward neural network (SOM-BP) algorithm to build a battery pack failure mode diagnosis model;

[0219] The step S4 comprises the following sub-steps:

[0220] S41. Normalize the fault data set and scale the data features to a small specific interval [0,1].

[0221] S42, combining the normalized data with the SOM-BP algorithm to build a battery pack system fault mode recognition model;

[0222] S43. Use cross entropy as the loss function to optimize the battery pack system fault pattern recognition model.

[0223] When the SOM-BP model is trained or tested in steps S42 and S43, the normalized fault data is used as the input of the SOM-BP model, and the corresponding fault mode is used as the output of the SOM-BP model.

[0224] The SOM-BP model in step S4 is usually composed of two parts: a feedforward SOM layer and a feedback BP layer. The input data is first passed to the SOM layer. Each neuron has a weight vector. When a new input arrives, it is compared with the weight vectors of all neurons to find the closest neuron (i.e., the winning neuron). The weights of the winning neuron and its neighboring neurons are updated according to the Kohonen learning rule, so that the structure of the entire network gradually forms an orderly mapping. The output of the SOM layer is used as the input of the BP layer. The BP network uses a back-propagation algorithm to adjust the weights. During the training process, the BP network updates the weights using the gradient descent method by comparing the difference (i.e., error) between the predicted output and the actual target value to reduce the error until the stop condition is met.

[0225] The BP neural network consists of three parts, namely the input layer, hidden layer and output layer. The neurons in the input layer are responsible for receiving data values ​​and forward propagating them to the neurons in the middle layer of the neural network, namely the hidden layer. The weighted sum of the hidden layer is finally forward propagated to the output layer, which will display the output results of the neural network. The forward propagation and back propagation mentioned here are the processes of training deep learning models. The BP neural network model alternately performs forward propagation and back propagation, and iterates the model parameters based on the gradient calculated by back propagation.

[0226] The neurons in the input layer are responsible for receiving data values ​​and forward propagating them to the neurons in the middle layer of the neural network, i.e. the hidden layer. The weighted sum of the hidden layer is finally forward propagated to the output layer, which will display the output results of the neural network. During training, the SOM-BP model alternately performs forward propagation and back propagation, and iterates the model parameters according to the gradient calculated by back propagation.

[0227] The forward propagation is:

[0228] A0=z0

[0229] Z i =W i T A i-1 +B i , i=1…n

[0230] A i =f i (Z i ), i=1…n

[0231] Among them, Z0 is the input of layer 0, A0 is the output of layer 0, and Z i is the input of the i-th layer, W i is the weight matrix of the i-th layer, B i is the bias matrix of the i-th layer, A iis the output of the i-th layer, n is the total number of hidden layers and output layers, f i () is the activation function of the i-th layer, the hidden layer selects the Tansig function as the activation function, and the Purelin function is used as the activation function between the hidden layer and the output layer.

[0232] After a forward propagation, the error between the output of the current SOM-BP model and the target output is calculated. After the error is obtained, the weight matrix and the bias matrix are adjusted through back propagation. After adjustment, forward propagation is performed again until the error meets the requirements.

[0233] During the training and testing process, the parameters of the deep learning model that are adjusted also include: the number of hidden layers, the number of neurons in each hidden layer, and the learning rate.

[0234] The cross entropy loss function is used as the loss function. The cross entropy loss is a widely used loss function in classification problems, especially when dealing with multi-classification problems. It measures the difference between the probability distribution predicted by the model and the probability distribution of the true label. In machine learning and deep learning, the cross entropy loss function is often used to optimize the parameters of the classification model.

[0235] For multi-classification problems, the cross entropy loss usually uses the softmax function to convert the output of the model into a probability distribution and then calculates the cross entropy. The formula is:

[0236]

[0237] Where: M is the total number of categories. ic is an indicator variable. If the true category of sample i is c, then y ic =1, otherwise y ic =0. is the probability that the model predicts that sample i belongs to category c.

[0238] S5. Adjust the learning rate, number of trainings, target training error and other parameters to optimize the recognition accuracy of the model until the highest accuracy battery pack fault pattern recognition model is obtained;

[0239] S6. Use the battery pack fault pattern recognition model to monitor various signals of the vehicle in real time, send a warning signal when an abnormal signal is identified, and warn of the fault mode.

[0240] Experimental results:

[0241] 1. Figure 2It is the neighbor distance graph of the SOM model. In the SOM network, the neighbor distance usually refers to the Euclidean distance between two neurons on the two-dimensional or high-dimensional grid of the network. For a two-dimensional SOM, the neighbor distance can be obtained by calculating the square root of the sum of the squares of the differences in the horizontal and vertical positions of the two neurons on the grid. The neighbor distance is a core concept in SOM. It not only affects the learning process of the network, but also determines how the network maintains the topological structure of the input data in a low-dimensional space. By adjusting the neighbor distance, SOM can effectively reduce the dimension and visualize the data. Figure 3 Sample hit count graph for SOM. Sample hit count (or “hit count”) usually refers to the number of times each neuron is selected as the winning neuron during the SOM training process. This concept is an important part of understanding how SOM learns and organizes its neurons.

[0242] 2. Figure 4 The result graph of the correlation R value of the SOM-BP model. The Pearson correlation coefficient (R) is a value between -1 and 1, which is used to measure the strength and direction of the linear relationship between two variables X and Y. When R = 1, it means that there is a completely positive linear relationship between X and Y, that is, Y increases as X increases. When R = -1, it means that there is a completely negative linear relationship between X and Y, that is, Y decreases as X increases. When R = 0, it means that there is no linear relationship between X and Y. The higher the R value (whether positive or negative), the stronger the predictive ability of the model is generally considered to be, because it shows that the model can better explain the variability of the dependent variable. The results show that the training set, test set, validation set and all data have high R values, showing extremely high recognition accuracy of the pattern.

[0243] 3. Figure 5 This is the recognition confusion matrix diagram of the SOM-BP model. The left side is the recognition accuracy of the training set, and the right side is the recognition accuracy of the test set. The results show that the constructed SOM-BP model has a high recognition accuracy (93.8%-100%) and can accurately identify various failure modes of the battery pack system.

[0244] In summary, this embodiment comprehensively considers the problem of fault pattern recognition and early warning when a battery pack system fails during long-term use. The results show that the established SOM-BP model can accurately identify various fault modes of the battery pack system and can be used for pattern recognition and early warning when a battery pack system fails during long-term use. In addition, this SOM-BP model method can be used in the design of a battery safety early warning system to promptly detect various faults generated by the battery pack system during use and provide early warnings, ensuring that the occupants have sufficient processing time to ensure that there are no potential safety hazards in the battery pack system during vehicle driving.

Claims

1. A method for identifying a battery pack failure mode, characterized in that: The following steps are involved: 1) Based on the fault characteristics, n fault modes are defined and the fault data set under each fault mode is obtained. 2) Use the box plot to filter the fault data set and obtain the fault optimal data set; 3) Use the self-organizing map-feedforward neural network algorithm to build a battery pack failure mode diagnosis model; 4) Using the fault optimization data set to train the battery pack fault mode diagnosis model to obtain a trained battery pack fault mode recognition model; 5) Monitor the vehicle operation signal and transmit it to the battery pack fault mode recognition model to obtain the battery pack fault recognition result; the battery pack fault recognition result includes no fault and fault; when the battery pack fault recognition result is faulty, it also includes a warning signal and a fault mode.

2. A method for battery pack fault mode identification according to claim 1, characterized in that: The failure modes include but are not limited to abnormal voltage sampling, abnormal mileage sampling, abnormal temperature sampling, abnormal temperature sensor connection, insulation failure, sudden internal short circuit, inconsistent discharge of different battery cells, and abnormal self-discharge.

3. A method for battery pack fault mode identification according to claim 2, characterized in that: The voltage sampling anomaly refers to a failure in the battery sampling system that causes distortion of voltage data collected by a single or multiple sensors; The mileage sampling abnormality refers to the malfunction of the battery sampling system causing distortion of the mileage data collected by one or more sensors; The temperature sampling anomaly refers to a failure in the battery sampling system that causes distortion of temperature data collected by a single or multiple sensors; The temperature sensor connection abnormality refers to the loose connection between battery cells or battery packs in the high-voltage circuit, causing distortion of battery temperature data; The insulation failure refers to the insulation resistance of the battery system exceeding the preset range due to vehicle wading, battery cell leakage, and insulation layer aging; The sudden internal short circuit refers to the sudden thermal runaway of the battery; the causes of the sudden thermal runaway of the battery include abnormal conduction of the positive and negative electrodes inside the battery cell; The inconsistent discharge of different cells refers to the different discharge depths of different cells during use; The self-discharge anomaly refers to the situation that the energy of the battery is not discharged into the external circuit but is lost in other ways, causing a single or multiple cells to be in an obvious outlier state and posing a safety hazard.

4. A method for battery pack fault mode identification according to claim 1, characterized in that: The steps to use the box plot to filter the fault data set include: 1) Use the interquartile range (IQR) to construct a box plot. 2) Use box plots to divide the fault data set into four subsets to measure the statistical dispersion and variability of the data; the four subsets are separated by the first quartile Q1, the second quartile Q2, and the third quartile Q3. 3) Outlier identification: If any element value in the fault data set is less than Q1-1.5IQR, or greater than Q3+1.5IQR, then the element is an outlier; 4) Remove outliers in the faulty dataset.

5. The method for battery pack fault mode identification according to claim 1, characterized in that: After obtaining the fault optimization data set, the fault data set is also normalized.

6. A method for battery pack fault mode identification according to claim 1, characterized in that: The battery pack failure mode diagnosis model combines self-organizing mapping and back propagation, including a feedforward SOM network and a feedback BP network; The neurons of the feedforward SOM network all have a weight vector; After receiving the input data, the feedforward SOM network compares the weight vectors of all neurons and finds the neuron whose weight is closest to the input data as the winning neuron; the weights of the winning neuron and its neighboring neurons are updated based on the Kohonen learning rule, so that the structure of the feedforward SOM network forms an orderly mapping; The input data is a vehicle operation signal; The input of the feedback BP network is the output of the SOM layer, and the output is the battery pack fault identification result.

7. A method for battery pack fault mode identification according to claim 6, characterized in that: The feedback BP network includes an input layer, a hidden layer and an output layer; Among them, the neurons in the input layer receive the output data of the SOM network and forward propagate the received data to the neurons in the hidden layer; The hidden layer weights the received data and forward propagates it to the output layer; The output layer outputs the battery pack fault identification result; During the forward propagation process, the input matrix, weight matrix, and bias matrix of the feedback BP network are as follows: Z0=(i1,i2,i3,…i m ) (1) IN n =(In n 1 ,IN n 2 ,IN n 3 ,…IN n m ) (2) B n =(B n 1 ,B n 2 ,B n 3 ,…B n m ) (3) Where m represents the number of sample sets; n represents the total number of hidden layers and output layers; Z0 represents the input matrix, which is the fault data set; W n and B n Represent the weight matrix and bias matrix of the nth layer respectively; i m is the i-th sample set; W n m , B n m is the weight matrix and bias matrix of the nth layer corresponding to the mth sample set; The forward propagation process of the feedback BP network is as follows: A0=Z0(4) Z i =W i T A i-1 +B i ,i=1…n(5) A i =f i (Z i ),i=1…n(6) In the formula, Z0 is the input of layer 0; A0 is the output of layer 0, layer 0 is the input layer; Z i is the input of the i-th layer; W i is the weight matrix of the i-th layer; B i is the bias matrix of the i-th layer; A i is the output of the i-th layer; n is the total number of hidden layers and output layers; f i () is the activation function of the i-th layer; A i-1 is the output of the i-1th layer.

8. The method for battery pack fault mode identification according to claim 1, characterized in that: The vehicle operation signal includes but is not limited to voltage data, mileage data, temperature data, insulation resistance, and battery cell discharge depth.

9. A method for battery pack fault mode identification according to claim 1, characterized in that: The steps of training the battery pack failure mode diagnosis model using the fault optimization dataset include: 1) Divide the fault optimization data set into a training set and a test set; 2) The battery pack failure mode diagnosis model is trained using the training set; during the training process, the feedback BP network uses the back propagation algorithm to adjust the weights, and the gradient descent method is used to update the weights by comparing the difference between the predicted output and the actual target value to reduce the error until the stopping condition is met; 3) Use the test set to test the battery pack failure mode diagnosis model. If the difference between the output result of the battery pack failure mode diagnosis model and the actual result is greater than a preset value, re-acquire the fault optimization data set and return to step 1).

10. A method for battery pack fault mode identification according to claim 9, characterized in that: The loss function L of the battery pack failure mode diagnosis model is as follows: Where: M is the total number of categories; y ic is an indicator variable. If the true category of sample i is c, then y ic =1, otherwise y ic =0. is the probability that the model predicts that sample i belongs to category c; N is the number of samples.

Images