Vehicle lithium battery fault early warning method based on multivariate prediction and anomaly detection

Abstract

The invention discloses a vehicle lithium battery fault early warning method based on multivariate prediction and anomaly detection. The method comprises the steps of collecting operation data of a vehicle battery system through a big data platform; performing data cleaning on the collected data and forming a data set; dividing the data set into a training set and a test set; constructing a prediction model through the training set; inputting the test set into the prediction model, and obtaining a prediction result by taking the highest monomer voltage and the highest probe temperature as prediction label values; and outputting an early warning result. According to the invention, high-precision synchronous forward prediction is carried out on the voltage and the temperature of the battery, and an experiment result shows that compared with an existing battery management system (BMS) early warning method, voltage over-limit early warning can be realized 4.5 minutes in advance, abnormal voltage rise and fall early warning before under-voltage can be realized 83.5 minutes in advance, temperature alarm can be realized 24.6 minutes in advance, and the method has the advantages of high reliability, high reliability, high reliability and the like. And temperature early warning is realized 35.1 minutes before a fault occurs, so that sufficient response time is provided for a driver, and the safety performance of the vehicle is remarkably enhanced.

Description

Technical Field

[0001] The present invention relates to the field of battery management technology fault diagnosis, and in particular to a vehicle lithium battery fault early warning method based on multivariate prediction and abnormality detection. Background Art

[0002] Accurate battery fault diagnosis and early warning before dangerous faults occur are the key to ensuring the safe, stable and reliable operation of electric vehicle power battery systems, and provide theoretical, methodological and technical support for the precise operation and maintenance of battery systems. At present, power battery fault diagnosis and early warning methods are mainly divided into three categories: statistical analysis-based methods, battery model-based methods and data-driven methods. The statistical analysis-based method uses various statistical methods such as entropy, Gaussian distribution, correlation coefficient, maximum likelihood, etc. to directly analyze the collected data through the data information collected by different sensors, and finally realizes fault diagnosis by setting reasonable thresholds. The core of the battery model-based method is to establish an accurate battery model and realize fault alarm by comparing the residual between the model prediction value and the actual value. In contrast, the data-driven method uses techniques such as machine learning and statistical analysis to analyze the historical and real-time data of the battery to detect and predict potential faults.

[0003] However, statistical analysis methods usually rely on some simple thresholds or specific statistical indicators, but when faced with complex, nonlinear, and multivariate lithium-ion battery fault diagnosis tasks, they are often unable to effectively detect in the early stages. The disadvantage of the battery model-based method is that it relies on a high-precision battery model and has a weak early warning capability for faults. The data-driven method overcomes the difficulty of establishing complex battery models, especially electrochemical models, and uses a large amount of vehicle operation data for analysis, which has significant advantages in dealing with nonlinear systems. Although there have been a lot of studies on data-driven methods, most of them focus on using historical data for fault diagnosis, while the prediction of data often only achieves real-time prediction or single-step prediction (i.e., predicting data at a future time point in advance), which makes the research biased towards diagnosis after the fault occurs, while fault warning can only achieve real-time or short-term early warning. In addition, most studies only consider the impact of a single output result on the fault, resulting in the warning effect being limited to a single dimension, such as only diagnosing or warning voltage or temperature faults, and failing to fully consider the progressive relationship between voltage faults and temperature faults, so the warning effect is often single and limited. The vehicle battery management system (BMS) itself often issues an alarm only when various levels of faults (generally divided into three levels) occur or when there is a relative lag. For the research on multi-fault early warning, due to the problem of reduced accuracy of early warning caused by the decrease in precision in multivariate long-term prediction, at this stage, the research on long-term early warning of multiple faults in lithium-ion batteries is still relatively scarce, or it is impossible to achieve high-precision long-term early warning of multiple faults. Establishing methods that meet actual application scenarios and developing technologies that can achieve accurate long-term early warning of multiple faults in lithium-ion batteries are of great significance to ensuring personal safety and promoting the development of electric vehicles. Summary of the invention

[0004] In order to solve the above-mentioned problems of single warning fault for two important parameters, voltage and temperature, decreased accuracy in warning of multiple faults, and inability to provide long-term warning, the present invention provides a vehicle lithium battery fault warning method based on multivariate prediction and anomaly detection.

[0005] To implement the above technical solution, the details are as follows:

[0006] S1. Collect the operating data of the vehicle battery system through the big data platform;

[0007] The vehicle battery system includes: a battery system of a normal vehicle and a vehicle that has had a fault alarm;

[0008] The fault types include: voltage fault and temperature fault;

[0009] The voltage faults include: battery cell overvoltage, undervoltage, and rapid voltage change;

[0010] The temperature faults include: temperature rise too fast, temperature exceeding the set threshold;

[0011] The operating data of the vehicle battery system collected include: vehicle ID, sampling time, total voltage of the battery pack (TV), maximum voltage of the single cell (HCV), minimum voltage of the single cell (LCV), SOC, total current (C), maximum temperature of the probe (HT) and minimum temperature of the probe (LT);

[0012] S2, clean the collected data and generate a data set;

[0013] Due to the complex operating conditions, the collected data needs to be cleaned as necessary;

[0014] Data cleaning methods include:

[0015] (1) Linear interpolation is used to fill in missing data of less than three frames, and missing segments of more than three frames are removed; that is, the sampling frequency is once every ten seconds, which is six data per minute. If 1-3 data are missing in this segment, linear interpolation is used to fill the missing data of this segment, and if more than 3 data are missing, this segment is directly deleted;

[0016] (2) Compare the total voltage, total current, maximum and minimum voltage of the single cell, and maximum and minimum temperature of the probe with the data collected within the normal range; the normal range of SOC is [200V, 500V], [-1000A, 1000A], [2V, 5V], [-40℃, 210℃], [0, 100], and data outside the range is considered as abnormal sensor data and should be discarded;

[0017] (3) Use wavelet transform to reduce noise of all data;

[0018] The data set is generated as follows: the cell maximum voltage (HCV) and the probe maximum temperature (HT) are randomly selected as the last two columns, and the battery pack total voltage (TV), total current (C), SOC, cell minimum voltage (LCV), and probe minimum temperature (LT) are not in a fixed order, forming a matrix X with a dimension of [N,7]. The data set is divided by identifying the vehicle ID, and the vehicle ID itself is not used as a data set.

[0019] The expression is as follows:

[0020]

[0021] The maximum voltage of the battery cell is selected as the predicted label value Y1, and the maximum temperature of the probe is selected as the predicted label value Y2; the expression is as follows:

[0022]

[0023] Where N represents the total length of the matrix composed of all data, which is equal to the sum of the number of rows in the training set and test set matrices;

[0024] S3, divide the data set into training set and test set;

[0025] Specifically, the normal vehicle data is used as the training set to obtain the training matrix X tr , the faulty vehicle is used as the test set to obtain the test matrix X te , the training set and the test set keep the same data form;

[0026] That is, the data matrix consisting of normal vehicles No. 1 and No. 2 is used as the training set, and the data matrix consisting of faulty vehicle No. 3 is used as the test set;

[0027] The expression is as follows:

[0028]

[0029]

[0030] In the formula, n represents the total number of rows in the training set matrix;

[0031] S4. Build a prediction model through the training set; the steps are as follows:

[0032] S4.1. Input the training set into the TCN model and perform causal convolution to extract input features. The output formula of causal convolution is:

[0033]

[0034] In the formula, represents the output sequence at time step t and output channel c out The value on c in is the number of input channels, which is equal to the number of input features and has a value of 7; c out is the number of output channels, which is equal to the number of convolution kernels. Its value is 256, which means that the extracted features are converted into 256-dimensional data. z is the size of the convolution kernel. Its value is 3. In time convolution, this value determines how many time steps the convolution kernel covers. i represents the position of the convolution kernel in the time dimension. is the weight of the convolution kernel, indicating that in the output channel c out , input channel c in , the weight parameter at the convolution kernel position i, the initial value is a random value with a mean of 0 and a standard deviation of 0.01 that conforms to the normal distribution. These weights determine how the convolution kernel performs weighted summation of the input features to generate the output; is the input data at time step ti and input channel c inHere ti represents the input value of a lag relative to the current time step t; Represents the bias term of each output channel. The initial value is 0 and there is no fixed value. The bias term will be continuously adjusted during the training process. The bias term is used to add an additional linear offset to the weighted sum result of the convolution to increase the flexibility of the model.

[0035] S4.2, use the output after extracting the input features as the input of LSTM to obtain the hidden state of the current time step t;

[0036] Specifically, the output after extracting the input features is used as the input of LSTM. After the features enter the LSTM, the input gate, forget gate, candidate cell state, cell state update, output gate, hidden state update calculation are performed in sequence. The long-term dependency is further captured through the LSTM model, and the prediction value is finally output through the fully connected layer. The expression is as follows:

[0037]

[0038] In the formula, i t represents the activation value of the input gate, which is used to control the amount of new information added to the cell state; σ represents the sigmoid activation function, whose output value is between 0 and 1, which is used to control the proportion of input information added; W i represents the weight matrix of the input gate; z t-1 Represents the hidden state of the previous time step, the value is 128; represents the output sequence at time step t and output channel c out The value on is used as the input of the LSTM model; b means concatenating the hidden state of the previous time step with the features extracted by TCN at the current time step; b i represents the bias term of the input gate; f t is the activation value of the forget gate; W f is the weight matrix of the forget gate; b f is the bias term of the forget gate; is the candidate cell state; tanh represents the hyperbolic tangent activation function; W c is the weight matrix used to generate candidate cell states; b c is the bias term of the candidate cell state; c t is the cell state at the current time step; f t ⊙c t-1 is the ratio of the previous cell state f t , indicating the selectively forgotten part; is the candidate state update under the input gate state; ⊙ represents element-wise multiplication; o t is the activation value of the output gate; W ois the weight matrix of the output gate; b o is the bias term of the output gate; z t is the hidden state of LSTM at the current time step t; tanh(c t ) indicates a nonlinear change in the cell state;

[0039] S4.3, LSTM calculates the hidden state of the current time step t and passes it to the fully connected layer, mapping the hidden state to the target prediction space to obtain the predicted value. The expression is as follows:

[0040]

[0041] In the formula, is the predicted result of voltage or temperature at time step t; W y is the weight matrix of the fully connected layer; z t is the hidden state of LSTM at the current time step t; b y is the bias term of the fully connected layer. Since binary data (voltage and temperature) are predicted, the dimension of the bias term is 2, which means that the hidden state is mapped to two output targets.

[0042] S4.4. After the training set is input into the model, all weight matrices and bias items in S4.1-S4.3 will be initialized and assigned. In the forward propagation, the model will use the currently initialized parameter values ​​to calculate the input with the above formula until the final prediction result is output. The prediction value obtained by forward propagation The error between the real label and the real label is calculated through the loss function, and then back propagation is performed to calculate the gradient of the loss function with respect to each parameter. Back propagation will pass back the influence of the loss function on each parameter. Then, the weight and bias parameters obtained in the above steps are updated through the Adam optimizer to reduce the loss function. The present invention uses the mean square error (MSE) as the training set loss function, and the expression is as follows:

[0043]

[0044] In the formula, n represents the length of the training set data, here n value is 907578, Y t represents the true value of voltage or temperature at time step t, is the predicted voltage or temperature value at time t;

[0045] S5, input the test set into the prediction model, and obtain the prediction result by taking the highest monomer voltage and the highest probe temperature as the prediction label value;

[0046] Specifically, after the test set is input into the prediction model, the model will repeat the formula obtained by inputting the training set into the model to calculate and call the obtained training parameters. The steps to obtain the prediction result with the highest monomer voltage and the highest probe temperature as the prediction label value are as follows:

[0047] S5.1. Input the test set into the TCN model and perform causal convolution to extract input features. The expression of extracted features is as follows:

[0048]

[0049] In the formula, Represents the input sequence X t-τ+1:t Output value after feature extraction; X t-τ+1:t represents the test set input sequence of the past τ time steps; τ is the time window size, and its value is 126;

[0050] S5.2. Pass the features extracted from TCN as input to LSTM to obtain the hidden state z of the time step t ; The expression is as follows:

[0051]

[0052] In the formula, z t-τ+1:t Represents the hidden state sequence obtained after accepting feature input;

[0053] S5.3. Generate the predicted value of the next Δ time steps by inputting the hidden state into the fully connected layer; the expression is as follows:

[0054]

[0055] In the formula, Indicates the maximum voltage of the monomer and the maximum temperature of the probe in the next Δ time steps, where the value of Δ is 24; [HCV t+1:t+Δ ,HT t+1:t+Δ ] is the prediction vector of voltage and temperature for the next Δ time steps; W y is the weight matrix of the fully connected layer; z t is the hidden state of LSTM at the current time step t. Here, only the hidden state obtained at the last time step t is used; b y It is the bias term of the fully connected layer. The bias term will be continuously adjusted during the training process. The bias term is used to add an additional linear offset to the weighted sum result of the convolution to increase the flexibility of the model.

[0056] Specifically, when the test set enters the TCN-LSTM model, the data is divided into M samples, each of which contains a sliding window of length 126 and a label output of prediction length 24. As sample data 1 enters the model, the trained model parameters are called to obtain the highest single cell voltage label data. Then sample data 2 enters the model and repeats the above process to obtain label data And so on, for the last sample Get label data M is the number of samples, and its value is Nn-126-24+1=29766; the specific prediction process is as follows:

[0057]

[0058]

[0059] Similarly, the predicted data of the probe's maximum temperature can be obtained according to the above method. like The prediction process is:

[0060]

[0061] In the formula, X te represents the test set matrix; Indicates the first sample of the test set entering the TCN-LSTM model, and so on Indicates the Mth sample of the test set entering the model, which is also the last sample; n indicates the data length of the training set, which is equal to the number of matrix rows; N indicates the total length of the training set and test set data; Y1 (1) Indicates the highest single cell voltage prediction data vector obtained by the first sample, and so on. (M) represents the highest monomer voltage prediction data vector obtained by the Mth sample;

[0062] In order to better quantify the difference between the predicted value and the true value obtained by the test set, the root mean square error (RMSE) is used as the evaluation indicator for the prediction performance of the model test set. The calculation formula of RMSE is as follows:

[0063]

[0064] In the formula, Nn represents the length of the test set data, Y t Indicates the true value of voltage or temperature, is the predicted voltage or temperature value at time step t;

[0065] S6, output warning results;

[0066] Specifically, when the predicted voltage or temperature value exceeds the set upper and lower limits, an overvoltage, undervoltage or temperature abnormality alarm is triggered; when the predicted value does not exceed the set upper and lower limits, but the predicted value is detected by the local anomaly detection algorithm (LOF) and the LOF outlier factor value exceeds 1, an abnormal voltage rise or fall alarm or abnormal temperature rise or fall alarm is triggered;

[0067] Here are the steps:

[0068] S6.1, standardize the prediction results;

[0069] Specifically, after obtaining the predicted values ​​of voltage data and temperature data, the input data needs to be standardized first to ensure that the data is compared on the same scale; the standardization formula is:

[0070]

[0071] In the formula, is the predicted voltage or temperature value obtained at time t; μ j is the mean value of voltage or temperature characteristic j; σ j is the standard deviation of the voltage or temperature characteristic j; z j is the standardized voltage or temperature value;

[0072] S6.2. The standardized voltage data or temperature data are combined into data sets D and U respectively, and the local outlier factor of each point in the data sets D and U is calculated to obtain the warning result; the details are as follows:

[0073] S6.2.1. Define the Euclidean distance; that is, take any data point p in the set and any point o in D or U except p, and define the Euclidean distance between point p and point o as d (p,o) , for any two points p=(p1,p2,...,p m ) and o=(o1,o2,...,o m )The Euclidean distance formula in m-dimensional space is:

[0074]

[0075] S6.2.2. Define the kth distance; that is, define the distance between point p and its nearest k points as d 1～k (p), d 1～k The maximum value in (p) is defined as d k (p), d k (p) is defined as the kth distance of point p;

[0076] d k (p) must satisfy the following two conditions: (1) In addition to point p, there are at least k points o' in D or U such that d k (p,o′)≤dk (p,o); (2) In D or U, there are at most k-1 points o' other than point k such that d(p,o') < d(p,o); here the value of k is 25, which means calculating the distances to the 25 adjacent points;

[0077] S6.2.3. Define the kth distance neighborhood N k (p); that is, the kth distance field N of point p k (p) The set of all points whose distance to point p is less than or equal to the kth distance to point p; the expression for the kth distance domain is:

[0078] N k (p)={o′|d(p,o′)≤d k (p)}

[0079] Where d(p,o′) is the Euclidean distance between point p and point o′, d k (p) the kth distance of point p;

[0080] S6.2.4. Calculate the kth reachable distance d k,reach (p,o); the calculation formula for the kth reachable distance is:

[0081] d k,reach (p,o)=max{d(p,o),d k (o)}

[0082] Where d(p,o) is the Euclidean distance between point p and point o; d k (o) is the kth distance of point o; max means taking the maximum value of the two;

[0083] S6.2.5. Calculate the kth local reachable density lrd of point p k (p); the local reachable density calculation formula is:

[0084]

[0085] In the formula, |N k (p)| is the area N of point p k (p) the number of points; d k,reach (p,o) is the kth reachable distance from point p to point o;

[0086] S6.2.6. Calculate the kth local outlier factor LOF of point p k (p); The calculation formula of the kth local outlier factor is as follows:

[0087]

[0088] Where lrd k(o) The kth local reachable density of point o; lrd k (p) is the kth local reachable density of point p; |N k (p)| is the area N of point p k (p) the number of points;

[0089] S6.2.7. Based on the calculated local outlier factor value, separate the points into outliers and normal points, set the threshold of the local outlier factor to 1, and when the calculated LOF k If (p) is greater than 1, it is considered an outlier;

[0090] The rule for outputting the warning result is that when the predicted value of the single cell maximum voltage is greater than the upper cut-off voltage, an overvoltage alarm is triggered; when it is less than the lower cut-off voltage, an undervoltage alarm is triggered; when the predicted value is within the normal value range but the LOF outlier factor value is greater than 1, an abnormal voltage fluctuation warning is triggered; when the predicted value of the single cell maximum temperature is greater than the upper temperature limit, an abnormal temperature warning is triggered; when the temperature value is within the normal range but the LOF outlier factor value is greater than 1, an abnormal temperature fluctuation warning is triggered;

[0091] The upper cut-off voltage is 4.25V, and the lower cut-off voltage is 3.4V;

[0092] The upper temperature limit is 55°C.

[0093] Beneficial effects of the present invention

[0094] The present invention proposes a fault warning method for automotive lithium batteries based on multivariate prediction and anomaly detection. The experimental results show that based on accurate prediction, accurate multi-step (24 steps, i.e. 4 minutes in advance) prediction of future data of voltage faults and temperature faults is achieved, thereby expanding the warning range from voltage faults to temperature faults that have a greater impact on safety, providing drivers with more sufficient emergency response time. The experimental results show that compared with the existing battery management system (BMS) warning method, the present invention can warn of voltage over-limit 4.5 minutes in advance, realize abnormal voltage rise and fall warning before undervoltage 83.5 minutes in advance, alarm temperature abnormality 24.6 minutes in advance, and realize temperature warning 35.1 minutes before the fault occurs. BRIEF DESCRIPTION OF THE DRAWINGS

[0095] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0096] Figure 1 is a flow chart of the method of the present invention;

[0097] Figure 2 is a schematic diagram of the method of the present invention;

[0098] Figure 3 : is a model structure diagram of the present invention, wherein part (a) is a structural principle diagram of a temporal convolutional network (TCN); part (b) is a residual convolution block diagram of a TCN model; and part (c) is a structural schematic diagram of a long short-term memory neural network (LSTM);

[0099] Figure 4 The predicted result of the maximum probe temperature obtained by the implementation of the present invention and the comparison result with the true value, wherein part (a) represents the true value of the temperature; part (b) represents the predicted value of the temperature; and part (c) represents the error between the true value and the predicted value;

[0100] Figure 5 The predicted result of the maximum voltage of a single cell obtained by the implementation of the present invention and the comparison result with the true value, wherein part (a) represents the true value of the voltage, part (b) represents the predicted value of the voltage, and part (c) represents the error between the true value and the predicted value;

[0101] Figure 6 is the overvoltage warning result obtained by implementing the present invention;

[0102] Figure 7 It is the undervoltage warning result obtained by implementing the present invention;

[0103] Figure 8 It is the temperature warning result obtained by implementing the present invention. DETAILED DESCRIPTION

[0104] The present invention is further described in detail below in conjunction with specific embodiments.

[0105] like Figure 1 , Figure 2 As shown, a vehicle lithium battery fault warning method based on multivariate prediction and anomaly detection includes the following steps:

[0106] S1. Collect the operating data of the vehicle battery system through the big data platform;

[0107] The vehicle battery system includes: a battery system of a normal vehicle and a vehicle that has had a fault alarm;

[0108] The fault types include: voltage fault and temperature fault;

[0109] The voltage faults include: battery cell overvoltage, undervoltage, and rapid voltage change;

[0110] The temperature faults include: temperature rise too fast, temperature exceeding the set threshold;

[0111] Specifically, based on the big data platform, vehicle operation data is collected and the actual operation data of the power battery is extracted; the collected vehicles include: normal vehicles and vehicles that have had fault alarms; the fault types include: voltage faults and temperature faults; voltage faults include: battery cell overvoltage, undervoltage, and voltage changes too fast; temperature faults include: temperature rise too fast, and the temperature exceeds the set threshold (-20℃~55℃);

[0112] The operating data of the vehicle battery system collected include: vehicle ID, sampling time, total voltage of the battery pack (TV), maximum voltage of the single cell (HCV), minimum voltage of the single cell (LCV), SOC, total current (C), maximum temperature of the probe (HT) and minimum temperature of the probe (LT);

[0113] In this embodiment, the real data of the battery systems of three electric vehicles running for half a year is obtained through the big data detection platform; the data collection process strictly follows the "Electric Vehicle Remote Service and Management Technical Specification" protocol, including two normally operating vehicles (car 1 and car 2) and a vehicle with a fault alarm (car 3); car 1 runs in a high temperature environment in summer, car 2 runs in a low temperature environment in winter, and car 3 has a BMS fault alarm. The fault types include voltage fault (single cell overvoltage, undervoltage, voltage change rate is too fast) and temperature fault (temperature rise is too fast, temperature exceeds the set threshold). The data sampling frequency of the three vehicles is once every 10 seconds, the battery charging cut-off voltage (upper cut-off voltage) is 4.25V, the discharge cut-off voltage (lower cut-off voltage) is 3.4V, and the upper and lower temperature limits range from -20°C to 55°C;

[0114] S2, clean the collected data and generate a data set;

[0115] Due to the complex operating conditions, the collected data needs to be cleaned as necessary;

[0116] Data cleaning methods include:

[0117] (1) Linear interpolation is used to fill in missing data of less than three frames, and missing segments of more than three frames are removed; that is, the sampling frequency is once every ten seconds, which is six data per minute. If 1-3 data are missing in this segment, linear interpolation is used to fill the missing data of this segment, and if more than 3 data are missing, this segment is directly deleted;

[0118] (2) Compare the total voltage, total current, maximum and minimum voltage of the single cell, and maximum and minimum temperature of the probe with the data collected within the normal range; the normal range of SOC is [200V, 500V], [-1000A, 1000A], [2V, 5V], [-40℃, 210℃], [0, 100], and data outside the range is considered as abnormal sensor data and should be discarded;

[0119] (3) Use wavelet transform to reduce noise of all data;

[0120] The data set is generated as follows: the cell maximum voltage (HCV) and the probe maximum temperature (HT) are randomly selected as the last two columns, and the battery pack total voltage (TV), total current (C), SOC, cell minimum voltage (LCV), and probe minimum temperature (LT) are not in a fixed order, forming a matrix X with a dimension of [N,7]. The data set is divided by identifying the vehicle ID, and the vehicle ID itself is not used as a data set.

[0121] The expression is as follows:

[0122]

[0123] The maximum voltage of the battery cell is selected as the predicted label value Y1, and the maximum temperature of the probe is selected as the predicted label value Y2; the expression is as follows:

[0124]

[0125] Wherein, N represents the total length of the matrix composed of all data, which is equal to the sum of the number of rows of the training set and the test set matrix; in this embodiment, N is 937433;

[0126] Some of the data recorded in the constructed data set are shown in Table 1:

[0127] Table 1 Part of the data for constructing the dataset

[0128]

[0129]

[0130] S3, divide the data set into training set and test set;

[0131] Specifically, the normal vehicle data is used as the training set to obtain the training matrix X tr , the faulty vehicle is used as the test set to obtain the test matrix X te , the training set and the test set keep the same data form;

[0132] That is, the data matrix consisting of normal vehicles No. 1 and No. 2 is used as the training set, and the data matrix consisting of faulty vehicle No. 3 is used as the test set;

[0133] The expression is as follows:

[0134]

[0135] Wherein, n represents the total number of rows in the training set matrix; in this embodiment, n is 907578;

[0136] S4. Build a prediction model through the training set; the steps are as follows:

[0137] S4.1. Input the training set into the TCN model and perform causal convolution to extract input features. The output formula of causal convolution is:

[0138]

[0139] In the formula, represents the output sequence at time step t and output channel c out The value on c in is the number of input channels, which is equal to the number of input features and has a value of 7; c out is the number of output channels, which is equal to the number of convolution kernels. Its value is 256, which means that the extracted features are converted into 256-dimensional data. z is the size of the convolution kernel. Its value is 3. In time convolution, this value determines how many time steps the convolution kernel covers. i represents the position of the convolution kernel in the time dimension. is the weight of the convolution kernel, indicating that in the output channel c out , input channel c in , the weight parameter at the convolution kernel position i, the initial value is a random value with a mean of 0 and a standard deviation of 0.01 that conforms to the normal distribution. These weights determine how the convolution kernel performs weighted summation of the input features to generate the output; is the input data at time step ti and input channel c in Here ti represents the input value of a lag relative to the current time step t; Represents the bias term of each output channel. The initial value is 0 and there is no fixed value. The bias term will be continuously adjusted during the training process. The bias term is used to add an additional linear offset to the weighted sum result of the convolution to increase the flexibility of the model.

[0140] S4.2, use the output after extracting the input features as the input of LSTM to obtain the hidden state of the current time step t;

[0141] Specifically, the output after extracting the input features is used as the input of LSTM. After the features enter the LSTM, the input gate, forget gate, candidate cell state, cell state update, output gate, hidden state update calculation are performed in sequence. The long-term dependency is further captured through the LSTM model, and the prediction value is finally output through the fully connected layer. The expression is as follows:

[0142]

[0143] In the formula, i trepresents the activation value of the input gate, which is used to control the amount of new information added to the cell state; σ represents the sigmoid activation function, whose output value is between 0 and 1, which is used to control the proportion of input information added; W i represents the weight matrix of the input gate; z t-1 Represents the hidden state of the previous time step, the value is 128; represents the output sequence at time step t and output channel c out The value on is used as the input of the LSTM model; b means concatenating the hidden state of the previous time step with the features extracted by TCN at the current time step; b i represents the bias term of the input gate; f t is the activation value of the forget gate; W f is the weight matrix of the forget gate; b f is the bias term of the forget gate; is the candidate cell state; tanh represents the hyperbolic tangent activation function; W c is the weight matrix used to generate candidate cell states; b c is the bias term of the candidate cell state; c t is the cell state at the current time step; f t ⊙c t-1 is the ratio of the previous cell state f t , indicating the selectively forgotten part; is the candidate state update under the input gate state; ⊙ represents element-wise multiplication; o t is the activation value of the output gate; W o is the weight matrix of the output gate; b o is the bias term of the output gate; z t is the hidden state of LSTM at the current time step t; tanh(c t ) indicates a nonlinear change in the cell state;

[0144] S4.3, LSTM calculates the hidden state of the current time step t and passes it to the fully connected layer, mapping the hidden state to the target prediction space to obtain the predicted value. The expression is as follows:

[0145]

[0146] In the formula, is the predicted result of voltage or temperature at time step t; W y is the weight matrix of the fully connected layer; z t is the hidden state of LSTM at the current time step t; b y is the bias term of the fully connected layer. Since binary data (voltage and temperature) are predicted, the dimension of the bias term is 2, which means that the hidden state is mapped to two output targets.

[0147] S4.4. After the training set is input into the model, all weight matrices and bias items in S4.1-S4.3 will be initialized and assigned. In the forward propagation, the model will use the currently initialized parameter values ​​to calculate the input with the above formula until the final prediction result is output. The prediction value obtained by forward propagation The error between the real label and the real label is calculated through the loss function, and then back propagation is performed to calculate the gradient of the loss function with respect to each parameter. Back propagation will pass back the influence of the loss function on each parameter. Then, the weight and bias parameters obtained in the above steps are updated through the Adam optimizer to reduce the loss function. The present invention uses the mean square error (MSE) as the training set loss function, and the expression is as follows:

[0148]

[0149] In the formula, n represents the length of the training set data, here n value is 907578, Y t represents the true value of voltage or temperature at time step t, is the predicted voltage or temperature value at time t;

[0150] The main function of the time convolution network TCN in the constructed TCN-LSTM hybrid neural network voltage and temperature prediction model is to capture the local time dependency of the input sequence, which is achieved through multi-layer time residual convolution blocks. Each convolution block includes two convolution layers, ReLU activation function and Dropout layer, the size of the convolution kernel is 3, the number of convolution kernels is 256, and the window moving step is 1. In TCN, the output of each layer is used as the input of the next layer, and the features in the time series are extracted layer by layer. Finally, the output shape is converted into a shape and dimension suitable for the LSTM model input through a fully connected layer mapping, and the length and short The long-term memory neural network LSTM is designed to process long-term dependencies in time series data. The LSTM model consists of an input layer, a hidden layer with 128 hidden units, a linear layer, and an output layer. The data after TCN convolution is input into the input layer and hidden layer of LSTM, and finally the final output is obtained through the ReLU activation function and the fully connected layer. The hyperparameters of the model include the number of hidden layers, the number of batches, the number of training iterations, the step size of the sliding time window, the prediction step size, and the learning rate. The corresponding parameters are set to 128, 16, 50, 126, 24, and 0.001 respectively.

[0151] The training model uses the constructed training set. The parameters saved in the model training include: the convolution kernel weights and bias items of the TCN model part, and the weights of the input gate, forget gate, candidate memory unit, output gate, and the bias of the input gate, forget gate, candidate memory unit, and output gate of the LSTM model part;

[0152] Further, such as Figure 3 As shown in the figure, part a shows the structural principle diagram of the temporal convolutional network (TCN). When the dilation rate is 1, the model performs causal convolution without skipping time steps, and when it is greater than 1, dilated convolution is performed; part b is the residual convolution block diagram of the TCN model; part c is a schematic diagram of the structure of the long short-term memory neural network (LSTM);

[0153] In this embodiment, the training set used by the model is only the data of two electric vehicles running for half a year. During the model training process, the amount of training set data and the data operation range can be increased so that the model can fully learn the data characteristics of the vehicle in different environments and driving behavior habits. For different models, the above steps can be repeated to train multiple models;

[0154] Furthermore, the normal vehicle data used to train the model should include as much vehicle data as possible and vehicle data with a wider operating range, so that the model can learn the impact of different driving habits, different operating conditions and external environments on battery data changes, so that the model can obtain more parameters, thereby achieving higher prediction accuracy;

[0155] Furthermore, the training model adopts a “train-test” approach, and model parameters can be updated at any time or regularly by adding training vehicles or improving data quality;

[0156] Furthermore, for higher training accuracy, different vehicle data of different models can be collected for different vehicle models to conduct common feature training of vehicles;

[0157] S5, input the test set into the prediction model, and obtain the prediction result by taking the highest monomer voltage and the highest probe temperature as the prediction label value;

[0158] Specifically, after the test set is input into the prediction model, the model will repeat the formula obtained by inputting the training set into the model to calculate and call the obtained training parameters. The steps to obtain the prediction result with the highest monomer voltage and the highest probe temperature as the prediction label value are as follows:

[0159] S5.1. Input the test set into the TCN model and perform causal convolution to extract input features. The expression of extracted features is as follows:

[0160]

[0161] In the formula, Represents the input sequence X t-τ+1:t Output value after feature extraction; X t-τ+1:t represents the test set input sequence of the past τ time steps; τ is the time window size, and its value is 126;

[0162] S5.2. Pass the features extracted from TCN as input to LSTM to obtain the hidden state z of the time step t ; The expression is as follows:

[0163]

[0164] In the formula, z t-τ+1:t Represents the hidden state sequence obtained after accepting feature input;

[0165] S5.3. Generate the predicted value of the next Δ time steps by inputting the hidden state into the fully connected layer; the expression is as follows:

[0166]

[0167] In the formula, Indicates the maximum voltage of the monomer and the maximum temperature of the probe in the next Δ time steps, where the value of Δ is 24; [HCV t+1:t+Δ ,HT t+1:t+Δ ] is the prediction vector of voltage and temperature for the next Δ time steps; W y is the weight matrix of the fully connected layer; z t is the hidden state of LSTM at the current time step t. Here, only the hidden state obtained at the last time step t is used; b y It is the bias term of the fully connected layer. The bias term will be continuously adjusted during the training process. The bias term is used to add an additional linear offset to the weighted sum result of the convolution to increase the flexibility of the model.

[0168] Specifically, when the test set enters the TCN-LSTM model, the data is divided into M samples, each of which contains a sliding window of length 126 and a label output of prediction length 24. As sample data 1 enters the model, the trained model parameters are called to obtain the highest single cell voltage label data. Then sample data 2 enters the model and repeats the above process to obtain label data And so on, for the last sample Get label data M is the number of samples, and its value is Nn-126-24+1=29766; the specific prediction process is as follows:

[0169]

[0170] Similarly, the predicted data of the probe's maximum temperature can be obtained according to the above method. like The prediction process is:

[0171]

[0172] In the formula, X te represents the test set matrix; Indicates the first sample of the test set entering the TCN-LSTM model, and so on Indicates the Mth sample of the test set entering the model, which is also the last sample; n indicates the data length of the training set, which is equal to the number of matrix rows; N indicates the total length of the training set and test set data; Y1 (1) Indicates the highest single cell voltage prediction data vector obtained by the first sample, and so on. (M) represents the highest monomer voltage prediction data vector obtained by the Mth sample;

[0173] In order to better quantify the difference between the predicted value and the true value obtained by the test set, the root mean square error (RMSE) is used as the evaluation indicator for the prediction performance of the model test set. The calculation formula of RMSE is as follows:

[0174]

[0175] In the formula, Nn represents the length of the test set data, Y t Indicates the true value of voltage or temperature, is the predicted voltage or temperature value at time step t;

[0176] The results of the temperature prediction of the test set are as follows Figure 4 As shown, part a shows the true value of temperature, part b is the predicted value of temperature, and part c is the error between the true value and the predicted value (error = true value - predicted value). The RMSE of temperature prediction is 0.4616℃;

[0177] The voltage prediction results are as follows: Figure 5 As shown, part a shows the true value of the voltage, part b is the predicted value of the voltage, and part c is the error between the true value and the predicted value. The RMSE of the voltage prediction is 0.0117V, which proves that the prediction method proposed in the present invention still has high accuracy when maintaining 24-step forward prediction;

[0178] S6, output warning results;

[0179] Specifically, when the predicted voltage or temperature value exceeds the set upper and lower limits, an overvoltage, undervoltage or temperature abnormality alarm is triggered; when the predicted value does not exceed the set upper and lower limits, but the predicted value is detected by the local anomaly detection algorithm (LOF) and the LOF outlier factor value exceeds 1, an abnormal voltage rise or fall alarm or abnormal temperature rise or fall alarm is triggered;

[0180] The details are as follows:

[0181] S6.1, standardize the prediction results;

[0182] Specifically, after obtaining the predicted values ​​of voltage data and temperature data, the input data needs to be standardized first to ensure that the data is compared on the same scale; the standardization formula is:

[0183]

[0184] In the formula, is the predicted voltage or temperature value obtained at time t; μ j is the mean value of voltage or temperature characteristic j; σ j is the standard deviation of the voltage or temperature characteristic j; z j is the standardized voltage or temperature value;

[0185] S6.2. The standardized voltage data or temperature data are combined into data sets D and U respectively, and the local outlier factor of each point in the data sets D and U is calculated to obtain the warning result; the details are as follows:

[0186] S6.2.1. Define the Euclidean distance; that is, take any data point p in the set and any point o in D or U except p, and define the Euclidean distance between point p and point o as d (p,o) , for any two points p=(p1,p2,...,p m ) and o=(o1,o2,...,o m )The Euclidean distance formula in m-dimensional space is:

[0187]

[0188] S6.2.2. Define the kth distance; that is, define the distance between point p and its nearest k points as d 1～k (p), d 1～k The maximum value in (p) is defined as d k (p), d k (p) is defined as the kth distance of point p; d k (p) must satisfy the following two conditions: (1) In addition to point p, there are at least k points o' in D or U such that d k (p,o′)≤d k (p,o); (2) In D or U, there are at most k-1 points o' other than point k such that d(p,o') < d(p,o); here the value of k is 25, which means calculating the distances to the 25 adjacent points;

[0189] S6.2.3. Define the kth distance neighborhood N k (p); i.e. the kth distance field N of point p k (p) The set of all points whose distance to point p is less than or equal to the kth distance to point p; the expression for the kth distance domain is:

[0190] N k (p)={o′|d(p,o′)≤d k (p)}

[0191] Where d(p,o′) is the Euclidean distance between point p and point o′, d k (p) the kth distance of point p;

[0192] S6.2.4. Calculate the kth reachable distance d k,reach (p,o); the calculation formula for the kth reachable distance is:

[0193] d k,reach (p,o)=max{d(p,o),d k (o)}

[0194] Where d(p,o) is the Euclidean distance between point p and point o; d k (o) is the kth distance of point o; max means taking the maximum value of the two;

[0195] S6.2.5. Calculate the kth local reachability density lrd of point p k (p); the local reachable density calculation formula is:

[0196]

[0197] In the formula, |N k (p)| is the area N of point p k (p) the number of points; d k,reach (p,o) is the kth reachable distance from point p to point o;

[0198] S6.2.6. Calculate the kth local outlier factor LOF of point p k (p); The calculation formula of the kth local outlier factor is as follows:

[0199]

[0200] Where lrd k (o) The kth local reachable density of point o; lrd k (p) is the kth local reachable density of point p; |N k (p)| is the area N of point p k (p) the number of points;

[0201] S6.2.7. Based on the calculated local outlier factor value, separate the points into outliers and normal points, set the threshold of the local outlier factor to 1, and when the calculated LOF k If (p) is greater than 1, it is considered an outlier;

[0202] The voltage fault warning result obtained in this embodiment is as follows: Figure 6 and Figure 7 As shown, from Figure 6 It can be seen that the terminal data of the voltage prediction value (frames 29400 to 29915) is 4.26V at frame 29444, which has exceeded the upper limit cut-off voltage of the battery 4.25V. According to the output rule of the warning result in step 6, the overvoltage warning is triggered at this time, and the BMS alarm point of the vehicle appears at frame 29447, which proves that compared with the original BMS detection method of the vehicle, the warning method proposed by the present invention realizes the overvoltage alarm 4.5 minutes earlier [(29447-29444+24)×10÷60]; Figure 7 It can be seen that the voltage prediction value is 3.6V at frame 257, which is not lower than the lower cut-off voltage. However, the calculated LOF outlier factor value is 1.35, which exceeds the set threshold and triggers the voltage abnormality alarm. The BMS triggers the undervoltage alarm at frame 734, proving that this method realizes the abnormal trend warning of the voltage before the undervoltage 83.5 minutes in advance [(734-257-24)×10÷60].

[0203] The temperature fault warning result obtained in this embodiment is as follows: Figure 8 As shown in the figure, it can be seen that the temperature anomaly was first detected at frame 29728 and the abnormal point was marked. At this time, the LOF calculation value of the point was 752.34, which exceeded the set LOF anomaly detection threshold, while the temperature prediction value was 26.7℃, which did not exceed the upper temperature limit. After that, the battery temperature rose significantly and rapidly. However, compared with the voltage data at this time, it can be seen that the actual voltage value is within the normal range, while the predicted voltage value only showed an abnormal upward trend, but the value is still within the normal range. Therefore, focusing only on voltage data is limited for fault warning;

[0204] From the first time the temperature detected an abnormal value (29728 frames) to the data termination point 29915 frames, the multi-step prediction model based on TCN-LSTM and the LOF local anomaly detection algorithm achieved a temperature warning time of 35.1 minutes [(29915-29728+24)×10÷60], which was 24.6 minutes [(29852-29728+24)×10÷60] earlier than the BMS warning, providing drivers with sufficient time to respond to emergencies.

Claims

1. A vehicle lithium battery fault warning method based on multivariate prediction and anomaly detection, characterized by: The following steps are involved: S1. Collect the operating data of the vehicle battery system through the big data platform; The vehicle battery system includes: a battery system of a normal vehicle and a vehicle that has had a fault alarm; The fault types include: voltage fault and temperature fault; The voltage faults include: battery cell overvoltage, undervoltage, and rapid voltage change; The temperature faults include: temperature rise too fast, temperature exceeding the set threshold; The operating data of the battery system includes: vehicle ID, sampling time, total voltage of the battery pack, maximum voltage of a single cell, minimum voltage of a single cell, SOC, total current, maximum temperature of the probe, and minimum temperature of the probe; S2, clean the collected data and generate a data set; S3, divide the data set into training set and test set; S4, build a prediction model through the training set; S5, input the test set into the prediction model, and obtain the prediction result by taking the highest monomer voltage and the highest probe temperature as the prediction label value; S6. Output the warning result.

2. The method for early warning of automotive lithium battery failure based on multivariate prediction and anomaly detection according to claim 1, characterized in that: The data cleaning of the collected data includes: (1) Linear interpolation is used to fill in the missing data of missing frames less than three frames, and the missing fragments of more than three frames are eliminated; (2) Compare the total voltage, total current, maximum and minimum voltage of the single cell, and maximum and minimum temperature of the probe with the data collected within the normal range; the normal range of SOC is [200V, 500V], [-1000A, 1000A], [2V, 5V], [-40℃, 210℃], [0, 100], and data outside the range is considered as abnormal sensor data and should be discarded; (3) Use wavelet transform to reduce noise on all data.

3. The method for early warning of vehicle lithium battery failure based on multivariate prediction and anomaly detection according to claim 1, characterized in that: The data set is generated specifically as follows: the total voltage of the battery pack, the maximum voltage of a single cell, the minimum voltage of a single cell, the SOC, the total current, the maximum temperature of the probe and the minimum temperature of the probe are formed into a matrix with a dimension of [N,7]; wherein the maximum voltage of the single cell and the maximum temperature of the probe are taken as the last two columns.

4. The method for early warning of vehicle lithium battery failure based on multivariate prediction and anomaly detection according to claim 1, characterized in that: The data set is divided in such a way that normal vehicle data is used as a training set to obtain a training matrix, and faulty vehicle data is used as a test set to obtain a test matrix.

5. The method for early warning of vehicle lithium battery failure based on multivariate prediction and anomaly detection according to claim 1, characterized in that: The steps of constructing a prediction model through a training set are as follows: S4.

1. Input the training set into the TCN model and perform causal convolution to extract input features. The output formula of causal convolution is: In the formula, represents the output sequence at time step t and output channel c out The value on c in is the number of input channels, which is equal to the number of input features; c out is the number of output channels, which is equal to the number of convolution kernels; z is the size of the convolution kernel; i represents the position of the convolution kernel in the time dimension; is the weight of the convolution kernel, indicating that in the output channel c out , input channel c in , the weight parameter at the convolution kernel position i; is the input data at time step td·i and input channel c in The value on d represents the dilation rate. When d is 1, causal convolution is performed, and when d is not 1, dilated convolution is performed. Represents the bias term of each output channel; S4.

2. The output after extracting the input features is used as the input of LSTM. After the features enter the LSTM, the input gate, forget gate, candidate cell state, cell state update, output gate, hidden state update calculation are performed in sequence. The LSTM model is used to further capture long-term dependencies, and finally the prediction value is output through the fully connected layer. The calculation formula of LSTM is as follows: In the formula, i t represents the activation value of the input gate; σ represents the sigmoid activation function; W i represents the weight matrix of the input gate; z t-1 represents the hidden state of the previous time step; represents the output sequence at time step t and output channel c out The value on is used as the input of the LSTM model; b means concatenating the hidden state of the previous time step with the features extracted by TCN at the current time step; b i represents the bias term of the input gate; f t is the activation value of the forget gate; W f is the weight matrix of the forget gate; b f is the bias term of the forget gate; is the candidate cell state; tanh represents the hyperbolic tangent activation function; W c is the weight matrix used to generate candidate cell states; b c is the bias term of the candidate cell state; c t is the cell state at the current time step; f t ⊙c t-1 is the ratio of the previous cell state f t , indicating the selectively forgotten part; is the candidate state update under the input gate state; ⊙ represents element-wise multiplication; o t is the activation value of the output gate; W o is the weight matrix of the output gate; b o is the bias term of the output gate; z t is the hidden state of LSTM at the current time step t; tanh(c t ) indicates a nonlinear change in the cell state; S4.3, LSTM calculates the hidden state z of the current time step t t It is then passed to the fully connected layer to map the hidden state to the target prediction space to obtain the predicted value, which is expressed as follows: In the formula, is the predicted result of voltage or temperature at time step t; W y is the weight matrix of the fully connected layer; z t is the hidden state of LSTM at the current time step t; b y is the bias term of the fully connected layer; S4.

4. After the training set is input into the model, all weight matrices and bias items in the previous steps will be initialized and assigned values. In the forward propagation, the model will use the currently initialized parameter values ​​to calculate the above formula for the input until the final prediction result is output. The prediction value obtained by forward propagation The error between the real label and the real label is calculated through the loss function, and then back propagation is performed to calculate the gradient of the loss function with respect to each parameter. Back propagation will pass back the influence of the loss function on each parameter. Then, the weight and bias parameters obtained in the above steps are updated through the Adam optimizer to reduce the loss function. The present invention uses the mean square error (MSE) as the training set loss function, and the expression is as follows: In the formula, n represents the length of the training set data, Y t represents the true value of voltage or temperature at time step t, is the predicted voltage or temperature value at time t.

6. The method for early warning of vehicle lithium battery failure based on multivariate prediction and anomaly detection according to claim 1, characterized in that: The steps of inputting the test set into the prediction model and obtaining the prediction result by taking the highest monomer voltage and the highest probe temperature as the prediction label value are as follows: S5.

1. Input the test set into the TCN model and perform causal convolution to extract input features. The expression of extracted features is as follows: In the formula, Represents the input sequence X t-τ+1:t Output value after feature extraction; X t-τ+1:t represents the test set input sequence for the past τ time steps; τ is the time window size; S5.

2. Pass the features extracted from TCN as input to LSTM to obtain the hidden state z of the time step t ; The expression is as follows: In the formula, z t-τ+1:t Represents the hidden state sequence obtained after accepting feature input; S5.

3. Generate the predicted value of the next Δ time steps by inputting the hidden state into the fully connected layer; the expression is as follows: In the formula, Indicates the maximum voltage of the monomer and the maximum temperature of the probe in the future Δ time steps; [HCV t+1:t+Δ ,HT t+1:t+Δ ] is the prediction vector of voltage and temperature for the next Δ time steps; W y is the weight matrix of the fully connected layer; z t is the hidden state of LSTM at the current time step t. Here, only the hidden state obtained at the last time step t is used; b y is the bias term of the fully connected layer.

7. The method for early warning of vehicle lithium battery failure based on multivariate prediction and anomaly detection according to claim 1, characterized in that: The steps of outputting the warning result are as follows: S6.1, standardize the prediction results; S6.

2. The standardized voltage data or temperature data are combined into data sets D and U respectively, and the local outlier factor of each point in the data sets D and U is calculated to obtain the warning result; the details are as follows: S6.2.

1. Define the Euclidean distance: Take any data point p in the set and any point o in D or U except p, and define the Euclidean distance between point p and point o as d (p,o) ; S6.2.

2. Define the kth distance: The distance between point p and its nearest k points is defined as d 1～k (p), d 1～k The maximum value in (p) is defined as d k (p), d k (p) is defined as the kth distance of point p; the value of k is 25, which means calculating the distances of the 25 adjacent points; d k (p) must satisfy the following two conditions: (1) In addition to point p, there are at least k points o' in D or U such that d k (p,o′)≤d k (p,o); (2) In D or U, except for the kth point, there are at most k-1 points o' such that d(p,o')<d(p,o); S6.2.

3. Define the kth distance neighborhood N k (p); the kth distance field N of point p k (p) The set of all points whose distance to point p is less than or equal to the kth distance to point p; the expression for the kth distance domain is: N k (p)={o′|d(p,o′)≤d k (p)} Where d(p,o′) is the Euclidean distance between point p and point o′, d k (p) the kth distance of point p; S6.2.

4. Calculate the kth reachable distance d k,reach (p,o); the calculation formula for the kth reachable distance is: d k,reach (p,o)=max{d(p,o),d k (o)} Where d(p,o) is the Euclidean distance between point p and point o; d k (o) is the kth distance of point o; max means taking the maximum value of the two; S6.2.

5. Calculate the kth local reachability density lrd of point p k (p); the local reachable density calculation formula is: In the formula, |N k (p)| is the area N of point p k (p) the number of points; d k,reach (p,o) is the kth reachable distance from point p to point o; S6.2.

6. Calculate the kth local outlier factor LOF of point p k (p); The calculation formula of the kth local outlier factor is as follows: Where lrd k (o) The kth local reachable density of point o; lrd k (p) is the kth local reachable density of point p; |N k (p)| is the area N of point p k (p) the number of points; S6.2.

7. Based on the calculated local outlier factor value, separate the points into outliers and normal points, set the threshold of the local outlier factor to 1, and when the calculated LOF k If (p) is greater than 1, it is considered an outlier; The rule for outputting the warning result is that when the predicted value of the single cell maximum voltage is greater than the upper cut-off voltage, an overvoltage alarm is triggered, and when it is less than the lower cut-off voltage, an undervoltage alarm is triggered; When the predicted value is within the normal range but the outlier factor value is greater than 1, an abnormal voltage fluctuation warning is triggered; When the probe's maximum temperature prediction value is greater than the upper temperature limit, a temperature abnormality warning is triggered; When the temperature value is within the normal range but the LOF outlier factor value is greater than 1, an abnormal temperature fluctuation warning is triggered; The upper cut-off voltage is 4.25V, and the lower cut-off voltage is 3.4V; The upper temperature limit is 55°C.

Images