Anomaly detection method and device based on real-time operation data of new energy vehicles

Abstract

The embodiment of the present invention provides an abnormality detection method and device based on the real-time operation data of new energy vehicles, including the following steps: in accordance with the requirements of GB / T 32960, based on the real-time operation data of the power battery system of the new energy vehicle and the historical health data of a small sample; combining the monitoring parameter data of the power battery system with the multivariate state estimation technology to obtain the abnormality detection model and the residuals of each monitoring parameter in the abnormality detection model, the residuals include healthy residuals and actual residuals; performing a sequential probability ratio test on the residuals, and obtaining the abnormality detection result of the power battery system of the new energy vehicle based on the index of the sequential probability ratio test. It can accurately detect the abnormal state of the power battery system when it is degraded. The training process is efficient and time-saving.

Description

Technical Field

[0001] The present invention relates to the technical field of abnormality detection of power batteries for new energy vehicles, and in particular to an abnormality detection method and device based on real-time operation data of new energy vehicles. Background Art

[0002] The operational safety of power batteries for new energy vehicles has always been one of the core issues that plague the development of new energy vehicles. The simple safety measures (sensor monitoring, contactors, insulation layers, voltage, current and temperature limit operations) added to the traditional BMS for the safety of power batteries are no longer sufficient to protect power batteries as new energy vehicles become increasingly intelligent and integrated, and the requirements for driving range are getting higher and higher. The hardware and software implementation of BMS has also become increasingly difficult. As a strong support for BMS, the fault diagnosis algorithm is mainly responsible for effectively and early detection of faults. At present, although the research on fault diagnosis algorithms has made certain progress in improving safety, there are certain limitations in operational safety, especially based on the real-time operating data of new energy vehicles (collected in accordance with the requirements of GB / T 32960-2016). The research on fault diagnosis algorithms for the operational safety of power battery systems for new energy vehicles is still in its infancy and lacks systematic research results.

[0003] The applicant has discovered that there are at least the following problems in the prior art: it is impossible to accurately and efficiently detect abnormalities in new energy vehicles. Summary of the invention

[0004] The technical problem solved by the embodiments of the present invention is the problem that anomaly detection of new energy vehicles cannot be achieved accurately and efficiently.

[0005] To achieve the above objectives, on the one hand, an embodiment of the present invention provides an anomaly detection method based on real-time operation data of new energy vehicles, comprising the following steps:

[0006] Obtain the real-time operation data and historical health data of the monitoring parameters of the new energy vehicle power battery system, and obtain the real-time operation data and historical health data of the key monitoring parameters based on failure analysis;

[0007] The real-time operation data of the key monitoring parameters and the historical health data of the key monitoring parameters are combined with the multivariate state estimation technology to obtain the anomaly detection model and the residuals of each monitoring parameter in the anomaly detection model, wherein the residuals include healthy residuals and actual residuals;

[0008] The residual is subjected to a sequential probability ratio test, and based on the index of the sequential probability ratio test, an abnormality detection result of the new energy vehicle power battery system is obtained.

[0009] On the other hand, an embodiment of the present invention provides an abnormality detection device based on real-time operation data of new energy vehicles, comprising:

[0010] An acquisition unit, used to acquire real-time operation data of monitoring parameters and historical health data of monitoring parameters of a new energy vehicle power battery system, and obtain real-time operation data of key monitoring parameters and historical health data of key monitoring parameters based on failure analysis;

[0011] A calculation unit, used to combine the real-time operation data of the key monitoring parameters and the historical health data of the key monitoring parameters with the multivariate state estimation technology to obtain the anomaly detection model and the residuals of each monitoring parameter in the anomaly detection model, wherein the residuals include healthy residuals and actual residuals;

[0012] The detection unit is used to perform a sequential probability ratio test on the residual, and obtain an abnormality detection result of the new energy vehicle power battery system based on an index of the sequential probability ratio test.

[0013] The above technical solution has the following beneficial effects: This application selects an accurate but not overly complex set of power battery monitoring parameters by studying and analyzing the failure mechanism of internal and external faults of power batteries, and proposes an efficient anomaly detection method based on real-time operation data to balance the effectiveness of BMS fault diagnosis and the completeness of historical data, which can accurately detect the abnormal state of the power battery system when it is degraded. The training process is efficient and time-saving. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, 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 work.

[0015] Figure 1 is a flow chart of an abnormality detection method based on real-time operation data of new energy vehicles provided by an embodiment of the present invention;

[0016] Figure 2 It is a structural schematic diagram of an abnormality detection device based on real-time operation data of new energy vehicles provided by an embodiment of the present invention;

[0017] Figure 3 This is a flow chart of a first implementation method of an anomaly detection method based on real-time operation data of new energy vehicles provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0019] The present invention provides an abnormality detection method based on real-time operation data of new energy vehicles, such as Figure 1 As shown, the following steps are included:

[0020] S101: Acquire real-time operation data of monitoring parameters and historical health data of monitoring parameters of a new energy vehicle power battery system, and obtain real-time operation data of key monitoring parameters and historical health data of key monitoring parameters based on failure analysis;

[0021] S102: combining the real-time operation data of the key monitoring parameters and the historical health data of the key monitoring parameters with the multivariate state estimation technology to obtain an anomaly detection model and the residuals of each monitoring parameter in the anomaly detection model, wherein the residuals include healthy residuals and actual residuals;

[0022] S103: performing a sequential probability ratio test on the residual, and obtaining an abnormality detection result of the new energy vehicle power battery system based on an index of the sequential probability ratio test.

[0023] like Figure 3 As shown, the real-time operation data of key monitoring parameters and the historical health data of key monitoring parameters are combined with multivariate state estimation technology to obtain the anomaly detection model and the residuals of each monitoring parameter in the anomaly detection model. The residuals include healthy residuals and actual residuals, specifically including:

[0024] Establish an observation matrix based on real-time operating data of key monitoring parameters;

[0025] Calculate the estimation matrix of the observation matrix;

[0026] The actual residual is calculated using the estimated matrix.

[0027] The real-time operation data of the key monitoring parameters and the historical health data of the key monitoring parameters are combined with the multivariate state estimation technology to obtain the abnormal detection model and the residuals of each monitoring parameter in the abnormal detection model, wherein the residuals include healthy residuals and actual residuals, and also include:

[0028] forming training data based on historical health data of the key monitoring parameters;

[0029] Establishing a memory matrix based on the training data and forming remaining training data;

[0030] Calculate an estimation matrix of the remaining training data according to the memory matrix and the remaining training data;

[0031] The health residuals are calculated using the estimated matrix.

[0032] in:

[0033] Selection of key monitoring parameter data:

[0034] The first step of MSET-based power battery anomaly detection is to select the key monitoring parameters of the power battery. By analyzing the working principle of the power battery, the key monitoring parameters of the power battery are selected, and based on the expected failure relationship model, the relevant data parameters of the power battery system of SAIC-GM-Wuling new energy vehicles are selected.

[0035] Obtaining the observation matrix:

[0036] The observation matrix composed of the key monitoring parameter data of the power battery is collected by the data acquisition system and is denoted as X obs .

[0037] In the observation matrix X obs In the above example, the horizontal axis represents the time state and the vertical axis represents the monitoring parameters. obs It is an n×m matrix, where n represents the number of monitoring parameters and m represents the number of time states.

[0038] The observation matrix X obs :

[0039]

[0040] x ji It means that j monitoring parameters are in time state t i The observed value at . Monitoring parameter x j The sequence of observations at each time state can be represented as a vector X j ,

[0041] X j =[x j1 ,x j2 ,…,x jm ].

[0042] Similarly, at a certain time state t i So the vector X(t i ) can be expressed as shown below,

[0043] X(t i )=[x 1i ,x 2i ,…,x ni ]T .

[0044] Selection of training data:

[0045] The data matrix composed of various monitoring parameters of the power battery under normal operating conditions is called training data, which is represented by the capital letter T. The so-called normal operating state refers to the operating state of the power battery in an environment with low failure rate. For most electronic products or systems, the failure rate follows the bathtub curve, that is, in the absence of external interference, the product or system is in a high failure rate period in the early and late stages of operation, while the mid-stage of operation is a relatively low failure rate period. Therefore, this article, based on the operating characteristics of the power battery for actual engineering applications, considers the mid-stage of the full life of the power battery as a normal operating state. In other words, the normal operating state of the power battery refers to its operating state in a low failure rate period. Of course, from the perspective of monitoring parameter data, the normal operating state of the power battery refers to the stable change state of each monitoring parameter, and the stable change state also includes the normal degradation state of each monitoring parameter.

[0046] The training data T can also form a similar data matrix. Assuming that the training data T contains a total of k time states, the training data T can be expressed as:

[0047]

[0048] Creation of memory matrix:

[0049] From the training data T, according to the corresponding selection steps, a data matrix is ​​created, which is called the memory matrix and is represented by the capital letter D.

[0050] Assume that the selected observation matrix X obs , the number of time states contained in the memory matrix D is k1, then the memory matrix D can be expressed as an n×k1 data matrix, where n represents the number of monitoring parameters and k1 represents the number of time states contained in the memory matrix D. Then the memory matrix D can be expressed as:

[0051]

[0052] Steps for selecting the memory matrix D:

[0053] Step 1: According to the number of monitoring parameters n and the actual application, the number of time states to be selected from the training data T is determined subjectively, that is, the number of time states k1 contained in the memory matrix D;

[0054] Step 2: Select the time states when each monitoring parameter is at its extreme value from the training data T. Generally, the number of time states k1 contained in the memory matrix D is greater than or equal to the number of time states when the monitoring parameter is at its extreme value, and cannot be less than the number of time states when the monitoring parameter is at its extreme value. If the number of time states when the monitoring parameter is at its extreme value is equal to k1, then the creation of the memory matrix D is completed; otherwise, proceed to Step 3;

[0055] Step 3 If the number of time states at the extreme value of the selected monitoring parameter is less than k1, continue to select according to the following rules:

[0056] Calculate the Euclidean norm of the training data T at this time, recorded as vector T norm ;

[0057] The vector T norm The vector obtained after arranging in descending or ascending order is denoted as vector T sort ;

[0058] According to the method of equidistant sampling, from vector T sort In , the number of time states of the remaining memory matrix D is selected.

[0059] Regarding the formation of the remaining training data:

[0060] From the training data T, the time states contained in the memory matrix D are removed, and the remaining time states form the remaining training data, which are represented by the capital letter L.

[0061] Assume that the number of time states contained in the remaining training data L is k2. Then, combined with the training data T, the memory matrix D and the remaining training data L have the following equation relationship:

[0062]

[0063] In the formula, k=k1+k2.

[0064] About calculating the estimation matrix of the observation matrix:

[0065] According to the observation matrix X obs Calculate X obs The estimation matrix is ​​the observation matrix X obs The estimated matrix is ​​denoted as X est .

[0066] X est =D·W,

[0067] Where W is the weight vector, which represents a similarity measure between the current state estimate and the memory matrix D. The weight vector W can be obtained by the following equation:

[0068] minε=min(|X obs -X est |)=min(|X obs -D·W|),

[0069] Right now

[0070] minε 2 =min[(X obs -D.W) T ·(X obs -D·W)]

[0071] When ε is under the constraint of minimization, let When , the minimum square error estimation solution of the weight vector W can be expressed as:

[0072] W=(D T ·D) -1 ·(D T ·X obs )

[0073] So, we can get X est for:

[0074] X est =D·(D T ·D) -1 ·(D T ·X obs )

[0075] Compute the estimation matrix for the remaining training data:

[0076] According to the definition of the remaining training data L, the estimated matrix of the remaining training data L is calculated, that is, the estimated matrix of the remaining training data L, denoted as L est .

[0077] L est =D·W′,

[0078] Where W′ is the weight vector, which also represents a similarity measure between the current state estimate and the memory matrix. So in the same way, the weight vector W′ can be obtained by the following equation:

[0079] minε=min(|LL est |) = min(|LD·W′|),

[0080] Right now

[0081] minε 2 =min[(LD·W′) T ·(LD·W′)]

[0082] Similarly, when ε is under the constraint of minimization, that is, The minimum square error estimation solution of the weight vector W′ can be expressed as:

[0083] W′=(D T ·D) -1 ·(D T ·L)

[0084] So by the same token, we can get L est for:

[0085] L est =D·(D T ·D) -1 ·(D T ·L)

[0086] The residual is subjected to a sequential probability ratio test to obtain an abnormality detection result of the new energy vehicle power battery system, specifically including:

[0087] If Λ(Y n )≥B, then accept hypothesis H1, H1 is the probability density function 1;

[0088] If Λ(Y n )≤A, then accept hypothesis H0, H0 is the probability density function 2;

[0089] If A<Λ(Y n )<B, continue sampling, and then repeat the comparison;

[0090] Among them, Λ(Y n ) represents the likelihood ratio function, A and B are constants, and 0<A<1<B<∞, Y n are n independent and identically distributed samples.

[0091] Calculate the actual residual, including:

[0092] R X =X est -X obs ;

[0093] Among them, X obs is the observation matrix, X est is the estimation matrix, R X is the actual residual.

[0094] R L =L est -L;

[0095] L is the estimation matrix of the remaining training data; L est is the estimated matrix of the remaining training data, R L is the health residual.

[0096] Sequential rank and probability ratio test study:

[0097] SRPRT has weaker requirements on sample data because it requires the potential distribution of sample data to be symmetric about its median or mean. It is precisely because of this feature that the SRPRT method is easier to satisfy in practical applications and is more reasonable than other assumptions based on specific distributions. The SRPRT method is based on Wilcoxon signed rank statistics. To calculate the likelihood function ratio in SPRT, SRPRT uses the Wilcoxon signed rank statistic W under the assumption H i The probability of the following is P(W i ; m) to replace the likelihood function L(Y n ;H i ), so the corresponding likelihood function ratio of SRPRT is:

[0098]

[0099] The key to SRPRT is to use the Wilcoxon signed rank sum statistic of the observed sample instead of the conventional statistic based on a specific distribution in the SPRT method. However, using the SRPRT method for testing, only the mean test can be performed.

[0100] Assume that the original residual sample is X = [x1, x2, ..., x n ], then the steps for calculating the signed rank and statistic and its probability distribution are as follows:

[0101] Step 1 Assume that the hypothesis testing problem is:

[0102] H0:μ=μ0,H1:μ=μ1.

[0103] Step 2: Assume H i Next, each original residual sample x j Subtract the assumed mean μ i , at this time, we get the new residual sample sequence x 1,i ,x 2,i ,…,x n,i ,that is

[0104] x j,i =x j -μ i ,

[0105] Where j (j=1,2,…,n) is the sample number and i (i=0,1) is the hypothesis number.

[0106] Step 3: The obtained new residual sample sequence x 1,i ,x 2,i ,…,x n,i Rearrange them according to their absolute values ​​so that

[0107]

[0108] Step 4: Remove the rearranged samples with an absolute value of zero. Assume that after removing the rearranged samples with an absolute value of zero, there are m rearranged samples left. Then, assign a rank value R to each rearranged sample according to its position in the entire sequence. For samples with the same absolute value in the rearranged sequence, assign their rank value to the average of their rank values ​​(for example, assign the same average rank value j+1 to the three rearranged samples with the same absolute value and arrangement numbers j, j+1, and j+2); assign a sign indicator function ψ according to the positive or negative value of the rearranged sample value. Corresponding rearranged samples The rank value R and indicator function ψ are shown in the following two equations respectively.

[0109]

[0110]

[0111] In the formula, k = 1, 2,…, n.

[0112] Step 5: Get hypothesis H i The Wilcoxon signed rank statistic W under i + and W i - As shown in the following formulas, they are discrete random variables that obey a certain distribution function.

[0113]

[0114] W i - =m(m+1) / 2-W i + .

[0115] For a two-sided hypothesis test, the test statistic W should be taken i =min(W i + ,W i - ). When the sample size is small (m≤20), P(W i , m) can be obtained by looking up the table; however, for large sample sizes (m>20), when there is no sample group with the same absolute value in the sample sequence, the statistic W i The approximate mean is T = m(m+1) / 4, and the variance is V T =m(m+1)(2m+1) / 24 normal distribution, so we have,

[0116]

[0117] Accordingly, it is not difficult to obtain P(W i ,m) is calculated as follows:

[0118]

[0119] However, when there are multiple non-zero rearranged samples (called sample groups) with the same absolute value in the sample sequence, the variance V T It needs to be corrected. Let t i is the number of samples in the i-th sample group with the same absolute value in the sequence, and g is the number of sample groups with the same absolute value in the sequence, then the corrected variance V T It can be expressed as follows:

[0120]

[0121] In summary, when SRPRT is used, the corresponding likelihood ratio function is

[0122]

[0123] Step 6: Result verification. As with the SPRT method, after obtaining the likelihood ratio function, constants A and B are introduced, 0<A<1<B<∞. Then, for which hypothesis H0 or H1 is accepted, by comparing Λ(Y n ) with the upper limit B and the lower limit A. The specific decision plan is as follows:

[0124] (1) If Λ(Y n )≥B, then accept hypothesis H1;

[0125] (2) If Λ(Y n )≤A, then accept hypothesis H0;

[0126] (3) If A<Λ(Y n )<B, then no decision can be made, continue sampling, and then repeat the comparison.

[0127] in,

[0128]

[0129] Where α and β are the probabilities of making the first and second type errors in the test, respectively.

[0130] The present invention also provides an abnormality detection device based on real-time operation data of new energy vehicles, such as Figure 2 As shown, including:

[0131] An acquisition unit 21 is used to acquire real-time operation data of monitoring parameters and historical health data of monitoring parameters of a new energy vehicle power battery system, and obtain real-time operation data of key monitoring parameters and historical health data of key monitoring parameters based on failure analysis;

[0132] A calculation unit 22, used to combine the real-time operation data of the key monitoring parameters and the historical health data of the key monitoring parameters with the multivariate state estimation technology to obtain an anomaly detection model and the residuals of each monitoring parameter in the anomaly detection model, wherein the residuals include healthy residuals and actual residuals;

[0133] The detection unit 23 is used to perform a sequential probability ratio test on the residual, and obtain an abnormality detection result of the new energy vehicle power battery system based on an index of the sequential probability ratio test.

[0134] The computing unit specifically includes:

[0135] Establish an observation matrix based on real-time operating data of key monitoring parameters;

[0136] Calculate the estimation matrix of the observation matrix;

[0137] The actual residual is calculated using the estimated matrix.

[0138] The computing unit further includes:

[0139] forming training data based on historical health data of the key monitoring parameters;

[0140] Establishing a memory matrix based on the training data and forming remaining training data;

[0141] Calculate an estimation matrix of the remaining training data according to the memory matrix and the remaining training data;

[0142] The health residuals are calculated using the estimated matrix.

[0143] The detection unit specifically comprises:

[0144] If Λ(Y n )≥B, then accept hypothesis H1, H1 is the probability density function 1;

[0145] If Λ(Y n )≤A, then accept hypothesis H0, H0 is the probability density function 2;

[0146] If A<Λ(Y n )<B, continue sampling, and then repeat the comparison;

[0147] Among them, Λ(Y n ) represents the likelihood ratio function, A and B are constants, and 0<A<1<B<∞, Yn are independent and identically distributed samples.

[0148] Calculate the residuals, including:

[0149] R X =X est -X obs ;

[0150] Among them, X obs is the observation matrix, X est is the estimation matrix, R X is the actual residual.

[0151] R L =L est -L;

[0152] L is the estimation matrix of the remaining training data; L est is the estimated matrix of the remaining training data, R L is the health residual.

[0153] The principles and technologies adopted by an anomaly detection device based on real-time operation data of new energy vehicles are the same as those of an anomaly detection method based on real-time operation data of new energy vehicles. The principles and processes of an anomaly detection method based on real-time operation data of new energy vehicles have been described in detail in the previous article, which will not be repeated here.

[0154] This application selects an accurate but not too complex set of power battery monitoring parameters by studying and analyzing the failure mechanisms of internal and external power batteries. Aiming at the trade-off between the effectiveness of BMS fault diagnosis and the completeness of historical data, this application proposes an efficient anomaly detection method based on real-time operation data, which can accurately detect abnormal states when the power battery system is degraded. The training process is efficient and time-saving.

[0155] The above technical solution of the embodiment of the present invention is described in detail below in conjunction with specific application examples. For technical details not introduced during the implementation process, please refer to the relevant description in the previous text.

[0156] Embodiment 1:

[0157] At present, based on the real-time operation data of the power battery system of new energy vehicles (collected in accordance with the requirements of GB / T 32960-2016), the research on the fault diagnosis algorithm for the operation safety of the power battery system of new energy vehicles is still in its infancy, and lacks systematic research results. This application is based on the actual situation of limited computing space and time of the BMS of new energy vehicles, and aims at the trade-off between the accuracy and complexity of BMS fault modeling, the trade-off between the effectiveness of BMS fault diagnosis and the completeness of historical data, and the trade-off between the real-time nature of BMS fault diagnosis and the availability of resources. The technical solutions of this project are as follows:

[0158] 1) By sorting out the internal and external failure mechanisms of power batteries, we conducted research on accurate power battery failure modeling, and then developed an accurate but not overly complex power battery failure model based on the power battery-related data and component expected failure relationship model of the SAIC-GM-Wuling new energy vehicle monitoring and early warning platform;

[0159] 2) Based on the power battery fault model, we study the multivariate state estimation technology and propose an efficient anomaly detection algorithm suitable for new energy vehicle BMS: Improved Multivariate State Estimation Technique (IMSET);

[0160] 3) Based on the research of anomaly detection algorithm, the actual residual and healthy residual of IMSET are input into the nonparametric sequential probability ratio test (NSPRT), combined with the universal threshold set offline, to realize adaptive threshold fault diagnosis;

[0161] 4) Based on the above research, the research results are verified on the SAIC-GM-Wuling new energy vehicle monitoring and early warning platform based on real-time operation data. The proposed research method is improved through actual verification and applied to the next generation of new energy vehicles. This application has made application results that break through technical bottlenecks in the aspects of "Analysis of failure mechanism of internal and external faults of power batteries", "Research on efficient anomaly detection algorithm based on IMSET", and "Research on adaptive threshold fault diagnosis algorithm based on IMSET and NSPRT".

[0162] Embodiment 2:

[0163] An anomaly detection method based on real-time operation data of new energy vehicles, such as Figure 3As shown in the figure, in order to make the selected training data T effective when there is no historical data, it is necessary to update it. The purpose of the update is to make the training data T contain all dynamic changes of each monitoring parameter under normal operating conditions as much as possible, especially its state during normal degradation. However, the update without historical data is different from the update with historical data. This project proposes two update methods based on actual engineering applications:

[0164] Unquantified updates:

[0165] According to the results of real-time detection, the time state when each detection result is a healthy state is added to the training data T. This updating method is called variable-quantity updating.

[0166] Quantitative update:

[0167] In the form of a floating window, only the data in the time state detected as healthy in the floating window is added to the initial training data T each time. This updating method is called quantitative updating.

[0168] When the range of change of the monitoring parameters is not very large, if the non-quantitative update method is used to update the training data T, then as the detection continues, the training data T must be updated on the basis of the last detection, and the health status detected at this time will be added to the training data T of the last detection. However, since the range of change of the monitoring parameters is not very large, the data included in the obtained training data T will be repeated in the long run, and the calculation speed will become slower and slower. At this time, if the quantitative update method is used, not only the training data T meets the purpose of including the dynamic changes of the monitoring parameters, but also in actual engineering applications, the computing power of the BMS is limited, so the quantitative update will greatly improve the calculation speed. Therefore, when the range of change of the monitoring parameters is not very significant, the quantitative update method is obviously better than the non-quantitative update method.

[0169] (2) Selection and optimization of nonlinear operators

[0170] If the weight vectors W and W′ are to exist, D T D must be reversible, but D T A necessary but not sufficient condition for D to be reversible is that the number of columns of the memory matrix D should be smaller than the number of rows.

[0171] In other words, assuming that the memory matrix D is represented as an n×k1 data matrix, where n represents the number of monitoring parameters and k1 represents the number of time states contained in the memory matrix D, then D TIf D is to be reversible, the number of columns of the memory matrix D (i.e., the number of time states k1 it contains) should be less than the number of rows (i.e., the number of monitoring parameters n). However, in actual engineering applications, this condition is difficult to meet. Especially in the field of engineering applications, in order to obtain enough statistical information on monitoring parameters, the number of time states contained in the training data T must be greater than the number of monitoring parameters. In addition, in order to make the estimated matrix X est Although the number of time states contained in the memory matrix D is not necessarily the more the better, the number of time states it contains must not be less than the number of monitoring parameters. Therefore, in order to solve the correlation problem between data in different time states in actual engineering applications, and thus lead to D T The irreversibility of the D matrix limits the application scope of the least square error estimation method, so the state estimation method based on IMSET introduces a nonlinear operator.

[0172] Nonlinear operators are also called nonlinear mappings. They use advanced pattern recognition methods to fit input data or vectors and calculate the normalized similarity between different data or vectors. Nonlinear operators are commonly used in equations with the symbol express.

[0173] Nonlinear operators also have a professional name, namely kernel function, which is generally represented by K in equations. h (x,y) represents.

[0174] For an x,y∈R d The kernel function K h (x,y), it has the following properties:

[0175] When x=y, the kernel function K h (x,y) gets the maximum value;

[0176] |K h (x,y)| decreases as |xy| decreases;

[0177] K h (x,y) is a general function with 2d variables.

[0178] The most common nonlinear operators are as follows:

[0179] Euclidean norm;

[0180]

[0181] City block distance;

[0182]

[0183] Linear correlation coefficient;

[0184]

[0185] In the formula and They represent the means of x and y respectively.

[0186] Gaussian operator;

[0187]

[0188] Where h is the filter coefficient or also called bandwidth.

[0189] When using nonlinear operators After replacing the linear vector product operator, equations (3.9), (3.10), (3.14) and (3.15) are as follows:

[0190]

[0191]

[0192]

[0193]

[0194] Nonlinear Operator It is a key technology for condition monitoring based on IMSET. SmartSignal has developed a commercial equipment condition monitoring software (SmartSignal eCM TM ) mainly refers to nonlinear operators. From this, we can also know how important the introduction and selection of nonlinear operators are for state monitoring based on IMSET.

[0195] In fact, when using the Gaussian operator, it also involves a parameter that needs to be manually defined - the bandwidth h. Since it contains potential instability, it involves the question of how to select the bandwidth h so that the estimation effect obtained when using the Gaussian operator is optimized. In actual engineering applications, in order to better use the Gaussian operator, its estimation performance must be optimized. A feasible solution is to introduce ridge regularization. Then, after using ridge regularization, equations (3.22) to (3.25) are as follows:

[0196]

[0197]

[0198]

[0199]

[0200] Where λ is the ridge regularization parameter (λ>0).

[0201] In general, finding the optimal value λ is a one-dimensional optimization problem that can be solved using nonlinear methods such as conjugate gradient descent.

[0202] Based on the above research, by selecting data from the expected fault relationship model as monitoring parameters, and then estimating the state based on IMSET, the actual residuals and healthy residuals of the monitoring parameters are obtained. Furthermore, based on the offline test data, the actual residual threshold is obtained. By comparing the actual residual with the residual threshold, anomaly detection can be achieved.

[0203] This application selects an accurate but not too complex set of power battery monitoring parameters by studying and analyzing the failure mechanisms of internal and external power batteries. Aiming at the trade-off between the effectiveness of BMS fault diagnosis and the completeness of historical data, this application proposes an efficient anomaly detection method based on real-time operation data, which can accurately detect abnormal states when the power battery system is degraded. The training process is efficient and time-saving.

[0204] It should be understood that the specific order or hierarchy of steps in the disclosed process is an example of an exemplary method. Based on design preferences, it should be understood that the specific order or hierarchy of steps in the process can be rearranged without departing from the scope of protection of the present disclosure. The attached method claims present the elements of the various steps in an exemplary order and are not intended to be limited to the specific order or hierarchy described.

[0205] In the above detailed description, various features are grouped together in a single embodiment to simplify the disclosure. This method of disclosure should not be interpreted as reflecting an intention that the embodiments of the claimed subject matter require more features than are clearly stated in each claim. On the contrary, as reflected in the appended claims, the invention is in a state of having less than all the features of the disclosed individual embodiments. Therefore, the appended claims are hereby expressly incorporated into the detailed description, with each claim standing on its own as a separate preferred embodiment of the invention.

[0206] The disclosed embodiments are described above to enable any person skilled in the art to implement or use the present invention. Various modifications of these embodiments are obvious to those skilled in the art, and the general principles defined herein may also be applied to other embodiments without departing from the spirit and scope of the present disclosure. Therefore, the present disclosure is not limited to the embodiments given herein, but is consistent with the broadest scope of the principles and novel features disclosed in this application.

[0207] The above description includes examples of one or more embodiments. Of course, it is impossible to describe all possible combinations of components or methods for the purpose of describing the above embodiments, but it should be recognized by those skilled in the art that the various embodiments may be further combined and arranged. Therefore, the embodiments described herein are intended to cover all such changes, modifications and variations that fall within the scope of protection of the appended claims. In addition, with respect to the term "comprising" used in the specification or claims, the word is covered in a manner similar to the term "including", just as "including," is explained as a transitional word in the claims. In addition, any term "or" used in the specification of the claims is intended to mean "non-exclusive or".

[0208] Those skilled in the art may also understand that the various illustrative logical blocks, units, and steps listed in the embodiments of the present invention may be implemented by electronic hardware, computer software, or a combination of the two. In order to clearly demonstrate the interchangeability of hardware and software, the various illustrative components, units, and steps described above have generally described their functions. Whether such functions are implemented by hardware or software depends on the specific application and the design requirements of the entire system. Those skilled in the art may use various methods to implement the described functions for each specific application, but such implementation should not be understood as exceeding the scope of protection of the embodiments of the present invention.

[0209] The various illustrative logic blocks or units described in the embodiments of the present invention can be implemented or operated by a general-purpose processor, a digital signal processor, an application-specific integrated circuit (ASIC), a field programmable gate array or other programmable logic device, a discrete gate or transistor logic, a discrete hardware component, or any combination of the above. The general-purpose processor can be a microprocessor, and optionally, the general-purpose processor can also be any conventional processor, controller, microcontroller or state machine. The processor can also be implemented by a combination of computing devices, such as a digital signal processor and a microprocessor, a plurality of microprocessors, one or more microprocessors combined with a digital signal processor core, or any other similar configuration.

[0210] The steps of the method or algorithm described in the embodiments of the present invention can be directly embedded in hardware, a software module executed by a processor, or a combination of the two. The software module can be stored in a RAM memory, a flash memory, a ROM memory, an EPROM memory, an EEPROM memory, a register, a hard disk, a removable disk, a CD-ROM, or other storage media of any form in the art. Exemplarily, the storage medium can be connected to the processor so that the processor can read information from the storage medium and can write information to the storage medium. Optionally, the storage medium can also be integrated into the processor. The processor and the storage medium can be arranged in an ASIC, and the ASIC can be arranged in a user terminal. Optionally, the processor and the storage medium can also be arranged in different components in the user terminal.

[0211] In one or more exemplary designs, the above functions described in the embodiments of the present invention can be implemented in hardware, software, firmware or any combination of the three. If implemented in software, these functions can be stored on a computer-readable medium, or transmitted in the form of one or more instructions or codes on a computer-readable medium. Computer-readable media include computer storage media and communication media that facilitate the transfer of computer programs from one place to another. The storage medium can be any available medium that can be accessed by any general or special computer. For example, such computer-readable media can include but are not limited to RAM, ROM, EEPROM, CD-ROM or other optical disk storage, disk storage or other magnetic storage devices, or any other medium that can be used to carry or store program codes in the form of instructions or data structures and other forms that can be read by general or special computers, or general or special processors. In addition, any connection can be appropriately defined as a computer-readable medium, for example, if the software is transmitted from a website site, server or other remote resource through a coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL) or wirelessly, such as infrared, wireless and microwave, it is also included in the defined computer-readable medium. The disk and disc include compact disk, laser disk, optical disk, DVD, floppy disk and Blu-ray disk. Disks usually copy data magnetically, while discs usually copy data optically with lasers. The above combination can also be included in computer readable media.

[0212] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. An abnormality detection method based on real-time operation data of new energy vehicles, characterized in that: The following steps are involved: Obtain the real-time operation data and historical health data of the monitoring parameters of the new energy vehicle power battery system, and obtain the real-time operation data and historical health data of the key monitoring parameters based on failure analysis; The real-time operation data of the key monitoring parameters and the historical health data of the key monitoring parameters are combined with the multivariate state estimation technology to obtain the anomaly detection model and the residuals of each monitoring parameter in the anomaly detection model, wherein the residuals include healthy residuals and actual residuals; Performing a sequential probability ratio test on the residual, and obtaining an abnormality detection result of the new energy vehicle power battery system based on an index of the sequential probability ratio test; The real-time operation data of the key monitoring parameters and the historical health data of the key monitoring parameters are combined with the multivariate state estimation technology to obtain the abnormal detection model and the residuals of each monitoring parameter in the abnormal detection model. The residuals include healthy residuals and actual residuals, specifically including: Establish an observation matrix based on real-time operating data of key monitoring parameters; Calculate the estimation matrix of the observation matrix; Calculating actual residuals using the estimated matrix; The real-time operation data of the key monitoring parameters and the historical health data of the key monitoring parameters are combined with the multivariate state estimation technology to obtain the abnormal detection model and the residuals of each monitoring parameter in the abnormal detection model, wherein the residuals include healthy residuals and actual residuals, and also include: forming training data based on historical health data of the key monitoring parameters; Establishing a memory matrix based on the training data and forming remaining training data; Calculate an estimation matrix of the remaining training data according to the memory matrix and the remaining training data; calculating health residuals using the estimation matrix; Calculate the residuals, including: R X =X est -X obs ; Among them, X obs is the observation matrix, X est is the estimation matrix, R X is the actual residual; R L =L est -L; L is the estimation matrix of the remaining training data; L est is the estimated matrix of the remaining training data, R L is the health residual.

2. The method for detecting anomalies based on real-time operation data of new energy vehicles according to claim 1, characterized in that: The residual is subjected to a sequential probability ratio test, and based on the index of the sequential probability ratio test, an abnormality detection result of the new energy vehicle power battery system is obtained, specifically including: If Λ(Y n )≥B, then accept hypothesis H1, H1 is the probability density function 1; If Λ(Y n )≤A, then accept hypothesis H0, H0 is the probability density function 2; If A<Λ(Y n )<B, continue sampling, and then repeat the comparison; Among them, Λ(Y n ) represents the likelihood ratio function, A and B are constants, and 0<A<1<B<∞, Y n are n independent and identically distributed samples.

3. An abnormality detection device based on real-time operation data of new energy vehicles, characterized in that: include: An acquisition unit, used to acquire real-time operation data of monitoring parameters and historical health data of monitoring parameters of a new energy vehicle power battery system, and obtain real-time operation data of key monitoring parameters and historical health data of key monitoring parameters based on failure analysis; A calculation unit, used to combine the real-time operation data of the key monitoring parameters and the historical health data of the key monitoring parameters with the multivariate state estimation technology to obtain the anomaly detection model and the residuals of each monitoring parameter in the anomaly detection model, wherein the residuals include healthy residuals and actual residuals; A detection unit, used to perform a sequential probability ratio test on the residual, and obtain an abnormality detection result of the new energy vehicle power battery system based on an index of the sequential probability ratio test; The computing unit specifically includes: Establish an observation matrix based on real-time operating data of key monitoring parameters; Calculate the estimation matrix of the observation matrix; Calculating actual residuals using the estimated matrix; The computing unit further includes: forming training data based on historical health data of the key monitoring parameters; Establishing a memory matrix based on the training data and forming remaining training data; Calculate an estimation matrix of the remaining training data according to the memory matrix and the remaining training data; calculating health residuals using the estimation matrix; Calculate the residuals, including: R X =X est -X obs ; Among them, X obs is the observation matrix, X est is the estimation matrix, R X is the actual residual; R L =L est -L; L is the estimation matrix of the remaining training data; L est is the estimated matrix of the remaining training data, R L is the health residual.

4. The abnormality detection device based on real-time operation data of new energy vehicles according to claim 3 is characterized in that: The detection unit specifically comprises: If Λ(Y n )≥B, then accept hypothesis H1, H1 is the probability density function 1; If Λ(Y n )≤A, then accept hypothesis H0, H0 is the probability density function 2; If A<Λ(Y n )<B, continue sampling, and then repeat the comparison; Among them, Λ(Y n ) represents the likelihood ratio function, A and B are constants, and 0<A<1<B<∞, Y n are independent and identically distributed samples.

Images