Lithium battery cloud data driving voltage extreme value fault early warning optimization method

Abstract

The invention discloses a lithium battery cloud data driving voltage extreme value fault early warning optimization method, and belongs to the technical field of fault early warning and software algorithms. According to the method, by analyzing the relation between battery consistency and battery abnormity, battery fault judgment is converted into voltage consistency evaluation; according to the method, firstly, battery health feature extraction is carried out by utilizing feature extraction and feature selection, secondly, density noise space clustering particles are generated based on a minimum domain radius calculation principle, and voltage consistency is measured by optimizing the space clustering particles, tracking an individual extreme value and a group extreme value and updating an individual position. And then defining a parameter with an extreme value dimension as one to improve the generalization ability of the algorithm for actual working conditions, and finally determining a hyper-parameter in the proposed early warning algorithm by using an orthogonal experiment to realize battery system fault early warning. According to the invention, fault early warning of the lithium battery system can be realized, safety accidents caused by improper maintenance in the use process of the battery are avoided, and the occurrence rate of fault safety accidents of the lithium battery system is reduced.

Description

TECHNICAL FIELD

[0001] The lithium battery cloud data driven voltage extreme value fault early warning optimization method belongs to the technical field of fault early warning and software algorithm. BACKGROUND

[0002] Lithium-ion batteries are currently the best choice for electric vehicle power sources due to their high energy density, long cycle life, and low self-discharge rate. However, battery safety accidents caused by battery failures occur frequently, and although vehicles are equipped with battery management systems (BMS) for state monitoring and fault diagnosis, the BMS fault diagnosis method is usually based on simple rules, which is difficult to cope with complex fault mechanisms and achieve early warning of faults. How to early warn battery failures has become a research focus at home and abroad.

[0003] Currently, battery fault diagnosis and early warning methods are mainly divided into rule-based methods, model-based methods, signal-based methods, and data-driven methods. Rule-based methods trigger alarms by detecting whether the values of key parameters (such as voltage and temperature) exceed specified thresholds. However, due to the complex operating environment of electric vehicles, it is difficult to define appropriate thresholds. If the threshold is too low, the alarm may be too sensitive; on the other hand, if the threshold is too high, a serious fault may have occurred when the alarm occurs, making the significance of fault early warning limited.

[0004] With the development of big data and cloud computing, implementing battery fault early warning on the cloud platform of electric vehicles has become an increasingly important direction for automotive power battery management technology. Although current fault diagnosis algorithms can achieve fault early warning under certain conditions, there are still problems to be solved when applied to the vehicle cloud platform.

[0005] 1) Algorithm complexity and generalization problem. Cloud data is large in scale, and if the algorithm logic is too complex, it will reduce the diagnosis speed and increase the computing cost, making it difficult to be accepted in actual tasks. At the same time, the influence of the randomness of real working conditions needs to be considered.

[0006] 2) Data labeling problem. Supervised learning-based fault diagnosis methods are directly affected by the quality of data labeling. Currently, many supervised learning-based fault diagnosis studies use experiments to simulate battery faults, but lack comprehensiveness in real working conditions, and actual cloud fault diagnosis faces the problems of missing fault labels and low label accuracy, which greatly limits the application of supervised learning-based methods in cloud battery fault diagnosis.

[0007] 3) Early warning and positioning of faults. Evolutionary faults do not have obvious characteristics in the early stage, and traditional rule-based methods are difficult to identify faulty batteries and locate the faulty cells in the battery pack. SUMMARY

[0008] In view of the deficiencies of the prior art, the present application provides a lithium battery cloud data driven voltage extreme value fault early warning optimization method, which converts the judgment of battery failure into the problem of voltage consistency evaluation by analyzing the relationship between battery consistency and battery abnormalities. Firstly, the method uses feature extraction and feature selection to extract battery health features. Secondly, based on the minimum field radius calculation principle, density noise space clustering particles are generated. By optimizing the space clustering particles, the individual extreme value and the population extreme value are tracked to update the individual position to measure the voltage consistency. Then, the dimensionless parameter of the extreme value is defined to improve the generalization ability of the algorithm for actual working conditions. Finally, the hyperparameters in the proposed early warning algorithm are determined by orthogonal experiment to realize battery system fault early warning. The present application can avoid safety accidents caused by improper maintenance of the battery during use, and reduce the incidence of lithium battery system failure safety accidents.

[0009] The object of the present application is achieved in that

[0010] A lithium battery cloud data driven voltage extreme value fault early warning optimization method, comprising the following steps:

[0011] Step a, extracting the effective charging cycle voltage and voltage gradient characteristic value of the battery system; comprising:

[0012] The battery system includes m battery monomers;

[0013] The sampling points in the charging process are n;

[0014] The voltage matrix V and the voltage gradient matrix dV are:

[0015]

[0016] dV it = v it -v i(t-τ)

[0017]

[0018] Wherein, the voltage change amount of the voltage gradient time interval t is defined; Vi (1 < i < m, 1 < t < n) represents the voltage of the i th monomer battery at the t th sampling time; dV (1 < i < m, T < t < n) represents the voltage gradient of the i th monomer battery at the t th sampling time;

[0019] Step b, obtaining the voltage feature vector and f1 and f2 based on step a; wherein the voltage feature vectors f1 and f2 are:

[0020]

[0021] Step c, generating density noise spatial clustering particles based on the minimum neighborhood radius calculation principle of step b; wherein the minimum neighborhood radius is:

[0022]

[0023] In order to adapt to random working conditions, the neighborhood radius Eps is based on the average minimum distance d min,ave of the metric;

[0024] Step d, based on step c, the individual position is updated by tracking individual extreme value and population extreme value to measure voltage consistency by optimizing spatial clustering particles;

[0025] Step e, define the dimension one parameter of extreme value to improve the generalization ability of the algorithm for actual working conditions; wherein the dimension is:

[0026] The dimension one parameter distance ratio δ is defined as:

[0027]

[0028] Step f, the super parameter in the proposed early warning algorithm is determined by using orthogonal experiment to realize battery system fault early warning.

[0029] The above-mentioned lithium battery cloud data driven voltage extreme fault early warning optimization method, the lithium battery open circuit voltage curve calculation method is:

[0030] R v (t)=maxv it -minv it

[0031] α j =argmaxR vj ,j=1,2,…,k

[0032] R d v(t)=maxdv it -mindv it

[0033] β j =argmaxR dvj ,j=1,2,…,k

[0034] Wherein, R v (t)(1≤t≤n) represents the voltage range of all single bodies at the t sampling moment, R dv (t)(τ≤t≤n) represents the voltage gradient range of all single bodies at the t sampling moment, α j (1≤j≤k) and β j (1≤j≤k) are R v and R dvTime index of the jth largest element.

[0035] Advantages

[0036] Firstly, the method of the application uses feature extraction to extract battery health features and draw lithium battery voltage characteristic curves, so as to convert the relationship between battery consistency and battery abnormalities into a voltage consistency evaluation problem, reduce the timeliness of lithium battery cloud data driven fault recognition, and provide a feasible method for lithium battery cloud data driven fault recognition.

[0037] Secondly, density noise space clustering particles are generated based on the minimum field radius calculation principle, the space clustering particles are optimized, the individual extreme value and the group extreme value are tracked to update the individual position to measure the voltage consistency, and the extreme value dimension is defined as a parameter to improve the generalization ability of the algorithm to actual working conditions. The super parameters in the proposed early warning algorithm are determined by using orthogonal experiments, lithium battery system fault early warning is realized, and the lithium battery system fault early warning accuracy is improved.

[0038] Thirdly, the open circuit voltage calculation method of the lithium battery system is optimized, so that the sample data of the application is more accurate. BRIEF DESCRIPTION OF DRAWINGS

[0039] Figure 1 A flowchart of the lithium battery cloud data driven voltage extreme value fault early warning optimization method of the application;

[0040] Figure 2 An open circuit voltage (OCV) curve diagram of a lithium battery sample obtained by the application;

[0041] Figure 3 A simulation software panel screenshot obtained by the application; DETAILED DESCRIPTION

[0042] The specific embodiments of the application will be further described in detail below with reference to the accompanying drawings. Specific embodiment one

[0044] This embodiment is a lithium battery cloud data driven voltage extreme value fault early warning optimization method embodiment.

[0045] The lithium battery cloud data driven voltage extreme value fault early warning optimization method of this embodiment, a flowchart as shown in Figure 1 The method comprises the following steps:

[0046] Step a, extracting effective charging cycle voltage and voltage gradient feature values of the battery system; comprising:

[0047] The battery system comprises m battery monomers;

[0048] The charging process has n sampling points;

[0049] The voltage matrix V and the voltage gradient matrix dV are as follows:

[0050]

[0051] dV it = v it - v i(t-τ)

[0052]

[0053] where the voltage change amount at the voltage gradient time interval t is defined; Vi (1 < i < m, 1 < t < n) represents the voltage of the i-th single cell at the t-th sampling moment; dV (1 < i < m, T < t < n) represents the voltage gradient of the i-th single cell at the t-th sampling moment;

[0054] Step b: Based on the voltage feature vectors f1 and f2 obtained in step a; where the voltage feature vectors f1 and f2 are:

[0055]

[0056] Step c: Generate density noise space clustering particles based on the minimum neighborhood radius calculation principle in step b; where the minimum neighborhood radius is:

[0057]

[0058] To adapt to random working conditions, the neighborhood radius Eps is based on the measurement of the average minimum distance d min,ave of;

[0059] Step d: Measure the voltage consistency by optimizing the space clustering particles in step c, tracking the individual extreme value and the population extreme value to update the individual position.

[0060] Step e: Define a parameter with a dimensionless extreme value to improve the generalization ability of the algorithm for actual working conditions; where the dimension is:

[0061] The dimensionless parameter distance ratio δ is defined as:

[0062]

[0063] Step f: Use orthogonal experiments to determine the hyperparameters in the proposed early warning algorithm to achieve battery system fault early warning. Specific Embodiment Two

[0065] This embodiment is an embodiment of an optimization method for lithium battery cloud data-driven voltage extreme value fault early warning.

[0066] This embodiment presents a lithium battery cloud data-driven voltage extreme value fault early warning optimization method. Taking multiple sets of lithium battery cloud detection samples as an example, the simulation is performed on Matlab 2020b software to extract the feature values ​​of the lithium battery voltage matrix. The simulation program is recorded as follows.

[0067]

[0068]

[0069]

[0070]

[0071]

[0072]

[0073]

[0074] The open-circuit voltage (OCV) curve of the lithium battery sample obtained by running the program is shown below. Figure 2 As shown, the screenshot of the simulation software panel is as follows. Figure 3 As shown, in Figure 3 In the diagram, loops above the dashed line represent abnormal loops, while loops below the dashed line represent normal loops.

Claims

1. A cloud-based data-driven method for early warning and optimization of voltage extreme value faults in lithium batteries, characterized in that, It includes the following steps: Step a: Extract the effective charging cycle voltage and voltage gradient eigenvalue of the battery system; It includes: The battery system contains m battery cells; There are n sampling points during the charging process; The voltage matrix V and the voltage gradient matrix dV are: dV it =v it -v i(t-τ) Among them, the voltage change amount at the voltage gradient time interval t is defined; Vi(1 < i < m, 1 < t < n) represents the voltage of the i-th single cell at the t-th sampling moment; dV(1 < i < m, T < t < n) represents the voltage gradient of the i-th single cell at the t-th sampling moment; Step b: Based on step a, obtain the voltage feature vectors f1 and f2; among them, the voltage feature vectors f1 and f2 are: Step c: Generate density noise space clustering particles based on the minimum neighborhood radius calculation principle in step b; among them, the minimum neighborhood radius is: To adapt to stochastic operating conditions, the neighborhood radius Eps is based on the average minimum distance d. min,ave Measure; Step d: Based on step c, measure the voltage consistency by optimizing the space clustering particles, tracking the individual extreme value and the population extreme value to update the individual position; Step e: Define the parameter with the dimension of one for the extreme value to improve the generalization ability of the algorithm for actual working conditions; among them, the dimension is: The parameter distance ratio δ with the dimension of one is defined as: Step f: Use orthogonal experiments to determine the hyperparameters in the proposed warning algorithm to achieve battery system fault warning.

2. The lithium battery cloud data-driven voltage extreme value fault early warning optimization method according to claim 1, characterized in that, The calculation method of the open-circuit voltage curve of the lithium battery is: R v (t)=maxv it -minv it α j =argmaxR vj ,j=1,2,…,k R d v(t)=maxdv it -mindv it β j =argmaxR dvj ,j=1,2,…,k Among them, R v (t)(1≤t≤n) represents the voltage range of a single cell at sampling time t, R dv (t)(τ≤t≤n) represents the voltage gradient range of the individual cells at sampling time t, α j (1≤j≤k) and β j (1≤j≤k) represent R respectively v With R dv The time index of the j-th largest element in the middle.

Images