Contact network part failure association classification method based on three-way decision theory

Abstract

The invention discloses an overhead line system part failure association classification method based on a three-way decision theory. The method comprises the following steps: acquiring historical failure data of overhead line system parts; defining failure association between the parts; constructing a failure association data set among the contact network parts; constructing a failure association risk assessment model; constructing a risk amplification function model; evaluating a failure association risk level between the parts, and calculating a risk amplification function value; constructing a failure association classification model based on a three-way decision theory; classifying the failure association into attraction association, mutual exclusion association and uncertain association; according to the method, the risk level of failure association between the parts can be evaluated, and the failure propagation mechanism of the parts can be revealed from the perspective of how to cause failure risk changes of other parts after one part fails.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of electrified railway traction power supply technology, in particular to a catenary component failure correlation classification method based on three-way decision theory. BACKGROUND

[0002] As an important part of the traction power supply system, the catenary system of high-speed railway bears the key task of continuous and stable power supply for high-speed trains. Due to the large structure, complex operating environment and closely coupled stress relationship of the catenary system, once the components fail, they will often affect the state of other components through mechanical coupling or electrical transmission, and even cause a chain of failures, threatening the safe operation of the train.

[0003] At present, the failure analysis of the catenary system mainly relies on fault tree analysis (FTA), reliability block diagram (RBD), Bayesian network (BN) and other methods. These methods can quantify the reliability and risk level of the catenary system by establishing a structured logical model.

[0004] However, this kind of method generally needs to rely on prior modeling of catenary structure and failure logic, and it is difficult to deal with a large number of dynamic failure propagation relationships in the system. At the same time, existing methods are generally based on the implicit assumption that "component failure will inevitably increase system risk", that is, all failure correlations are attractive correlations. However, in actual operation, the failure of some components in the catenary system may reduce the failure risk of other components through stress release, load transfer or mechanical isolation, i.e. there are mutual exclusion correlations. Therefore, in order to solve the above problems, it is of great theoretical significance and engineering value to propose a failure correlation classification method for components. SUMMARY

[0005] In order to overcome the defects of the prior art and reveal the failure correlation characteristics between components, the present application proposes a catenary component failure correlation classification method based on three-way decision theory. This method classifies each failure correlation as attractive correlation, mutual exclusion correlation and uncertain correlation by constructing a failure correlation classification model based on three-way decision theory, so as to reveal the propagation mechanism of failure in the system.

[0006] The purpose of the present application can be achieved by the following technical solutions:

[0007] A catenary component failure correlation classification method based on three-way decision theory, comprising the following steps:

[0008] S1: Obtain a historical failure data set of catenary components;

[0009] S2: Define the failure correlation between components and construct a failure correlation data set between catenary components;

[0010] S3: constructing a failure correlation risk assessment model considering time decay coefficient;

[0011] S4: calculating the risk coefficient of each failure correlation according to the failure correlation risk assessment model, and constructing a failure correlation risk amplification function model based on the risk coefficient;

[0012] S5: calculating the risk amplification function value of the failure correlation according to the risk assessment model and the risk amplification function model;

[0013] S6: constructing a contact net component failure correlation classification model based on three-branch decision theory according to the risk amplification function value, including constructing upper approximation and lower approximation based on three-branch decision theory, constructing a target subset of failure correlation, and constructing a classification domain based on three-branch decision theory;

[0014] S7: constructing a three-branch decision classification rule according to the target subset of failure correlation and the classification domain based on three-branch decision theory, and classifying the failure correlation.

[0015] Further, in the historical failure data set of the contact net component, if the failure of component l i will cause the failure risk of component l j to change, then the failure correlation between component l i and l j is defined as failure correlation l i →l j .

[0016] Further, the specific method of constructing the failure correlation risk assessment model considering time decay coefficient includes:

[0017] S31: the calculation method of the reference risk function of any component l i is shown in formula (S-1):

[0018]

[0019] In formula (S-1), represents the sum of the severity weights of system failure events in the data set, N i represents the failure frequency of component l i , represents the system global risk smoothing item, D represents the observation days of the failure data set, N represents the number of component types in the data set, and R w (i) is greater, the risk level of component l i is higher;

[0020] S32: according to the reference risk function Rw (i), the failure correlation risk coefficient R(l i →l j ) considering the time decay coefficient is constructed as shown in formula (S-2):

[0021]

[0022] In formula (S-2), Δt represents the time difference of failures of the post-positioned component l j and the pre-positioned component l i , T represents the artificially set failure correlation time window, ω i and ω j respectively represent the severity weight of the components l i and l j under the condition of 0 < Δt < T, represents the severity weight of the kth failure event of the component l i in the failure data set, and λ represents the time decay coefficient.

[0023] Further, based on the baseline risk function of the component and the failure correlation risk coefficient, a failure correlation risk amplification function model is constructed as shown in formula (S-3):

[0024]

[0025] Further, the specific steps of the lower approximation and the upper approximation of the three-branch decision theory include:

[0026] S51: A failure correlation to be classified is represented as x, A = {f} is an attribute set, and the attribute f induces an equivalence relation ~ m on the domain U = [x1, x2,..., x f ]: The equivalence class [x] ～f = {y ∈ U | y ~ f x} represents all objects with the same attributes as x.

[0027] S52: For any subset X ∈ U, the lower approximation apr f (X) and the upper approximation are constructed as shown in formula (S-4) and (S-5) respectively:

[0028] apr f (X) = {x ∈ U | [x] ～f ∈ X} (S-4)

[0029]

[0030] In formula (S-4), aprf (X) represents all equivalence classes completely subordinated to X, in formula (S-5), represents all equivalence classes intersecting with X.

[0031] Further, for any invalid association, the target subset of the invalid association includes the target subset X with attractive invalid association Attract , the target subset X with repulsive invalid association Rpel and the target subset X with uncertain invalid association Uncertain , respectively as shown in formula (S-6), (S-7) and (S-8):

[0032] X Attract = {x∈U|α x >1+θ} (S-6)

[0033] X Rpel = {x∈U|α x <1-θ} (S-7)

[0034] X Uncertain =U\(X Attract ∪X Repel )={x∈U|1-θ≤α x ≤1+θ} (S-8)

[0035] In formula (S-6)-(S-8), α x represents the risk amplification function value of the invalid association to be classified, and θ represents the size of the uncertainty interval set artificially.

[0036] Further, in the three target subsets X Attract with attractive invalid association, X Rpel with repulsive invalid association and X Uncertain with uncertain invalid association, only the approximate determination domain is taken as the classification domain, and the rest is taken as the uncertain domain, to construct the classification domain based on the three-branch decision theory, respectively as shown in formula (S-9), (S-10) and (S-11):

[0037]

[0038] d Uncertain =U\(D Attract UD Repel ) (S-11)

[0039] In formula (S-9), d Attract represents the classification domain consistent with attractive association, in formula (S-10), d Repel represents the classification domain consistent with repulsive association, and in formula (S-11), dUncertain representing all the rest of the failure correlations.

[0040] Further, according to the classification field, a three-way decision classification rule is established as shown in formula (S-12):

[0041]

[0042] Further, the failure correlations are classified into attractive failure correlations, mutually exclusive failure correlations and uncertain failure correlations.

[0043] Compared with the prior art, the present application has the following technical effects:

[0044] (1) The present application comprehensively considers the time decay factor in the risk assessment of failure correlations;

[0045] (2) The present application classifies the failure correlations into attractive correlations, mutually exclusive correlations and uncertain correlations to reveal the propagation mechanism of failure in the system. BRIEF DESCRIPTION OF DRAWINGS

[0046] fig. 1 is the overall flowchart of the present application;

[0047] fig. 2 is the risk coefficient heat map of the failure correlations between the contact net parts;

[0048] fig. 3 is the classification situation of the failure correlations of each part as a front part. DETAILED DESCRIPTION

[0049] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments of the present application, any modification, equivalent replacement, improvement, etc. obtained by those skilled in the art without creative labor should be included in the protection scope of the present application.

[0050] As shown in fig. 1 , a contact net part failure correlation classification method based on three-way decision theory includes the following steps:

[0051] S1: Obtain a historical failure data set of contact net parts;

[0052] S2: Based on the historical failure data set of the contact net parts, if the failure of part l i will cause the failure risk of part l j to change, then define parts l i and l jFailure correlation exists between the two i →l j .

[0053] S3: Construct a failure correlation risk assessment model considering the time decay coefficient:

[0054] S31: For any component l i , the reference risk function calculation method is shown in formula (S-1):

[0055]

[0056] In formula (S-1), N i represents the failure frequency of component l i , represents the sum of severity weights of system failure events in the data set, represents the system global risk smoothing term, which is used to avoid underestimation of the risk of components with small sample size, D represents the observation days of the failure data set, N represents the number of components in the data set, and R w (i) is the reference risk function of component l i .

[0057] S32: According to the reference risk function R w (i) of the component described in S31, a failure correlation risk coefficient R(l i →l j ) considering the time decay coefficient is constructed, which is shown in formula (S-2):

[0058]

[0059] In formula (S-2), Δt represents the failure time difference between the post-component l j and the pre-component l i , T represents the artificially set failure correlation time window, ω i and ω j respectively represent the severity weights of components l i and l j under the condition of 0<Δt<T, represents the severity weight of the kth failure event of component l i in the failure data set, and λ represents the time decay coefficient.

[0060] S4: Based on the reference risk function of the component and the failure correlation risk coefficient, a failure correlation risk amplification function model is constructed, which is shown in formula (S-3):

[0061]

[0062] S5: Calculate the risk amplification function value associated with failure;

[0063] S6: Based on the risk amplification function value, construct a failure association classification model for overhead contact system components based on three-branch decision theory, including constructing upper and lower approximations based on three-branch decision theory, constructing a target subset of failure associations, and a classification domain based on three-branch decision theory.

[0064] The construction methods for the lower and upper approximations based on three-branch decision theory include:

[0065] S51: Let x represent a failure association to be classified, and A = {f} be a set of attributes, where attribute f is in the universe of discourse U = [x1, x2, ..., x]. m Induce an equivalence relation on [the above]. f : Equivalence class [x] ～f ={y∈U|y~ f x} represents an object that has all properties that are the same as x.

[0066] S52: For any subset X∈U, construct the following approximation apr f (X) and the upper approximation As shown in equations (S-4) and (S-5) respectively:

[0067] apr f (X)={x∈U|[x] ～f ∈X} (S-4)

[0068]

[0069] In equation (S-4), apr f (X) represents the equivalence class containing all classes that are completely subordinate to X, in equation (S-5), This represents an equivalence class that contains all elements that intersect with X.

[0070] Furthermore, for any failure association, the target subset of the failure association includes a target subset X with attractive failure associations. Attract A subset X of targets with mutually exclusive failure associations Rpel and the target subset X with uncertain failure associations Uncertain As shown in equations (S-6), (S-7), and (S-8) respectively:

[0071] X Attract ={x∈U|α x >1+θ} (S-6)

[0072] X Rpel= {x e U | a x <1-θ} (S-7)

[0073] X Uncertain = U\(X Attract ∪X Repel ) = {x e U | 1-θ ≤ a x ≤ 1+θ} (S-8)

[0074] In formula (S-6) to (S-8), a x represents the risk amplification function value of the failure correlation to be classified, and θ represents the size of the uncertainty interval set artificially.

[0075] Further, in the target subset X Attract with attractive failure correlation, the target subset X Rpel with mutually exclusive failure correlation, and the target subset X Uncertain with uncertain failure correlation, only the approximate determination domain is taken as the classification domain, and the rest is taken as the uncertain domain, to construct the classification domain based on the three-branch decision theory, as shown in formula (S-9), (S-10) and (S-11) respectively:

[0076]

[0077] d Uncertain = U\(D Attract UD Repel ) (S-11)

[0078] In formula (S-9), d Attract represents the classification domain consistent with attractive correlation, in formula (S-10), d Repel represents the classification domain consistent with mutually exclusive correlation, and in formula (S-11), d Uncertain represents all the rest of the failure correlation.

[0079] S7: According to the target subset of the failure correlation and the classification domain based on the three-branch decision theory, a three-branch decision classification rule is established as shown in formula (S-12):

[0080]

[0081] The method proposed in the present application is used, taking the Chengdu Railway Bureau as an example, the contact net part name and number thereof are shown in Table 1.

[0082] Table 1

[0083]

[0084] By constructing the historical failure correlation data set of contact net components of Chengdu Railway Bureau, the risk coefficients of each failure correlation and the risk amplification function are calculated, and a classification model based on three-branch decision theory is constructed to classify the failure correlations.

[0085] As shown in fig. 2 , the reference risk coefficients of each component are calculated by formula (S-1), and the risk coefficients of each failure correlation are calculated by formula (S-2).

[0086] Analysis fig. 2 It can be seen that more than half of the maximum risk coefficient values of the failure correlations exceed 0.5, and even a part of the maximum risk coefficient values of the failure correlations exceed 0.7. It is worth noting that the maximum risk coefficients of the failure correlations between component 9 (support device) and component 10 (positioning device) exceed 0.9, which indicates that there may be a high-risk failure propagation path between these two components, and we need to pay attention to it in the operation and maintenance process.

[0087] As shown in fig. 3 , the risk amplification function of each failure correlation is calculated by formula (S-3), and each failure correlation is classified by formula (S-4)-(S-12).

[0088] Analysis fig. 3 It can be seen that more than half of the failure correlations are classified as attractive failure correlations, which indicates that the failure of the preceding component in these failure correlations will increase the failure risk of the following component, and we should pay attention to such failure correlations in the operation and maintenance process.

[0089] In addition, a part of the failure correlations cannot be classified, which is uncertain failure correlation, which may be due to the weak mutual influence between these components or insufficient samples, which prompts us to adopt the strategy of delayed intervention and dynamic observation when facing this type of failure correlation in the operation and maintenance process, to avoid misjudgment due to insufficient information.

[0090] It is worth noting that a part of the failure correlations are classified as mutually exclusive failure correlations, which means that the failure of the preceding component in these failure correlations does not lead to the spread of risk, but rather reduces the failure risk of the following component, playing a local protection role. This counterintuitive phenomenon prompts us that in complex engineering systems, the failure of some components may provide a certain degree of protection to the system through mechanical decoupling or stress release, etc. This not only enriches our understanding of the failure propagation mechanism of the system, but also provides a new design idea for the operation and maintenance strategy: by identifying the failure correlations with mutual exclusion characteristics in the system, we can tolerate or guide the failure of non-critical components in appropriate scenarios to achieve the risk balance and optimization of the whole system.

[0091] Through the above analysis of the historical failure data of the overhead contact system parts, it can be seen that the method proposed in the application can not only evaluate the risk level of failure correlation, but also reveal the propagation mechanism of failure in the system from the perspective of attractive failure correlation and mutually exclusive failure correlation.

Claims

1. A method for classifying the failure associations of overhead contact line components based on three-branch decision theory, characterized in that, The classification method includes the following steps: S1: Obtain the historical failure dataset of overhead contact system components; S2: Define the failure associations between components and construct a dataset of failure associations between catenary components; S3: Construct a failure association risk assessment model that considers the time decay coefficient; S4: Construct a failure association risk amplification function model based on risk coefficient; S5: Based on the aforementioned risk assessment model and risk amplification function model, calculate the risk amplification function value associated with failure; S6: Based on the risk amplification function value, construct a failure association classification model for overhead contact system components based on three-branch decision theory, including constructing upper and lower approximations based on three-branch decision theory, constructing a target subset of failure associations, and a classification domain based on three-branch decision theory. S7: Based on the target subset of the failure association and the classification domain based on the three-branch decision theory, construct a three-branch decision classification rule to classify the failure association.

2. The failure association classification method for overhead contact line components based on three-branch decision theory according to claim 1, characterized in that, Based on the interaction characteristics between components, if component l i Failure will cause the components to l j The failure risk changes, defining component l i and l j There is a failure association. i →l j .

3. The failure association classification method for overhead contact line components based on three-branch decision theory according to claim 2, characterized in that, The specific methods for constructing the failure association risk assessment model that considers the time decay coefficient include: S31: Calculate the baseline risk function, using component l i For example, calculate component l i Benchmark risk function R ω The method of (i) is as shown in equation (1): In equation (1), N represents the sum of severity weights for system failure events within the dataset. i Indicates component l i Failure frequency This represents the system's global risk smoothing term, where D represents the number of observation days in the failure dataset, and N represents the number of component types in the dataset. S32: Based on the aforementioned benchmark risk function R w (i) Construct a failure association risk coefficient R(l) that considers the time decay coefficient. i →l j As shown in equation (2): In Equation (2), Δt represents the failure time difference between the后置零部件l j and the前置零部件l i , T represents the artificially set failure correlation time window, ω i and ω j respectively represent the severity weights of component l i and l j under the condition of 0 < Δt < T, represents the severity weight of component l i in the k-th failure event in the failure dataset, and λ represents the time decay coefficient.

4. The failure association classification method for overhead contact line components based on three-branch decision theory according to claim 3, characterized in that, The risk amplification function model for the failure association is shown in equation (3):

5. The failure association classification method for overhead contact line components based on three-branch decision theory according to claim 4, characterized in that, The construction methods for the lower and upper approximations based on three-branch decision theory include: S51: Let x represent a failure association to be classified, and A = {f} be a set of attributes, where attribute f is in the universe of discourse U = [x1, x2, ..., x]. m Induce an equivalence relation on [the above]. f : Equivalence classes This represents an object whose properties are the same as x. S52: For any subset X∈U, construct the following approximation apr f (X) and the upper approximation As shown in equations (4) and (5) respectively: In equation (4), apr f (X) represents the equivalence class containing all classes that are completely subordinate to X, in equation (5), This represents an equivalence class that contains all elements that intersect with X.

6. The failure association classification method for overhead contact line components based on three-branch decision theory according to claim 5, characterized in that, For any failure association, the target subset of the failure association includes the target subset X that has attractive failure associations. Attract A subset X of targets with mutually exclusive failure associations Rpel and the target subset X with uncertain failure associations Uncertain As shown in equations (6), (7) and (8) respectively: X Attract ={x∈U|α x >1+θ} (6) X Rpel ={x∈U|α x <1-θ} (7) X Uncertain =U\(X Attract ∪X Repel )={x∈U|1-θ≤α x ≤1+θ} (8); In equations (6) to (8), α x θ represents the risk amplification function value of the failure association to be classified, and θ represents the size of the uncertainty interval set by the user.

7. The failure association classification method for overhead contact line components based on three-branch decision theory according to claim 6, characterized in that, The classification domains based on three-branch decision theory include classification domains that are consistently attractive associations, classification domains that are consistently mutually exclusive associations, and all other invalid associations. The specific construction method includes: In the target subset X with attraction failure association Attract In this process, only the approximate value is taken as the definite domain of the classification domain, and the remaining objects are taken as the uncertain domain, thus constructing a classification domain d that is consistently attractively correlated. Attract As shown in equation (9): In the target subset X with mutually exclusive failure association Rpel In this process, only the approximate values ​​are taken as the definite domain of the classification domain, and the remaining objects are taken as the uncertain domain, thus constructing a classification domain d that is consistently mutually exclusive. Repel As in equation (10): In the target subset X with uncertain failure correlation Uncertain In this process, only the approximate value is taken as the definite domain of the classification domain, and the remaining objects are taken as the uncertain domain, thus constructing the classification domain d for all other failure associations. Uncertain As in equation (11): d Uncertain =U\(D Attract UD Repel ) (11)。 8. The failure association classification method for overhead contact line components based on three-branch decision theory according to claim 7, characterized in that, The three-branch decision classification rule D TWD (x) is shown in equation (12):

9. The failure association classification method for overhead contact line components based on three-branch decision theory according to claim 8, characterized in that, Failure associations are classified into attractive failure associations, mutually exclusive failure associations, and uncertain failure associations.

Images