A maintenance mode prediction method for electromechanical systems based on underlying fault characteristics

Abstract

The present invention provides a method for predicting the maintenance mode of an electromechanical system based on the underlying characteristics of the fault. The method includes: based on the maintenance record of the electromechanical system, taking the fault data as an example and the fault type as a label, constructing a correlation matrix between the instance and the label, clustering all the labels, calculating the Euclidean distance between the instance and the cluster center of the positive and negative instance sets of the label, and obtaining the mapping relationship between the instance and the label; constructing a binary classifier, taking the fault data of the faulty electromechanical system as an unknown instance, inputting the unknown instance into the binary classifier, and outputting a preliminary determination result of the maintenance mode of the faulty electromechanical system; introducing a calibration label, and using the underlying characteristics of the fault and the calibration label to update the preliminary determination result of the maintenance mode of the faulty electromechanical system, and obtaining the final determination result of the maintenance mode. The method provided by the present invention can effectively determine the maintenance mode based on the data collected from the complex electromechanical system, and has high accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of electromechanical system maintenance, and in particular to a method for predicting an electromechanical system maintenance mode based on underlying fault characteristics. Background Art

[0002] As complex electromechanical systems continue to develop in the direction of refinement and intelligence, the difficulty of condition monitoring and maintenance mode of complex electromechanical systems has increased. Complex electromechanical systems are prone to failures due to the influence of various factors such as human and environmental factors during long-term use. It is crucial to observe and detect the operating quality and operating status of system equipment. Based on some requirements of condition repair and life management, parts that exceed the wear range must be replaced in time, and equipment parameters that exceed maintenance standards must be adjusted in time. For example, the status of train traction system components directly affects driving safety. At present, the railway department has accumulated a large number of complex electromechanical system maintenance records. Based on the operating data and fault maintenance data of the monitoring items, how to scientifically, efficiently and accurately determine the maintenance method of complex electromechanical systems is currently a key issue that needs to be solved to ensure the stable operation of equipment.

[0003] At present, there is no effective method for determining the maintenance mode of a complex electromechanical system in the prior art. Summary of the invention

[0004] The embodiment of the present invention provides a method for predicting the maintenance mode of an electromechanical system based on the underlying characteristics of a fault, so as to realize effective maintenance work of a complex electromechanical system.

[0005] In order to achieve the above object, the present invention adopts the following technical scheme.

[0006] A method for predicting maintenance methods of electromechanical systems based on underlying fault characteristics, comprising:

[0007] Based on the maintenance records of electromechanical systems, fault data is used as an instance and fault type is used as a label. The correlation matrix between instances and labels is constructed, all labels are clustered, and the Euclidean distance between the instance and the cluster center of the positive and negative instance sets of the label is calculated to obtain the mapping relationship between the instance and the positive and negative instance sets of the label.

[0008] A binary classifier is constructed according to the correlation matrix between the instance and the label and the mapping relationship between the instance and the positive and negative instance sets of the label, the fault data of the faulty electromechanical system is used as an unknown instance, the unknown instance is input into the binary classifier, and the binary classifier outputs a preliminary determination result of the maintenance method of the faulty electromechanical system;

[0009] A calibration tag is introduced, and the preliminary determination result of the maintenance method of the faulty electromechanical system is updated using the underlying fault characteristics and the calibration tag to obtain a final determination result of the maintenance method.

[0010] Preferably, the maintenance record based on the electromechanical system takes fault data as an instance and fault type as a label, constructs a correlation matrix between the instance and the label, performs clustering processing on all labels, calculates the Euclidean distance between the instance and the cluster center of the positive and negative instance set of the label, and obtains the mapping relationship between the instance and the positive and negative instance set of the label, including:

[0011] Taking each fault data of the faulty electromechanical system as an instance, each fault type as a label, and the specific parameters corresponding to the fault type as eigenvalues, the feature matrix F between the instance and the eigenvalue and the correlation matrix R between the instance and the label are constructed:

[0012]

[0013]

[0014] Among them, m represents the number of all instances, n represents the number of all features, q represents the number of all labels, and a ij is the specific characteristic parameter of this instance, 1≤i≤m, 1≤j≤n, b ij It is -1 or 1, -1 means that the instance does not have the corresponding label, 1 means that the instance has the corresponding label, 1≤i≤m, 1≤j≤q;

[0015] Step 2: For each instance corresponding to a label, if it is related to the label, add it to the positive instance set P of the label; otherwise, add it to the negative instance set N;

[0016] Step 3: Calculate the label entropy E(k) for the positive instance set P and negative instance set N corresponding to each label. p 、E(k) n :

[0017]

[0018]

[0019] Among them, E(k) p represents the label entropy of the positive instance set corresponding to the kth label; hp(j) represents the ratio of the jth label vector appearing in the positive instance set corresponding to the label to the total label vector appearing in the positive instance set; E(k) n and hn(j) respectively represent the label entropy and vector ratio corresponding to the negative instance set of the kth label;

[0020] Step 4: Use the label entropy of the positive and negative instance sets to calculate the basic entropy of each label, and use the basic entropy to evaluate the minimum change of the label:

[0021]

[0022] w k represents the basic entropy of the kth label;

[0023] Step 5: Calculate the number of clusters adaptively based on the training data:

[0024]

[0025] Among them, c k Represents the number of clusters corresponding to the k-th label positive and negative instance set; where r k is the ratio parameter corresponding to the kth label; s is the smoothing parameter; |P| and |N| represent the cardinality of the positive and negative instance sets, respectively. The positive and negative instance sets corresponding to each label are clustered with the same number of clusters c. k Perform k-means clustering separately;

[0026] Step 6: Construct a vector from n to 2c for each label based on the clustering results. k Dimensional mapping:

[0027]

[0028] ε(x) k Represents a mapping constructed from the set of clustering results of the instance and the set of positive and negative instances corresponding to the jth label; is the Euclidean distance between the instance and the cluster centers of the positive instance set of the label, is the Euclidean distance between the instance and each cluster center of the negative instance set of the label. The mapping includes processing the relevant instance set and the irrelevant instance set of the label respectively to obtain different label feature results.

[0029] Preferably, the binary classifier is constructed according to the correlation matrix between the instance and the label and the mapping relationship between the instance and the positive and negative instance sets of the label, the fault data of the faulty electromechanical system is used as an unknown instance, the unknown instance is input into the binary classifier, and the binary classifier outputs a preliminary determination result of the maintenance method of the faulty electromechanical system, including:

[0030] Step 1: Construct a binary training set D based on the correlation matrix R between instances and labels and the mapping relationship between the clustering results of positive and negative instance sets corresponding to instances and labels. jk , the training set D jk contains the correlation between each pair of labels and instances. For q possible category labels {l1, l2, ..., l q}, generate q(q-1) / 2 binary classifiers through pairwise comparison;

[0031]

[0032]

[0033] Among them, D jk For each instance feature vector x i Binary training set of correlations with the jth label and the kth label; b ij represents the element in row i and column j in matrix R; b ik represents the i-th row and k-th column element in the matrix R; ε(x i ) j Represents the instance vector x corresponding to the jth label corresponding to the positive and negative instance clustering results i Mapping result; Ψ(Y i , l j , l k ) represents the instance x i Relative to the preference value of the jth label and the kth label, Y i represents the relevant label set of the i-th instance. For the training set corresponding to each instance, a binary classifier is induced using the binary learning algorithm B, i.e., g jk ←B(D jk );where g jk Represented by the training set D jk The mapping relationship from the distance set to the label pair correlation obtained by training, B is any simple binary classification learner;

[0034] Step 2: Each instance will participate in the learning process of the binary classifier, and the classification learning results are used as the voting basis of the voting method: for any instance x i ∈X, if g jk >0 then the system votes for l j , otherwise vote for l k , taking the fault data of the faulty electromechanical system as the unknown instance x u , the unknown instance x u Input to the binary classifier;

[0035] The binary classifier outputs the unknown instance x u The overall vote for each possible class label:

[0036]

[0037] Where f(x u , l j ) represents an unknown instance x u Possible class labels lj The voting results, Denotes the binary training set D kj The corresponding learning result vote, if g kj (x u )≤0 holds, then is 1, otherwise it is 0, and f(x u , l j ) as the preliminary determination result of the maintenance method for the faulty electromechanical system.

[0038] Preferably, the introduction of the calibration tag and the use of the fault underlying features and the calibration tag to update the preliminary determination result of the maintenance method of the faulty electromechanical system to obtain the final maintenance method determination result include:

[0039] Step 1: In the framework of paired comparison, according to the number of votes in the preliminary maintenance method voting, sort the label set from large to small to obtain a preliminary sorted label set, and add a virtual label l v As calibration labels, add virtual labels to each x u The corresponding training results are in the sorted label set;

[0040] Step 2: After adding the calibration labels, q new label pairs are formed (l j , l v ), and induces q auxiliary binary classifiers, label pairs (l j , l v ) The corresponding binary training set is constructed as follows:

[0041]

[0042] D jv ={(ε(x u ) j ,Φ(Y i , l j , l v ))|1≤i≤m}

[0043] D jv For each instance x i Correlation between the jth label and the calibration label of the binary training set, Φ(Y i , l j , l v ) represents the instance x i Relative to the relevant value of the jth label, ε(x u ) j Represents an unknown instance x uThe distance set mapping to the cluster center of the positive and negative instance sets corresponding to the jth label is the relationship between the unknown instance and the underlying features of the jth fault type. For each new training set, the binary classifier is induced using the binary learning algorithm B: jv ←B(D jv ). jv Represented by the training set D jv The mapping relationship from the distance set to the label pair correlation obtained by training, B can be any simple binary classification learner;

[0044] Step 3: Based on the classification learning results of the binary classifier, the unknown instance x u Possible class labels l j The voting results are updated as follows:

[0045]

[0046] in Denotes the binary training set D jv Voting on the learning results, if g jv (x u )>0 holds, then is 1;

[0047] Step 4: According to D jv The corresponding classification learner learning result, each unknown instance x u The corresponding virtual label l v Total voting results:

[0048]

[0049] Among them, f ′ (x u , l v ) is the updated voting result.

[0050] Step 5: According to the voting results of Step 4, the instance x to be determined u The corresponding maintenance method is determined as follows:

[0051] Y u = {l j |f′(x u , l j )>f′(x u , l v )|

[0052] Among them, Y u Indicates the final maintenance method determination result.

[0053] It can be seen from the technical solutions provided by the above-mentioned embodiments of the present invention that the method provided by the present invention can effectively determine the maintenance mode based on the data collected from the complex electromechanical system with high accuracy.

[0054] Additional aspects and advantages of the present invention will be given in part in the following description, which will become obvious from the following description, or may be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0056] Figure 1 A processing flow chart of a method for predicting maintenance methods of an electromechanical system based on underlying fault characteristics provided by an embodiment of the present invention.

[0057] Figure 2 A data collection terminal is provided in accordance with an embodiment of the present invention.

[0058] Figure 3 A number of cluster centers provided by an embodiment of the present invention.

[0059] Figure 4 This is the mapping result from the first instance to the first label of the present invention.

[0060] Figure 5 A preliminary maintenance mode voting result provided by an embodiment of the present invention.

[0061] Figure 6 A final maintenance method determination result provided by an embodiment of the present invention.

[0062] Figure 7 This is a comparison of the improved CLR provided by the present invention and the CLR before improvement in terms of various evaluation indicators.

[0063] Figure 8 This is a comparison of the improved CLR provided by the present invention and other algorithms in terms of various evaluation indicators. DETAILED DESCRIPTION

[0064] The embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be interpreted as limiting the present invention.

[0065] It will be understood by those skilled in the art that, unless expressly stated, the singular forms "one", "said", and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we refer to an element as being "connected" or "coupled" to another element, it may be directly connected or coupled to the other element, or there may be intermediate elements. In addition, the "connection" or "coupling" used herein may include wireless connection or coupling. The term "and / or" used herein includes any unit and all combinations of one or more associated listed items.

[0066] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as those generally understood by those skilled in the art in the art to which the present invention belongs. It should also be understood that terms such as those defined in common dictionaries should be understood to have meanings consistent with the meanings in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless defined as herein.

[0067] To facilitate understanding of the embodiments of the present invention, several specific embodiments will be further explained below with reference to the accompanying drawings, and each embodiment does not constitute a limitation on the embodiments of the present invention.

[0068] The processing flow of a method for predicting maintenance methods of an electromechanical system based on underlying fault characteristics provided by an embodiment of the present invention is as follows: Figure 1 As shown, the processing steps include the following:

[0069] Step S1: Based on the maintenance records of complex electromechanical systems, taking fault data as instances and fault types as labels, a correlation matrix R between instances and labels is constructed, and a clustering set of each label is obtained using an improved clustering method. The Euclidean distance between the instance and the cluster center is calculated to obtain a mapping relationship between the instance and the clustering results of the positive and negative instance sets corresponding to the label;

[0070] Step S2, constructing a binary classifier to make a preliminary determination of the maintenance method of the data to be determined;

[0071] Step S3: introduce the calibration tag and use the underlying fault characteristics to update the preliminary determination result to obtain the final maintenance determination result.

[0072] When constructing a binary training set, the method of the present invention uses the mapping of the underlying feature set of labels obtained from the positive and negative instance sets corresponding to the instances to the labels as a part of the training set.

[0073] Specifically, the above step S1 includes: Figure 2 A schematic diagram of a data acquisition terminal for a complex electromechanical system of a train traction system provided by an embodiment of the present invention is provided. According to the maintenance record of the complex electromechanical system, an improved clustering method is used to obtain a cluster set, and the Euclidean distance from each cluster set to the cluster center is calculated. The specific processing process includes:

[0074] Step 1: This invention takes each fault data as an instance, each fault type as a label, and the specific parameters corresponding to the fault type as eigenvalues, which include specific values ​​such as breakdown voltage. Construct the feature matrix F between the instance and the eigenvalue and the correlation matrix R between the instance and the label:

[0075]

[0076]

[0077] Among them, m represents the number of all instances, n represents the number of all features, q represents the number of all labels, and a ij (1≤i≤m, 1≤j≤n) is the specific characteristic parameter of this instance; b ij (1≤i≤m, 1≤j≤q) is -1 or 1, -1 indicates that the instance does not have a corresponding label, and 1 indicates that the instance has a corresponding label.

[0078] Step 2: According to the correlation between instances and labels in the correlation matrix R between instances and labels, the instance set is divided into a positive instance set P and a negative instance set N corresponding to each label. Specifically, for each instance corresponding to a label, if it is related to the label, it is added to the positive instance set of the label, otherwise it is added to the negative instance set.

[0079] Step 3: Calculate the label entropy E(k) for the positive instance set P and negative instance set N corresponding to each label. p 、E(k) n :

[0080]

[0081]

[0082] Among them, E(k) p represents the label entropy of the positive instance set corresponding to the kth label; hp(j) represents the ratio of the jth label vector appearing in the positive instance set corresponding to the label to the total label vector appearing in the positive instance set; E(k) n and hn(j) represent the label entropy and vector ratio corresponding to the negative instance set of the kth label respectively.

[0083] Step 4: The basic entropy of each label can be calculated using the label entropy of the positive and negative instance sets, and the minimum change of the label can be evaluated using the basic entropy:

[0084]

[0085] w k represents the basic entropy of the k-th label.

[0086] Step 5: The number of clusters adaptively adapted to the training data can be obtained through the above parameter calculation:

[0087]

[0088] Among them, c k represents the number of clusters corresponding to the positive and negative instance sets of the kth label; s is the smoothing parameter, which is set to 1 here to make it Laplace smoothing; |P| and |N| represent the cardinality of the positive and negative instance sets respectively. On the basis of a fixed number of clusters, the positive and negative instance sets corresponding to each label are clustered with the same number of clusters c. k Perform k-means clustering separately.

[0089] Step 6: The clustering results can reflect the relationship between instances and the deep characteristics of the labels, and the cluster center is an intuitive reflection of the clustering results. According to the clustering results, each label can be constructed from n dimensions to 2c k Dimensional mapping:

[0090]

[0091] ε(x) k Represents the mapping between the instance and the clustering result set corresponding to the j-th label, which is the set of positive and negative instances respectively; is the Euclidean distance between the instance and the cluster centers of the positive instance set of the label, is the Euclidean distance between the instance and the cluster centers of the negative instance set of the label. The mapping includes processing the relevant instance set and the irrelevant instance set of the label separately to obtain different label feature results. The difference in clustering results between the relevant instance set and the irrelevant instance set can represent the impact of the difference in the relevance between the instance and the label on the underlying features of the label. The combination of the two will make it easier to train and distinguish the underlying correlation between the instance and the label.

[0092] The number of cluster centers c calculated after the above steps k like Figure 3 As shown, the mapping result from the first instance to the first label is as follows Figure 4 shown.

[0093] Specifically, the above step S2 includes:

[0094] The mapping relationship ε(x) between the instance finally obtained in step S1 and the clustering result of the positive and negative instance sets corresponding to the labels k On this basis, the mapping relationship is used as a component of the binary training set construction.

[0095] Step 1: Construct a binary training set D based on the mapping relationship between the correlation matrix R between instances and labels and the clustering results of the positive and negative instance sets corresponding to the instances and labels. jk , the training set D jk contains the correlation between each pair of labels and instances. For q possible category labels {l1, l2..,., l q}, a total of q(q-1) / 2 binary classifiers can be generated through pairwise comparisons.

[0096]

[0097] D jk ={(ε(x i ) j ,Ψ(Y i , l j , l k ))}

[0098] Among them, D jk For each instance feature vector x i Binary training set of correlations with the jth label and the kth label; b ij represents the element in row i and column j in matrix R; b ik represents the i-th row and k-th column element in the matrix R; ε(x i ) j Represents the instance vector x corresponding to the jth label positive and negative instance clustering result i Mapping result; Ψ(Y i , l j , l k ) represents the instance x i Relative to the preference value of the jth label and the kth label, Y i represents the relevant label set of the i-th instance. For the training set corresponding to each instance, a binary classifier can be induced using the binary learning algorithm B, that is, g jk ←B(D jk ). Where g jk Represented by the training set D jk The mapping relationship from the distance set to the label pair correlation obtained by training, B can be any simple binary classification learner.

[0099] Step 2: Each instance will participate in the learning process of the binary classifier, and the classification learning results are used as the voting basis of the voting method: for any instance xi ∈X, if g jk >0, the system votes for l j , otherwise vote for l k , taking the fault data of the faulty electromechanical system as the unknown instance x u , the unknown instance x u Input to the binary classifier;

[0100] The binary classifier outputs the unknown instance x u The overall vote for each possible class label:

[0101]

[0102] Where f(x u , l j ) represents an unknown instance x u Possible class labels l j The voting results, Denotes the binary training set D kj The corresponding learning result vote, if g kj (x u )≤0 holds, then is 1, otherwise it is 0. u , l j ) as the preliminary determination result of the maintenance method for the faulty electromechanical system.

[0103] Taking 10 maintenance judgment results as an example, the preliminary maintenance mode voting results that can be obtained in Example 2 are as follows: Figure 5 shown.

[0104] Specifically, the above step S3 includes: the present invention provides a method of introducing a calibration tag to update the determination result using the underlying fault characteristics to obtain a final maintenance determination result, and the specific processing steps are as follows:

[0105] Step 1: In the framework of paired comparison, according to the number of votes in the preliminary maintenance method voting, sort the label set from large to small to obtain a preliminary sorted label set, and add a virtual label l v As calibration labels. Add virtual labels to each x u The corresponding training results are in the sorted label set. Among them, the calibration label l v As a split point between relevant and irrelevant labels: all relevant labels take precedence over l v , and l v It takes precedence over all irrelevant tags.

[0106] Step 2: After adding the calibration labels, q new label pairs are formed (l j , l v), and induces q auxiliary binary classifiers, similar to the binary training set corresponding to the original label pair, label pair (l j , l v ) The corresponding binary training set can be constructed as follows:

[0107]

[0108] D jv ={(ε(x u ) j ,Φ(Y i , l j , l v ))|1≤i≤m}

[0109] D jv For each instance x i Correlation between the jth label and the calibration label of the binary training set, Φ(Y i , l j , l v ) represents the instance x i Relative to the relevant value of the jth label, ε(x u ) j Represents an unknown instance x u The mapping of the distance set to the cluster center of the positive and negative instance sets corresponding to the jth label, that is, the relationship between the unknown instance and the underlying features of the jth fault type. For each new training set, the binary classifier is induced using the binary learning algorithm B: jv ←B(D jv ). jv Represented by the training set D jv The mapping relationship from the distance set to the label pair correlation obtained by training, B can be any simple binary classification learner;

[0110] Step 3: Based on the classification learning results of the binary classifier, the unknown instance x in Step 2 is u Possible class labels l j The voting results are updated as follows:

[0111]

[0112] in Denotes the binary training set D jv Voting on the learning results, if g jv (x u )>0 holds, then is 1.

[0113] Step 4: According to D jv The corresponding classification learner learning result, each unknown instance x uThe corresponding virtual label l v Total voting results:

[0114]

[0115] Among them, f ′ (x u , l v ) is the updated voting result.

[0116] Step 5: According to the voting results of Step 4, the instance x to be determined u The corresponding maintenance method is determined as follows:

[0117] Y u = {l j |f′(x u , l j )>f′(x u , l v )}

[0118] Among them, Y u Indicates the final maintenance method determination result.

[0119] Similarly, taking the same 10 maintenance determination results in Example 2 as an example, the final maintenance method determination that can be obtained in Example 3 is as follows: Figure 6 shown.

[0120] Using 70% of the data as a training set and 30% of the data as a test set, it can be found that the detection method provided by the present invention can accurately determine the maintenance method of a complex electromechanical system. Figure 7 The improved CLR provided by the present invention is compared with the CLR before improvement in terms of average precision (Average Precision), coverage (Coverage), Hamming loss (Hamming Loss), single error (One Error) and rank loss (Ranking Loss). Figure 8 This is the comparison between the improved CLR and other algorithms in the above evaluation indicators.

[0121] An embodiment of the present invention also provides a non-transitory computer-readable storage medium, which is used to store computer instructions. When the computer instructions are executed by a processor, a complex electromechanical system maintenance determination method, system and medium based on the underlying characteristics of the fault are implemented.

[0122] An embodiment of the present invention also provides a computer program (product), including a computer program, which, when running on one or more processors, is used to implement a complex electromechanical system maintenance determination method, system and medium based on underlying fault characteristics.

[0123] An embodiment of the present invention also provides an electronic device, comprising: a data acquisition terminal, a processor, a memory and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory, so that the electronic device executes a complex electromechanical system maintenance judgment method, system and medium based on the underlying characteristics of the fault.

[0124] To sum up, the embodiment of the present invention uses an improved calibrated label ranking CLR (Calibrated Label Ranking) algorithm combined with relevant operation and maintenance data of complex electromechanical systems to determine the maintenance method, and compares it with the algorithm before improvement and other various determination algorithms to verify the effectiveness of the present invention, and designs a corresponding maintenance method determination system to achieve rapid determination in actual work and effectively guide the actual maintenance work of complex electrical systems.

[0125] Those skilled in the art can understand that the accompanying drawings are only schematic diagrams of an embodiment, and the modules or processes in the accompanying drawings are not necessarily required to implement the present invention.

[0126] It can be known from the description of the above implementation methods that those skilled in the art can clearly understand that the present invention can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which can be stored in a storage medium such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in the various embodiments of the present invention or certain parts of the embodiments.

[0127] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device or system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. The device and system embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0128] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for predicting maintenance methods of electromechanical systems based on underlying fault characteristics, characterized in that: include: Based on the maintenance records of electromechanical systems, fault data is used as an instance and fault type is used as a label. The correlation matrix between instances and labels is constructed, all labels are clustered, and the Euclidean distance between the instance and the cluster center of the positive and negative instance sets of the label is calculated to obtain the mapping relationship between the instance and the positive and negative instance sets of the label. A binary classifier is constructed according to the correlation matrix between the instance and the label and the mapping relationship between the instance and the positive and negative instance sets of the label, the fault data of the faulty electromechanical system is used as an unknown instance, the unknown instance is input into the binary classifier, and the binary classifier outputs a preliminary determination result of the maintenance method of the faulty electromechanical system; Introducing a calibration tag, using the underlying fault features and the calibration tag to update the preliminary determination result of the maintenance method of the faulty electromechanical system, and obtaining a final determination result of the maintenance method; The maintenance record based on the electromechanical system takes fault data as an instance and fault type as a label, constructs a correlation matrix between the instance and the label, performs clustering on all labels, calculates the Euclidean distance between the instance and the cluster center of the positive and negative instance set of the label, and obtains the mapping relationship between the instance and the positive and negative instance set of the label, including: Taking each fault data of the faulty electromechanical system as an instance, each fault type as a label, and the specific parameters corresponding to the fault type as eigenvalues, the feature matrix F between the instance and the eigenvalue and the correlation matrix R between the instance and the label are constructed: Among them, m represents the number of all instances, n represents the number of all features, q represents the number of all labels, and a ij is the specific characteristic parameter of this instance, 1≤i≤m,1≤j≤n, b ij It is -1 or 1, -1 means that the instance does not have the corresponding label, 1 means that the instance has the corresponding label, 1≤i≤m,1≤j≤q; Step 2: For each instance corresponding to a label, if it is related to the label, add it to the positive instance set P of the label; otherwise, add it to the negative instance set N; Step 3: Calculate the label entropy E(k) for the positive instance set P and negative instance set N corresponding to each label. p 、E(k) n : Among them, E(k) p represents the label entropy of the positive instance set corresponding to the kth label; hp(j) represents the ratio of the jth label vector appearing in the positive instance set corresponding to the label to the total label vector appearing in the positive instance set; E(k) n and hn(j) respectively represent the label entropy and vector ratio corresponding to the negative instance set of the kth label; Step 4: Use the label entropy of the positive and negative instance sets to calculate the basic entropy of each label, and use the basic entropy to evaluate the minimum change of the label: w k represents the basic entropy of the kth label; Step 5: Calculate the number of clusters adaptively based on the training data: Among them, c k represents the number of clusters corresponding to the positive and negative instance sets of the kth label, s is the smoothing parameter; |P| and |N| represent the cardinality of the positive and negative instance sets respectively, and the same number of clusters c is used for the positive and negative instance sets corresponding to each label. k Perform k-means clustering separately; Step 6: Construct a vector from n to 2c for each label based on the clustering results. k Dimensional mapping: ε(x) k Represents a mapping constructed from the set of clustering results of the instance and the set of positive and negative instances corresponding to the jth label; is the Euclidean distance between the instance and the cluster centers of the positive instance set of the label, is the Euclidean distance between the instance and each cluster center of the negative instance set of the label, and the mapping includes processing the relevant instance set and the irrelevant instance set of the label respectively to obtain different label feature results; The binary classifier is constructed according to the correlation matrix between the instance and the label and the mapping relationship between the instance and the positive and negative instance sets of the label, the fault data of the faulty electromechanical system is used as an unknown instance, and the unknown instance is input into the binary classifier, and the binary classifier outputs a preliminary determination result of the maintenance method of the faulty electromechanical system, including: Step 1: Construct a binary training set D based on the correlation matrix R between instances and labels and the mapping relationship between the clustering results of positive and negative instance sets corresponding to instances and labels. jk , the training set D jk contains the correlation between each pair of labels and instances. For q possible category labels {l1,l2,…,l q }, generate q(q―1) / 2 binary classifiers through pairwise comparison; D jk ={(ε(x i ) j ,Ψ(Y i ,l j ,l k ))} Among them, D jk For each instance feature vector x i Binary training set of correlations with the jth label and the kth label; b ij represents the element in row i and column j in matrix R; b ik represents the i-th row and k-th column element in the matrix R; ε(x i ) j Represents the instance vector x corresponding to the jth label corresponding to the positive and negative instance clustering results i Mapping result; Ψ(Y i ,l j ,l k ) represents the instance x i Relative to the preference value of the jth label and the kth label, Y i represents the relevant label set of the i-th instance. For the training set corresponding to each instance, a binary classifier is induced using the binary learning algorithm B, i.e., g jk ←B(D jk );where g jk Represented by the training set D jk The mapping relationship from the distance set to the label pair correlation obtained by training, B is any simple binary classification learner; Step 2: Each instance will participate in the learning process of the binary classifier, and the classification learning results are used as the voting basis of the voting method: for any instance x i ∈X, if q jk >0 then the system votes for l j , otherwise vote for l k , taking the fault data of the faulty electromechanical system as the unknown instance x u , the unknown instance x u Input to the binary classifier; The binary classifier outputs the unknown instance x u The overall vote for each possible class label: Where f(x u ,l j ) represents an unknown instance x u Possible class labels l j The voting results, Denotes the binary training set D kj The corresponding learning result vote, if g kj (x u )≤0 holds, then is 1, otherwise it is 0, and f(x u ,l j ) as the preliminary determination result of the maintenance method for the faulty electromechanical system.

2. The method according to claim 1, characterized in that The introduction of the calibration tag and the use of the fault bottom layer characteristics and the calibration tag to update the preliminary determination result of the maintenance method of the faulty electromechanical system to obtain the final maintenance method determination result include: Step 1: In the framework of paired comparison, according to the number of votes in the preliminary maintenance method voting, sort the label set from large to small to obtain a preliminary sorted label set, and add a virtual label l v As calibration labels, add virtual labels to each x u The corresponding training results are in the sorted label set; Step 2: After adding the calibration labels, q new label pairs are formed (l j ,l v ), and induces q auxiliary binary classifiers, label pairs (l j ,l v ) The corresponding binary training set is constructed as follows: D jv ={(ε(x u ) j ,Φ(Y i ,l j ,l v ))│1≤i≤m} D jv For each instance x i Correlation between the jth label and the calibration label of the binary training set, Φ(Y i ,l j ,l v ) represents the instance x i Relative to the relevant value of the jth label, ε(x u ) j Represents an unknown instance x u The distance set mapping to the cluster center of the positive and negative instance sets corresponding to the jth label, that is, the relationship between the unknown instance and the underlying features of the jth fault type. For each new training set, the binary learning algorithm B is used to induce a binary classifier: g jv ←B(D jv ), g jv Represented by the training set D jv The mapping relationship from the distance set to the label pair correlation obtained by training, B is any simple binary classification learner; Step 3: Based on the classification learning results of the binary classifier, the unknown instance x u Possible class labels l j The voting results are updated as follows: in Denotes the binary training set D jv Voting on the learning results, if g jv (x u )>0 holds, then is 1; Step 4: According to D jv The corresponding classification learner learning result, each unknown instance x u The corresponding virtual label l v Total voting results: Among them, f'(x u ,l v ) is the updated voting result; Step 5: According to the voting results of Step 4, the instance x to be determined u The corresponding maintenance method is determined as follows: Y u ={l j |f′(x u ,l j )>f′(x u ,l v )} Among them, Y u Indicates the final maintenance method determination result.

Images