A method for predicting failure of complex electromechanical equipment, electronic equipment and storage medium

Abstract

The present invention provides a method, electronic device, and storage medium for predicting complex electromechanical equipment faults. The method comprises: processing accumulated historical equipment fault text data, constructing positive and negative instance sets, and calculating the number of fault label clusters based on fault type; clustering the positive and negative instance sets based on the fault label clustering results, calculating the correlations between labels, constructing an instance similarity matrix, and updating the instance similarity matrix using the label similarity matrix; constructing a matrix of similar instances and a matrix of non-similar instances based on the updated instance similarity matrix, calculating the corresponding weight for each instance, and obtaining closely related labels for each label; training a classification training model using the instance data, constructing unknown instances based on equipment performance, and predicting faults in the unknown instances using the trained classification training model. The present invention predicts equipment faults based on abnormal phenomena occurring in the equipment by mining fault association rules from historical fault text data.

Description

Technical Field

[0001] The present invention relates to the technical field of electromechanical equipment fault prediction, and in particular to a complex electromechanical equipment fault prediction method, electronic equipment, and storage medium. Background Art

[0002] With the development of my country's economy and engineering technology, the design and manufacturing of electromechanical systems are steadily moving towards precision. Due to the complex internal structures of modern large-scale electromechanical equipment, failures in key subsystems, such as control systems and power systems, are frequent, difficult to monitor, and often accidental. Complex electromechanical equipment also accumulates a vast amount of textual data on faults. Mining this historical textual data and analyzing the essential relationship between equipment performance and potential failures can help us understand the subtleties and proactively predict potential failures, providing greater safety for complex electromechanical equipment.

[0003] The essence of data-driven fault prediction for complex electromechanical equipment lies in converting high-dimensional feature vectors into state identifiers. Given the difficulty of accurately modeling complex electromechanical equipment, the method of constructing precise mathematical models based on sets of device state observation parameters for fault prediction has significant limitations in practical applications. This invention, however, uses big data analysis technology to combine labels with actual fault representations, enabling fault prediction for complex electromechanical equipment. Summary of the Invention

[0004] The implementation of the present invention provides a complex electromechanical equipment fault prediction method, electronic equipment and storage medium, so as to effectively predict possible faults of complex electromechanical systems.

[0005] In order to achieve the above-mentioned purpose, the present invention adopts the following technical solutions.

[0006] A method for predicting faults of complex electromechanical equipment, comprising:

[0007] Process the accumulated historical fault text data of the equipment, construct positive and negative instance sets, and calculate the number of fault label clusters based on the fault type;

[0008] Clustering the positive and negative instance sets according to the fault label clustering results, constructing an instance similarity matrix, calculating the correlation between labels, and updating the instance similarity matrix through the label similarity matrix;

[0009] Based on the updated instance similarity matrix, a matrix of similar instances and a matrix of non-similar instances are constructed, and the corresponding weight of each instance is calculated to obtain closely related labels for each label.

[0010] A classification training model is constructed using the similar instance matrix and the non-similar instance matrix, the classification training model is trained using instance data, unknown instances are constructed according to device performance, and the trained classification training model is used to predict failures of the unknown instances.

[0011] Preferably, the aforementioned processing of the accumulated historical fault text data of complex electromechanical equipment, constructing a positive and negative instance set, and calculating the number of fault label clusters according to the fault type includes:

[0012] Step 1: Sort out the historical text data of electromechanical equipment, use instances to represent each fault data, use labels to represent each fault type, and construct the relationship matrix E between instances and labels:

[0013]

[0014] Among them, m represents the number of all instances, n represents the number of all labels, and a ij (1≤i≤m,1≤j≤m) is 0 or 1, 0 means that the instance does not have the corresponding label, and 1 means that the instance has the corresponding label;

[0015] Step 2: Divide the instance set into a positive instance set P and a negative instance set N, where P represents the set of instances with labels under each label, and N represents the set of instances without labels under each label;

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

[0017]

[0018]

[0019] Among them, E(k p ) represents the label entropy of the kth label of the positive instance set P, where h kp (j) represents the ratio of the jth label vector in P corresponding to the kth label to the total label vectors in P, E(k n ) represents the label entropy of the kth label of the negative instance set N, where h kn (j) represents the ratio of the jth label vector appearing in N corresponding to the kth label to the total label vectors appearing in N;

[0020] Step 4: Calculate the basic entropy w of the kth label k :

[0021]

[0022] Among them, w k represents the basic entropy of the kth label, minE(k) represents the label entropy E(k) corresponding to the kth label p ) and E(k n ) whichever is smaller;

[0023] Step 5: Calculate the ratio parameter r of the kth label k :

[0024]

[0025] Among them, r k represents the ratio parameter of the kth label, and s is the ratio smoothing parameter;

[0026] Step 6: Calculate the number of clusters m for the kth label k :

[0027] m k =r k ×min{|P|,|N|}

[0028] Among them, m k Indicates the number of clusters for the k-th label.

[0029] Preferably, clustering the positive and negative instance sets according to the fault label clustering result, constructing an instance similarity matrix, calculating the correlation between labels, and updating the instance similarity matrix using the label similarity matrix includes:

[0030] Step 1: Perform k-means clustering on the positive and negative instance sets P and N respectively, where the number of clusters of positive and negative instances of the kth label is r k ;

[0031] Step 2: Construct the instance similarity matrix MI based on the cluster number of the instance in the clustering result:

[0032]

[0033] Among them, MI kij represents the element in row i and column j of the instance similarity matrix of the kth instance, c k (x i ) represents x i The number of clusters in the set of positive and negative instances corresponding to the kth label, x i represents the i-th instance in E, c k (x j ) represents x j The number of clusters in the set of positive and negative instances corresponding to the kth label, x jrepresents the jth instance in E;

[0034] Step 3: Calculate label similarity MC:

[0035]

[0036] Among them, MC kj represents the label similarity between the kth label and the jth label, γ represents the cosine similarity parameter, l k represents the kth label in E, l j represents the jth label in E;

[0037] Step 4: Update the instance similarity matrix according to the label similarity matrix to obtain MI_new:

[0038]

[0039] Among them, MI_new represents the updated instance similarity matrix, MC kj Indicates the label similarity between the kth label and the jth label, MI j Represents the j-th column element in the original instance similarity matrix MI;

[0040] Step 5: Create a mapping to obtain the properties of a specific tag:

[0041]

[0042] Among them, x i represents the i-th instance in E, FI k (x i ) represents the instance x i The Euclidean distance set to each cluster center of the P and N clustering results corresponding to the k-th label is represented by the function d(·). Represents xi and The Euclidean distance of Indicates the coordinates of the first cluster center of P corresponding to the kth label, Indicates the last cluster center coordinate of P corresponding to the kth label, Indicates the coordinates of the first cluster center of N corresponding to the kth label, Indicates the last cluster center coordinate of N corresponding to the kth label;

[0043] Step 6: Construct label similarity matrix A k :

[0044]

[0045]

[0046] Among them, A k represents the label similarity matrix, a i Represents x i Relative to The value of the label in represents the label matrix consisting of labels other than the label in the kth column, Indicates Y i Relative to label l k The value of l k Belongs to Y i ,but is 1, otherwise 0, Y i Indicates that x i Related related tag sets, express The first tag in express The kth label in .

[0047] Preferably, the method of constructing a matrix of similar instances and a matrix of non-similar instances for complex electromechanical devices based on the updated instance similarity matrix, calculating the corresponding weight of each instance, and obtaining closely related labels for each label includes:

[0048] Step 1: Preprocess the label similarity matrix. For the label in column k, store the instances belonging to the label in D k In D, instances that do not belong to this label are stored in nk middle;

[0049] Step 2: Construct similar instance matrices and non-similar instance matrices based on the updated instance similarity matrix, and calculate the Euclidean distance ranking n between similar instance matrices. r and the Euclidean distance ranking n between matrices of different instances m :

[0050] n r =sort(d(D k ,D k ))

[0051] n m =sort(d(D k ,D nk ))

[0052] Among them, n r Indicates the Euclidean distance ranking between similar instance matrices, n m represents the Euclidean distance ranking between different instance matrices, d represents the Euclidean distance between two matrices, D k Indicates A k The instance matrix of the label belonging to the kth column in D nk Indicates Ak The instance matrix that does not belong to the k-th column label;

[0053] Step 3: Calculate the corresponding weight ω of the b-th instance b :

[0054]

[0055] Among them, ω b Indicates the corresponding weight of the b-th instance, len(n r ) represents the matrix n r the number of rows,

[0056] len(n m ) represents the matrix n m The number of rows of matrix E, k is the number of neighbor labels, length(Dni) represents the number of rows of matrix Dni, m is the number of rows of matrix E, nr i Represents the matrix n r The i-th element in Represents the matrix n m The i-th element in ;

[0057] Step 4: Calculate the kth label l k Closely related tags: k :

[0058]

[0059] Among them, rank(l k ) represents the b The features returned by feature selection, t represents the number of close labels determined, express The kth label in ;

[0060] Preferably, the method of constructing a classification training model using the electromechanical equipment similar instance matrix and the non-similar instance matrix, training the classification training model using historical instance data of the electromechanical equipment, constructing unknown instances based on the state performance of the equipment under test, and predicting the failure of the unknown instances using the trained classification training model includes:

[0061] Construct a binary training set D:

[0062]

[0063] Among them, D k represents the k-th row element in the binary training set D, x i represents the i-th instance in E, FI k (x i ) represents the instance x iThe set of Euclidean distances to each cluster center of the P and N clustering results corresponding to the k-th label, Indicates Y i Relative to label l j The value of l k Belongs to Y i ,but is 1; otherwise is 0, Y i Indicates that x i Related set of related tags;

[0064] Construct binary training set D kj :

[0065]

[0066] Among them, D kj represents the binary training set, x i represents the i-th instance in E, FI k (xi) represents instance x i The set of Euclidean distances to each cluster center of the P and N clustering results corresponding to the k-th label, Indicates Y i Relative to label l k The value of l k Belongs to Y i ,but is 1, otherwise 0, Y i Indicates that x i Related set of related tags;

[0067] Construct a binary learner as a classification training model, the binary classifier is a support vector machine SVM, and the binary training set D is combined with the binary training set D kj Input to the binary learner, train the binary learner, and the binary learner outputs the mapping function f of the unknown instance i ;

[0068] Construct an unknown instance u based on the performance of the device under test and use the mapping function f i The unknown instance u is input into the mapping function f i , the mapping function f i Output the relevant label prediction result Y corresponding to the unknown instance u u :

[0069]

[0070] Among them, Y u is the relevant label prediction result of the unknown instance u, f i Mapping function constructed for binary learner, FIk (u) represents the Euclidean distance set from instance u to each cluster center of the P and N clustering results corresponding to the kth label.

[0071] A non-transitory computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, the method for predicting complex electromechanical equipment faults is implemented.

[0072] A computer program product includes a computer program, wherein when the computer program is run on one or more processors, the computer program is used to implement the complex electromechanical equipment fault prediction method.

[0073] An electronic device comprises: 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 to enable the electronic device to execute instructions for implementing the complex electromechanical equipment fault prediction method as described above.

[0074] It can be seen from the technical solutions provided by the above-mentioned embodiments of the present invention that the method of the present invention mines the fault association rules of historical fault text data, predicts equipment failures based on abnormal phenomena occurring in the equipment, and provides higher safety for complex electromechanical equipment.

[0075] Additional aspects and advantages of the present invention will be set forth in part in the following description, will become apparent from the following description, or may be learned by practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. 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 any creative work.

[0077] Figure 1 A processing flow chart of a method for predicting complex electromechanical equipment faults provided by an embodiment of the present invention.

[0078] Figure 2 This is a schematic diagram of an MI corresponding to a first tag provided in an embodiment of the present invention.

[0079] Figure 3 A schematic diagram of the tag similarity MC corresponding to a first tag provided in an embodiment of the present invention.

[0080] Figure 4A schematic diagram of an instance similarity matrix MI_new after a first label update provided by an embodiment of the present invention.

[0081] Figure 5 A provided by the embodiment of the present invention k Schematic diagram.

[0082] Figure 6 The embodiment of the present invention provides a first tag corresponding to n m Schematic diagram.

[0083] Figure 7 The embodiment of the present invention provides a first tag corresponding to n r Schematic diagram.

[0084] Figure 8 A schematic diagram of the final weights of each label provided by an embodiment of the present invention.

[0085] Figure 9 A schematic diagram of a binary training set D corresponding to a first label provided in an embodiment of the present invention.

[0086] Figure 10 A binary training set D constructed from a set of closely related labels of a first label provided in an embodiment of the present invention kj Schematic diagram. DETAILED DESCRIPTION

[0087] 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 are not to be construed as limiting the present invention.

[0088] It will be understood by those skilled in the art that, unless expressly stated otherwise, the singular forms "a", "an", "said" and "the" used herein may also include the plural forms. It should be further understood that the term "comprising" used in the description 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, "connected" or "coupled" as used herein may include wireless connections or couplings. The term "and / or" used herein includes any unit and all combinations of one or more associated listed items.

[0089] 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 commonly understood by those skilled in the art in the art to which the present invention pertains. It should also be understood that terms such as those defined in common dictionaries should be understood to have meanings consistent with their meanings in the context of the prior art and, unless defined as such herein, will not be interpreted in an idealized or overly formal sense.

[0090] 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.

[0091] In order to solve the problem of large amounts of accumulated fault text data in complex electromechanical equipment, existing fault predictions are unable to effectively utilize historical fault data and deeply explore the relationship between faults and the characteristics of equipment occurrence. The embodiments of the present invention provide a method, device, electronic device and storage medium for predicting faults in complex electromechanical equipment. The method processes the historical fault text data accumulated by the equipment, calculates the number of fault label clusters according to the fault type; constructs a set of positive and negative instances, calculates the correlation between labels, and constructs an instance similarity matrix and a label similarity matrix; uses the instance similarity matrix and the label similarity matrix to construct a classification training model, constructs unknown instances based on the equipment performance, and predicts the faults of the unknown instances through the classification training model.

[0092] The embodiment of the present invention provides a processing flow of a method for predicting a complex electromechanical equipment fault. Figure 1 As shown, the processing steps include the following:

[0093] Step S10: Process the historical fault text data accumulated by the equipment, construct a positive and negative instance set, and calculate the number of fault label clusters according to the fault type.

[0094] Step S20 : ​​clustering the positive and negative instance sets according to the fault label clustering result, calculating the correlation between labels, constructing an instance similarity matrix, and updating the instance similarity matrix using the label similarity matrix.

[0095] Step S30: construct a matrix of similar instances and a matrix of non-similar instances based on the updated instance similarity matrix, calculate the corresponding weight of each instance, and obtain closely related labels for each label.

[0096] Step S40: construct a classification training model using the similar instance matrix and the non-similar instance matrix, train the classification training model using instance data, construct unknown instances based on device performance, and predict faults of the unknown instances using the trained classification training model.

[0097] Furthermore, the above step S10 specifically includes:

[0098] Step 1: Sort out the historical text data of complex electromechanical equipment, use instances to represent each fault data, use labels to represent each fault type, and construct the relationship matrix E between instances and labels:

[0099]

[0100] Among them, m represents the number of all instances, n represents the number of all labels, and a ii (1≤i≤m,1≤j≤m) is 0 or 1, 0 means that the instance does not have the corresponding label, and 1 means that the instance has the corresponding label;

[0101] Step 2: Divide the instance set into a positive instance set P and a negative instance set N, where P represents the set of instances with labels under each label, and N represents the set of instances without labels under each label;

[0102] Step 3: Calculate the label entropy E(k) of the kth label for the positive instance set P and the negative instance set N respectively. p )、E(k n ):

[0103]

[0104]

[0105] Among them, E(k p ) represents the label entropy of the kth label of the positive instance set P, where It represents the ratio of the jth label vector appearing in P corresponding to the kth label to the total label vector appearing in P, E(k n ) represents the label entropy of the kth label of the negative instance set N, where It represents the ratio of the jth label vector appearing in N corresponding to the kth label to the total label vector appearing in N;

[0106] Step 4: Calculate the basic entropy w of the kth label k :

[0107]

[0108] Among them, w k represents the basic entropy of the kth label, minE(k) represents the label entropy E(k) corresponding to the kth label p ) and E(k n ) whichever is smaller;

[0109] Step 5: Calculate the ratio parameter r of the kth label k :

[0110]

[0111] Among them, r k represents the ratio parameter of the kth label, and s is the ratio smoothing parameter;

[0112] Step 6: Calculate the number of clusters m for the kth label k :

[0113] m k =r k ×min{|P|,|N|}

[0114] Among them, m k Indicates the number of clusters for the k-th label.

[0115] In combination with the above implementation process, based on historical text data, the present invention takes some publicly available high-speed rail accident data as an example, selects 20 instances * 20 labels to construct E as shown in Table 1, and selects one data item for display. One of the data items is that a train discovered that due to the insensitive movement of the small cylinder axis of the toilet device, the toilet bowl cover was not tight, the odor was seriously backflowed, the odor in the carriage was strong, the toilet seat PLC (Programmable Logic Controller) was burned, and the fault could not be reset. After conversion, the format is:

[0116] [0,0,1,0,0,1,1,0,0,0,0,0,1,1,0,0,0,1,0,0,]

[0117] Table 1 E constructed from some high-speed rail fault data sets

[0118]

[0119]

[0120] Calculate the label entropy E(1 p )=2.926418,E(1 n )=2.995732, the basic entropy of the first label w1=0.494148, the ratio parameter r1 of the first label =0.498049, the number of clusters m1 of the first label =5, and the other labels can be calculated similarly.

[0121] Furthermore, the above step S20 specifically includes:

[0122] Step 1: Perform k-means clustering on the positive and negative instance sets P and N respectively, where the number of clusters of positive and negative instances of the kth label is r k ;

[0123] Step 2: Construct the instance similarity matrix MI based on the cluster number of the instance in the clustering result:

[0124]

[0125] Among them, MI kij represents the element in row i and column j of the instance similarity matrix of the kth instance, c k (x i ) represents x i The number of clusters in the set of positive and negative instances corresponding to the kth label, x i represents the i-th instance in E, c k (x j ) represents x j The number of clusters in the set of positive and negative instances corresponding to the kth label, x j represents the jth instance in E;

[0126] Step 3: Calculate label similarity MC:

[0127]

[0128] Among them, MC kj represents the label similarity between the kth label and the jth label, γ represents the cosine similarity parameter, l k represents the kth label in E, l j represents the jth label in E;

[0129] Step 4: Update the instance similarity matrix according to the label similarity matrix to obtain MI_new:

[0130]

[0131] Among them, MI_new represents the updated instance similarity matrix, MC kj Indicates the label similarity between the kth label and the jth label, MI j Represents the j-th column element in the original instance similarity matrix MI;

[0132] Step 5: Create a mapping to obtain the properties of a specific tag:

[0133]

[0134] Among them, x i represents the i-th instance in E, FI k (x i ) represents the instance x i The Euclidean distance set to each cluster center of the P and N clustering results corresponding to the k-th label is represented by the function d(·). Represents xi and The Euclidean distance, Indicates the coordinates of the first cluster center of P corresponding to the kth label, Indicates the last cluster center coordinate of P corresponding to the kth label, Indicates the coordinates of the first cluster center of N corresponding to the kth label, Indicates the last cluster center coordinate of N corresponding to the kth label;

[0135] Step 6: Construct label similarity matrix A k :

[0136]

[0137]

[0138] Among them, A k represents the label similarity matrix, a i Represents x i Relative to The value of the label in represents the label matrix consisting of labels other than the label in the kth column, Indicates Y i Relative to label l k The value of l k Belongs to Y i ,but is 1, otherwise 0, Y i Indicates that x i Related related tag sets, express The first tag in express The kth label in .

[0139] Taking the high-speed rail fault data in Example 1 as an example, the MI corresponding to the first tag is as shown in the attached Figure 2 As shown, the label similarity MC corresponding to the first label is as follows Figure 3 As shown, the instance similarity matrix MI_new after the first label update is as follows Figure 4 As shown, the mapping of the first label of the first instance constructed is shown in Table 2:

[0140] Table 2 The distance from the first instance to each cluster center under the first label

[0141]

[0142] The constructed A k like Figure 5 The construction of other instances and labels is similar.

[0143] The instance similarity matrix is ​​constructed based on the k-means clustering results, measuring the similarity between instances. It indicates the similarity between instances in the clustering results, that is, whether they belong to the same cluster. The label similarity matrix is ​​constructed based on cosine similarity, measuring the similarity between labels. The instance similarity matrix is ​​updated using the label similarity matrix, and the updated instance similarity matrix can be directly used in the spectral clustering algorithm, which can further explore the properties of the matrix data and improve the stability of the final results based on the k-means algorithm.

[0144] Furthermore, the above step S30 specifically includes:

[0145] This embodiment provides a method for constructing a matrix of similar instances and a matrix of non-similar instances, calculating the corresponding weight of each instance, and obtaining closely related labels for each label.

[0146] Step 1: Preprocess the label similarity matrix. For the kth column label, store the instances belonging to the label in D k In D, instances that do not belong to this label are stored in nk middle;

[0147] Step 2: Construct the similar instance matrix and the non-similar instance matrix based on the updated instance similarity matrix.

[0148] The closely related labels for each label are used to create a mapping, capturing the properties of that specific label. The kth row in the instance-related label set indicates which t labels are most closely related to the kth label, sorted by correlation from highest to lowest. The instance-related label set is used to construct the closely related label set for each label. The properties of the related labels in the label set are then used to construct the binary training set.

[0149] Calculate the Euclidean distance ranking n between similar instance matrices r and the Euclidean distance ranking n between matrices of different instances m :

[0150] n r =sort(d(D k ,D k ))

[0151] n m =sort(d(D k ,D nk ))

[0152] Among them, n r Indicates the Euclidean distance ranking between similar instance matrices, n mrepresents the Euclidean distance ranking between different instance matrices, d represents the Euclidean distance between two matrices, D k Indicates A k The instance matrix of the label belonging to the kth column in D nk Indicates A k The instance matrix that does not belong to the k-th column label;

[0153] Step 3: Calculate the corresponding weight ω of the bth instance b :

[0154]

[0155] Among them, ω b Indicates the corresponding weight of the b-th instance, len(n r ) represents the matrix n r The number of rows, len(n m ) represents the matrix n m The number of rows of matrix E, k is the number of neighbor labels, length(Dni) represents the number of rows of matrix Dni, m is the number of rows of matrix E, nr i Represents the matrix n r The i-th element in Represents the matrix n m The i-th element in;

[0156] Step 4: Calculate the kth label l k Closely related tags: k :

[0157]

[0158] Among them, rank(l k ) represents the b The features returned by feature selection, t represents the number of close labels determined, express The kth label in .

[0159] According to the results of Example 1 and Example 2, the n corresponding to the first label m like Figure 6 As shown, the first label corresponds to n r like Figure 7 As shown, the final weight of each label is as follows Figure 8 As shown, the closely related tags of the first tag are tag 18, tag 11 and tag 2.

[0160] Furthermore, the above step S40 specifically includes:

[0161] This embodiment provides a method for constructing a classification training model, constructing unknown instances based on device performance, and predicting failures of the unknown instances through classification training.

[0162] Step 1: Construct binary training set D:

[0163]

[0164] Among them, D k represents the k-th row element in the binary training set D, x i represents the i-th instance in E, FI k (x i ) represents the instance x i The set of Euclidean distances to each cluster center of the P and N clustering results corresponding to the k-th label, Indicates Y i Relative to label l k The value of l k Belongs to Y i ,but is 1, otherwise 0, Y i Indicates that x i Related set of related tags;

[0165] Step 2: Construct binary training set D kj :

[0166]

[0167] Among them, D kj represents the binary training set, x i represents the i-th instance in E, FI k (xi) represents instance x i The set of Euclidean distances to each cluster center of the P and N clustering results corresponding to the k-th label, Indicates Y i Relative to label l k The value of l k Belongs to Y i ,but is 1, otherwise 0, Y i Indicates that i Related set of related tags;

[0168] Step 3: Predict unknown instances based on the mapping function generated by the binary learner:

[0169] Construct a binary learner as a classification training model. Considering the characteristics of the input training data set with high structural feature dimension, small sample size and linear classification, the support vector machine (SVM) is selected as the binary classifier. Its structure includes two parts: training and prediction. That is, training the data with known classification to obtain classification rules and classifying and predicting the input data according to the training results. k (x i )and The constructed binary training set, where FI k (x i ) is built based on MI_new.

[0170] The binary training set D is combined with the binary training set D kj Input to the binary learner, train the binary learner, and the binary learner outputs the mapping function f of the unknown instance i ;

[0171] Input unknown instance u to get the related label prediction result Y u :

[0172]

[0173] Among them, Y u is the relevant label prediction result of the unknown instance u, f i Mapping function constructed for binary learner, FI k (u) represents the Euclidean distance set from instance u to each cluster center of the P and N clustering results corresponding to the kth label.

[0174] Combining examples 1, 2, and 3, the binary training set D corresponding to the first label is constructed as follows: Figure 9 As shown, the binary training set D is constructed from the set of closely related labels of the first label. kj For example Figure 10 As shown in the figure, the input unknown instance u is [1,1,?,0,1,1,1,1,0,1,?,0,1,1,0,1,0,1,1,0], and the known part of u is compared with the original data. The maximum probability of the complete data of the unknown instance u is [1,1,0,0,1,1,1,1,0,1,1,0,1,1,0,1,1,0], and the final prediction results are that the most likely fault types are the fault types corresponding to label 3, label 7, label 9, label 14, label 1, label 2, label 8, label 11, label 12, label 16, and label 17, from large to small.

[0175] In summary, the embodiments of the present invention propose a method, device, electronic device and storage medium for predicting faults of complex electromechanical equipment, which can effectively utilize historical fault descriptions, classify faults according to historical fault data, describe equipment phenomena, and realize fault prediction of complex electromechanical equipment. The method provided by the present invention can more clearly show the possibility of faults caused by equipment phenomena, prevent problems before they occur, and provide higher safety for complex electromechanical equipment.

[0176] Those skilled in the art will appreciate that the accompanying drawings are merely schematic diagrams of an embodiment, and the modules or processes in the accompanying drawings are not necessarily required to implement the present invention.

[0177] From the above description of the embodiments, it can be seen that those skilled in the art can clearly understand that the present invention can be implemented by means of software plus the necessary general-purpose hardware platform. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium such as ROM / RAM, a magnetic disk, or an optical disk, 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 various embodiments of the present invention or certain parts of the embodiments.

[0178] Each embodiment in this specification is described in a progressive manner. The same or similar parts between the embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments. In particular, for the device or system embodiments, since they are basically similar to the method embodiments, the description is relatively simple. For the relevant parts, refer to the partial description of the method embodiments. 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 can be selected according to actual needs to achieve the purpose of the scheme of this embodiment. A person of ordinary skill in the art can understand and implement it without making any creative efforts.

[0179] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A method for predicting faults of complex electromechanical equipment, characterized in that: include: The accumulated historical fault text data of complex electromechanical equipment is processed to construct positive and negative instance sets, and the number of fault label clusters is calculated according to the fault type; Clustering the positive and negative instance sets according to the fault label clustering results, constructing an instance similarity matrix, calculating the correlation between labels, and updating the instance similarity matrix through the label similarity matrix; Based on the updated instance similarity matrix, a matrix of similar instances and a matrix of non-similar instances are constructed, and the corresponding weight of each instance is calculated to obtain closely related labels for each label. A classification training model is constructed using the matrix of similar instances of electromechanical equipment and the matrix of non-similar instances. The classification training model is trained using historical instance data of electromechanical equipment. Unknown instances are constructed based on equipment performance, and the trained classification training model is used to predict faults of the unknown instances.

2. The method according to claim 1, characterized in that The aforementioned processing of the accumulated historical fault text data of complex electromechanical equipment, constructing a set of positive and negative instances, and calculating the number of fault label clusters according to the fault type includes: Step 1: Sort out the historical text data of electromechanical equipment, use instances to represent each fault data, use labels to represent each fault type, and construct the relationship matrix E between instances and labels: Among them, m represents the number of all instances, n represents the number of all labels, and a ij (1≤i≤m,1≤j≤m) is 0 or 1, 0 means that the instance does not have the corresponding label, and 1 means that the instance has the corresponding label; Step 2: Divide the instance set into a positive instance set P and a negative instance set N, where P represents the set of instances with labels under each label, and N represents the set of instances without labels under each label; Step 3: Calculate the label entropy E(k) of the kth label for the positive instance set P and the negative instance set N respectively. p )、E(k n ): Among them, E(k p ) represents the label entropy of the kth label of the positive instance set P, where Indicates the ratio of the jth label vector appearing in P corresponding to the kth label to the total label vector appearing in P, E(k n ) represents the label entropy of the kth label of the negative instance set N, where It represents the ratio of the jth label vector appearing in N corresponding to the kth label to the total label vector appearing in N; Step 4: Calculate the basic entropy w of the kth label k : Among them, w k represents the basic entropy of the kth label, minE(k) represents the label entropy E(k) corresponding to the kth label p ) and E(k n ) whichever is smaller; Step 5: Calculate the ratio parameter r of the kth label k : Among them, r k represents the ratio parameter of the kth label, and s is the ratio smoothing parameter; Step 6: Calculate the number of clusters m for the kth label k : m k =r k ×min{|P|,|N|} Among them, m k Indicates the number of clusters for the k-th label.

3. The method according to claim 2, characterized in that The clustering process of the positive and negative instance sets according to the fault label clustering result, constructing an instance similarity matrix, calculating the correlation between labels, and updating the instance similarity matrix using the label similarity matrix includes: Step 1: Perform k-means clustering on the positive and negative instance sets P and N respectively, where the number of clusters of positive and negative instances of the kth label is r k ; Step 2: Construct the instance similarity matrix MI based on the cluster number of the instance in the clustering result: Among them, MI kij represents the element in row i and column j of the instance similarity matrix of the kth instance, c k (x i ) represents x i The number of clusters in the set of positive and negative instances corresponding to the kth label, x i represents the i-th instance in E, c k (x j ) represents x j The number of clusters in the set of positive and negative instances corresponding to the kth label, x j represents the jth instance in E; Step 3: Calculate label similarity MC: Among them, MC kj represents the label similarity between the kth label and the jth label, γ represents the cosine similarity parameter, l k represents the kth label in E, l j represents the jth label in E; Step 4: Update the instance similarity matrix according to the label similarity matrix to obtain MI_new: Among them, MI_new represents the updated instance similarity matrix, MC kj Indicates the label similarity between the kth label and the jth label, MI j Represents the j-th column element in the original instance similarity matrix MI; Step 5: Create a mapping to obtain the properties of a specific tag: Among them, x i represents the i-th instance in E, FI k (x i ) represents the instance x i The Euclidean distance set to each cluster center of the P and N clustering results corresponding to the kth label is represented by the function d(·), d Represents xi and The Euclidean distance of Indicates the coordinates of the first cluster center of P corresponding to the kth label, Indicates the last cluster center coordinate of P corresponding to the kth label, Indicates the coordinates of the first cluster center of N corresponding to the kth label, Indicates the last cluster center coordinate of N corresponding to the k-th label; Step 6: Construct label similarity matrix A k : Among them, A k represents the label similarity matrix, a i Represents x i Relative to The value of the label in represents the label matrix consisting of labels other than the label in the kth column, Indicates Y i Relative to label l k The value of l k Belongs to Y i ,but is 1, otherwise 0, Y i Indicates that x i Related related tag sets, express The first tag in express The kth label in .

4. The method according to claim 3, characterized in that The method of constructing a matrix of similar instances and a matrix of non-similar instances for complex electromechanical equipment based on the updated instance similarity matrix, calculating the corresponding weight of each instance, and obtaining closely related labels for each label includes: Step 1: Preprocess the label similarity matrix. For the label in column k, store the instances belonging to the label in D k In D, instances that do not belong to this label are stored in nk middle; Step 2: Construct similar instance matrices and non-similar instance matrices based on the updated instance similarity matrix, and calculate the Euclidean distance ranking n between similar instance matrices. r and the Euclidean distance ranking n between matrices of different instances m : n r =sort(d(D k ,D k )) n m =sort(d(D k ,D nk )) Among them, n r Indicates the Euclidean distance ranking between similar instance matrices, n m represents the Euclidean distance ranking between different instance matrices, d represents the Euclidean distance between two matrices, D k Indicates A k The instance matrix of the label belonging to the kth column in D nk Indicates A k The instance matrix that does not belong to the k-th column label; Step 3: Calculate the corresponding weight ω of the b-th instance b : Among them, ω b Indicates the corresponding weight of the b-th instance, len(n r ) represents the matrix n r The number of rows, len(n m ) represents the matrix n m The number of rows of matrix E, k is the number of neighbor labels, length(Dni) represents the number of rows of matrix Dni, m is the number of rows of matrix E, nr i Represents the matrix n r The i-th element in Represents the matrix n m The i-th element in ; Step 4: Calculate the kth label l k Closely related tags: k : Among them, rank(l k ) represents the b The features returned by feature selection, t represents the number of close labels determined, express The kth label in .

5. The method according to claim 4, characterized in that The method of constructing a classification training model using the electromechanical equipment similar instance matrix and the non-similar instance matrix, training the classification training model using historical instance data of the electromechanical equipment, constructing unknown instances based on the state performance of the equipment under test, and predicting the failure of the unknown instances using the trained classification training model includes: Construct a binary training set D: Among them, D k represents the k-th row element in the binary training set D, x i represents the i-th instance in E, FI k (x i ) represents the instance x i The set of Euclidean distances to each cluster center of the P and N clustering results corresponding to the k-th label, Indicates Y i Relative to label l k The value of l k Belongs to Y i ,but is 1; otherwise is 0, Y i Indicates that x i Related set of related tags; Construct binary training set D kj : Among them, D kj represents the binary training set, x i represents the i-th instance in E, FI k (xi) represents instance x i The set of Euclidean distances to each cluster center of the P and N clustering results corresponding to the k-th label, Indicates Y i Relative to label l k The value of l k Belongs to Y i ,but is 1, otherwise 0, Y i Indicates that x i Related set of related tags; Construct a binary learner as a classification training model, the binary classifier is a support vector machine SVM, and the binary training set D is combined with the binary training set D kj Input to the binary learner, train the binary learner, and the binary learner outputs the mapping function f of the unknown instance i ; Construct an unknown instance u based on the performance of the device under test and use the mapping function f i The unknown instance u is input into the mapping function f i , the mapping function f i Output the relevant label prediction result Y corresponding to the unknown instance u u : Among them, Y u is the relevant label prediction result of the unknown instance u, f i Mapping function constructed for binary learner, FI k (u) represents the Euclidean distance set from instance u to each cluster center of the P and N clustering results corresponding to the kth label.

6. A non-transitory computer-readable storage medium, characterized in that The non-transitory computer-readable storage medium is used to store computer instructions, and when the computer instructions are executed by a processor, the complex electromechanical equipment fault prediction method according to any one of claims 1 to 5 is implemented.

7. A computer program product, characterized in that The invention comprises a computer program, which is used to implement the complex electromechanical equipment fault prediction method according to any one of claims 1 to 5 when the computer program is run on one or more processors.

8. An electronic device, characterized in that: include: 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 to enable the electronic device to execute instructions for implementing the complex electromechanical equipment fault prediction method as described in any one of claims 1 to 5.

Images