Drilling equipment maintenance cost assessment method based on maintenance text record mining
By performing domain-specific fine-tuning and clustering analysis of the BERT model of the drilling equipment maintenance text data, combined with the Markov process model, the accuracy of drilling equipment maintenance cost prediction is solved, and more effective maintenance cost prediction and fault prediction are achieved.
Patent Information
- Application Number
- CN202510189291.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-02-20
- Publication Date
- 2025-07-08
AI Technical Summary
In the prior art, maintenance cost prediction methods for drilling equipment are limited in effectiveness when processing unstructured data, especially when processing censored and incomplete data, it is difficult to meet the needs of complex maintenance activities.
The domain-specific fine-tuning based on the BERT model is used to vectorize the text data of the drilling equipment maintenance, perform cluster analysis, and combine it with the Markov process model to predict the equipment maintenance cost.
Improves the accuracy of drilling equipment maintenance cost prediction, enables more efficient prediction of failures and implements forward-looking maintenance to meet the needs of complex maintenance activities.
Smart Images
Figure CN120277169A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the technical field of operation and maintenance of drilling equipment, and particularly to a method for evaluating the maintenance cost of drilling equipment based on mining maintenance text records. Background Art
[0002] Drilling equipment is a special mechanical device used for operations such as drilling wells, circulating mud, and lifting drill pipes during the drilling construction process, mainly including key equipment such as top drives, mud pumps, and drawworks. These devices operate under high intensity and complex working conditions, and have high requirements for their reliability and maintenance. Currently, in the field of operation and maintenance of drilling equipment, the operating status and working condition data of the equipment are usually collected through sensors, and combined with statistical analysis methods, the maintenance cost is analyzed to optimize the maintenance strategy and reduce the downtime. However, in the face of unstructured data, the effectiveness of traditional statistical methods is limited. Especially when dealing with censored and incomplete data, it is difficult to define the expected value and outliers. This makes it impossible to meet the needs of complex maintenance activities relying solely on statistical data. Summary of the Invention
[0003] This application provides a method for evaluating the maintenance cost of drilling equipment based on mining maintenance text records to solve the problem that relying solely on statistical data in the prior art cannot meet the needs of complex maintenance activities.
[0004] In a first aspect, this application provides a model generation method, including:
[0005] Obtain multiple maintenance text data of the target model equipment, each maintenance text data including a maintenance record, the recording time of the maintenance record, and the maintenance cost;
[0006] According to the pre-trained BERT model, perform domain-specific fine-tuning on the text data in the field of drilling equipment maintenance to obtain a fine-tuned model, and vectorize each maintenance record through the fine-tuned model to obtain a numerical vector corresponding to each maintenance record;
[0007] Perform clustering processing on all maintenance records according to the numerical vector corresponding to each maintenance record to obtain a maintenance record cluster;
[0008] Determine the class label of each maintenance record according to the maintenance record cluster and the maintenance cost corresponding to each maintenance record;
[0009] Calculate the unknown parameters of the Markov process model according to the maintenance time and class label of each maintenance record to obtain an equipment maintenance cost prediction model.
[0010] In some embodiments, based on the pre-trained BERT model, domain-specific fine-tuning is performed on the text data in the field of drilling equipment maintenance to obtain a fine-tuned model. The fine-tuned model is used to vectorize each maintenance record, obtaining a numerical vector corresponding to each maintenance record, including:
[0011] For each maintenance record, preprocess the maintenance record to obtain text data. The preprocessing includes case conversion, special character removal, spelling check correction, and lemmatization;
[0012] Based on the pre-trained BERT model, domain-specific fine-tuning is performed on the text data in the field of drilling equipment maintenance to obtain a fine-tuned model. The fine-tuned model is used to vectorize the text data, obtaining a numerical vector corresponding to the text data.
[0013] In some embodiments, based on the numerical vectors corresponding to each maintenance record, clustering is performed on all maintenance records to obtain maintenance record clusters, including:
[0014] According to different initial number of clustering classes and the clustering criterion function corresponding to each initial number of clustering classes, perform clustering on all numerical vectors to obtain a k-WCSS graph;
[0015] Based on the elbow in the k-WCSS graph, obtain the target number of clustering classes;
[0016] According to the target number of clustering classes, obtain the maintenance record clusters corresponding to the target number of clustering classes.
[0017] In some embodiments, based on the maintenance record clusters and the maintenance costs corresponding to each maintenance record, determine the class labels of each maintenance record, including:
[0018] According to the maintenance costs of each maintenance record, obtain the maintenance cost distribution interval of each maintenance record cluster;
[0019] According to each maintenance cost distribution interval, obtain the class label of each maintenance record cluster;
[0020] According to the class labels of each maintenance record cluster, obtain the class labels of each maintenance record.
[0021] In some embodiments, based on the maintenance time and class label of each maintenance record, calculate the unknown parameters of the Markov process model to obtain a device maintenance cost prediction model, including:
[0022] According to the maintenance time and class label of each maintenance record, determine the change situation of the class label of the target model device over the maintenance time;
[0023] According to the change situation, determine each two adjacent class labels and the time interval between each two adjacent class labels;
[0024] Calculate the unknown parameters of the Markov process model according to every two adjacent category labels and the time interval between every two adjacent category labels, and obtain the equipment maintenance cost prediction model.
[0025] In some embodiments, calculating the unknown parameters of the Markov process model according to every two adjacent category labels and the time interval between every two adjacent category labels, and obtaining the equipment maintenance cost prediction model, including:
[0026] Determine the transition probability matrix of the category labels of the target model equipment changing with time according to every two adjacent category labels and the time interval between every two adjacent category labels;
[0027] Perform unknown parameter calculation on the Markov process model according to the preset initial state and the transition probability matrix to obtain the equipment maintenance cost prediction model.
[0028] In a second aspect, the present application provides a method for evaluating the maintenance cost of a drilling equipment based on mining maintenance text records, including:
[0029] Determine the time to be predicted for the target model equipment;
[0030] Input the time to be predicted and the initial state into the equipment maintenance cost prediction model to obtain the prediction result of the target model equipment at the time to be predicted. The prediction result includes each category label and the probability result corresponding to each category label. The equipment maintenance cost prediction model is the model obtained according to the method of the present application;
[0031] Determine the state of the target model equipment at the time to be predicted according to the probability result.
[0032] In a third aspect, the present application provides a model generation device, including:
[0033] An acquisition module, configured to acquire a plurality of maintenance text data of the target model equipment, and each maintenance text data includes a maintenance record, a recording time of the maintenance record, and a maintenance cost;
[0034] A vectorization representation module, configured to perform domain-specific fine-tuning on the text data in the field of drilling equipment maintenance according to a pre-trained BERT model to obtain a fine-tuned model, and perform vectorization processing on each maintenance record through the fine-tuned model to obtain a numerical vector corresponding to each maintenance record;
[0035] A clustering module, configured to perform clustering processing on all maintenance records according to the numerical vector corresponding to each maintenance record to obtain a maintenance record cluster;
[0036] A determination module, configured to determine a category label for each maintenance record according to a maintenance record cluster and the maintenance cost corresponding to each maintenance record;
[0037] A calculation module, configured to calculate unknown parameters of a Markov process model according to the maintenance time and category label of each maintenance record, so as to obtain a device maintenance cost prediction model.
[0038] In a fourth aspect, the present application provides a drilling equipment maintenance cost assessment device based on maintenance text record mining, including:
[0039] A determination time module, configured to determine a time to be predicted for a target model device;
[0040] A prediction module, configured to input the time to be predicted and an initial state into the device maintenance cost prediction model to obtain a prediction result of the target model device at the time to be predicted, where the prediction result includes each category label and a probability result corresponding to each category label, and the device maintenance cost prediction model is a model obtained according to the method of the present application;
[0041] A category label module, configured to determine a category label of the target model device at the time to be predicted according to the probability result.
[0042] In a fifth aspect, the present application provides an electronic device, including: a processor, and a memory communicatively connected to the processor;
[0043] The memory stores computer-executable instructions;
[0044] The processor executes the computer-executable instructions stored in the memory to implement the method of the present application.
[0045] In a sixth aspect, the present application provides a computer-readable storage medium, in which computer-executable instructions are stored, and when the computer-executable instructions are executed by a processor, they are used to implement the method of the present application.
[0046] The drilling equipment maintenance cost assessment method based on maintenance text record mining provided by this application obtains multiple maintenance text data of the target model equipment, and each maintenance text data includes a maintenance record, the recording time of the maintenance record, and the maintenance cost; according to the pre-trained BERT model, domain-specific fine-tuning is performed on the text data in the field of drilling equipment maintenance to obtain a fine-tuned model, and each maintenance record is vectorized through the fine-tuned model to obtain a numerical vector corresponding to each maintenance record; according to the numerical vector corresponding to each maintenance record, all maintenance records are clustered to obtain a maintenance record cluster; then, according to the maintenance record cluster and the maintenance cost corresponding to each maintenance record, the category label of each maintenance record is determined, and finally, according to the maintenance time and category label of each maintenance record, the unknown parameters of the Markov process model are calculated to obtain a device maintenance cost prediction model, which improves the accuracy of the model in predicting the maintenance cost of drilling equipment. BRIEF DESCRIPTION OF THE DRAWINGS
[0047] The drawings herein are incorporated into the specification and form a part of the specification, showing embodiments consistent with the present application, and are used together with the specification to explain the principles of the present application.
[0048] Figure 1 A schematic diagram of a scenario generated by a model provided by an embodiment of the present application;
[0049] Figure 2 A schematic flowchart of a model generation method provided by an embodiment of the present application;
[0050] Figure 3 A schematic flowchart of another model generation method provided by an embodiment of the present application;
[0051] Figure 4 A schematic flowchart of a drilling equipment maintenance cost assessment method based on maintenance text record mining provided by an embodiment of the present application;
[0052] Figure 5 A schematic structural diagram of a model generation device provided by an embodiment of the present application;
[0053] Figure 6 A schematic structural diagram of a drilling equipment maintenance cost assessment device based on maintenance text record mining provided by an embodiment of the present application;
[0054] Figure 7 A schematic structural diagram of an electronic device provided by an embodiment of the present application.
[0055] Through the above-mentioned drawings, specific embodiments of the present application have been shown, and there will be more detailed descriptions hereinafter. These drawings and textual descriptions are not intended to limit the scope of the concept of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. Detailed Implementation Modes
[0056] Exemplary embodiments will be described in detail herein, and examples thereof are shown in the accompanying drawings. When the following description refers to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application. On the contrary, they are merely examples of apparatuses and methods consistent with some aspects of the present application as detailed in the appended claims.
[0057] To clearly understand the technical solution of the present application, the solutions of the prior art will be introduced in detail first.
[0058] Drilling equipment is a mechanical device and equipment used for drilling construction in specific working conditions, mainly composed of a drilling rig, a mud pump, a mud purification device, a mud mixer, a derrick, etc. At present, in the field of reliability maintenance of drilling equipment, the maintenance cost of drilling equipment is mainly statistically analyzed through sensor data.
[0059] However, in the face of unstructured data, the effectiveness of traditional statistical methods is limited. Especially when dealing with censored and incomplete data, it is difficult to define expected values and outliers. This makes it impossible to meet the needs of complex maintenance activities relying solely on statistical data.
[0060] Maintenance work orders, as equipment maintenance records, provide rich context information. By using a classification model to mine these text data, faults can be predicted more effectively and proactive maintenance can be implemented. However, spelling mistakes and technical terms in maintenance records will affect the effect of real-time analysis. Therefore, natural language processing (NLP) technology has become the key to solving this problem.
[0061] Although NLP shows potential in maintenance text analysis, traditional methods often rely on standard corpora and are difficult to adapt to the specialized texts in technical fields.
[0062] In view of the problem that relying solely on statistical data cannot meet the needs of complex maintenance activities, the inventors found in their research that all maintenance records can be clustered to obtain maintenance record clusters, and then the category label of each maintenance record can be determined according to the maintenance record clusters and the maintenance cost corresponding to each maintenance record. Thus, a model training can be carried out according to the maintenance time and category label of the maintenance record to obtain an equipment maintenance cost prediction model.
[0063] The application scenarios of the model generation method provided by the embodiments of the present application will be introduced below.
[0064] Figure 1 A schematic diagram of a scenario for model generation provided by an embodiment of the present application is shown in Figure 1As shown in the figure, the scenario includes the maintenance text data of the target model device and the server. The maintenance text data includes maintenance records, the recording time of the maintenance records, and the maintenance costs. The server obtains multiple maintenance text data of the target model device, and each maintenance text data includes a maintenance record, the recording time of the maintenance record, and the maintenance cost; according to the pre-trained BERT model, domain-specific fine-tuning is performed on the text data in the field of drilling equipment maintenance to obtain a fine-tuned model, and each maintenance record is vectorized through the fine-tuned model to obtain a numerical vector corresponding to each maintenance record; according to the numerical vector corresponding to each maintenance record, clustering processing is performed on all maintenance records to obtain a maintenance record cluster; according to the maintenance record cluster and the maintenance cost corresponding to each maintenance record, the class label of each maintenance record is determined; according to the maintenance time and class label of each maintenance record, unknown parameter calculation is performed on the Markov process model to obtain a device maintenance cost prediction model.
[0065] Figure 2 It is a schematic flowchart of a model generation method provided by an embodiment of the present application. As Figure 2 shown, the method includes:
[0066] S201. Obtain multiple maintenance text data of the target model device, and each maintenance text data includes a maintenance record, the recording time of the maintenance record, and the maintenance cost.
[0067] The target model device may include engineering operation equipment, such as an excavator.
[0068] Among them, the maintenance text data may refer to the text data when maintaining the target model device. The maintenance record may refer to the specific content of maintaining the target model device, such as repairing or replacing a certain component. The recording time may refer to the time of maintaining the target model device, and the maintenance cost may refer to the expenses incurred when maintaining the target model device.
[0069] In the present application, in order to predict the maintenance cost of the target model device at a preset time, therefore, it is necessary to extract the maintenance text data from the historical maintenance data of the target model device, so that clustering processing can be performed according to the maintenance records.
[0070] S202. According to the pre-trained BERT model, domain-specific fine-tuning is performed on the text data in the field of drilling equipment maintenance to obtain a fine-tuned model, and each maintenance record is vectorized through the fine-tuned model to obtain a numerical vector corresponding to each maintenance record.
[0071] Among them, the fine-tuned model may refer to being obtained by performing domain-specific fine-tuning on the basis of the large-scale pre-trained BERT model according to the text data in the field of drilling equipment maintenance. Vectorizing the maintenance records through the fine-tuned model improves the accuracy of text representation and classification.
[0072] S203. Cluster all the maintenance records according to the numerical vectors corresponding to each maintenance record to obtain maintenance record clusters.
[0073] Among them, clustering can refer to a data analysis technique used to divide a set of objects into multiple groups or clusters, such that objects within the same cluster are more similar in a certain sense, while objects in different clusters are relatively less similar. Clustering is an unsupervised learning method, which means it does not require pre-labeled data.
[0074] In this application, by clustering all the maintenance records, the similarities between each maintenance record are mined to obtain multiple maintenance record clusters.
[0075] S204. Determine the class label for each maintenance record according to the maintenance record clusters and the maintenance costs corresponding to each maintenance record.
[0076] Among them, the class label can refer to the class label used to distinguish each maintenance record cluster. For example, assuming there are maintenance record cluster A and maintenance record cluster B, then maintenance record cluster A has the corresponding class label A, and maintenance record cluster B has the corresponding class label B.
[0077] In this application, after obtaining the maintenance record clusters, the overall cost situation of each maintenance record cluster can be determined according to the maintenance costs of each maintenance record in each maintenance record cluster, and the class label of the maintenance record cluster can be determined according to the overall cost situation.
[0078] S205. Calculate the unknown parameters of the Markov process model according to the maintenance time and class label of each maintenance record to obtain the equipment maintenance cost prediction model.
[0079] In this application, after obtaining the maintenance time and class label of each maintenance record, calculate the unknown parameters of the Markov process model, and substitute the calculated parameters back into the Markov process model to obtain the equipment maintenance cost prediction model.
[0080] A model generation method provided by an embodiment of the present application. By obtaining multiple maintenance text data of a target model device, each maintenance text data includes a maintenance record, the recording time of the maintenance record, and the maintenance cost; according to a pre-trained BERT model, domain-specific fine-tuning is performed on the text data in the field of drilling equipment maintenance to obtain a fine-tuned model, and each maintenance record is vectorized through the fine-tuned model to obtain a numerical vector corresponding to each maintenance record; according to the numerical vectors corresponding to each maintenance record, clustering processing is performed on all maintenance records to obtain maintenance record clusters; according to the maintenance record clusters and the maintenance costs corresponding to each maintenance record, the category labels of each maintenance record are determined; according to the maintenance time and category labels of each maintenance record, unknown parameter calculation is performed on the Markov process model to obtain a device maintenance cost prediction model, which improves the accuracy of the model's prediction of maintenance costs.
[0081] Figure 3 It is a schematic flowchart of another model generation method provided by an embodiment of the present application. As Figure 3 shown, the method includes:
[0082] S301. Obtain multiple maintenance text data of a target model device, and each maintenance text data includes a maintenance record, the recording time of the maintenance record, and the maintenance cost.
[0083] S302. For each maintenance record, preprocess the maintenance record to obtain text data. The preprocessing includes case conversion, special character removal, spelling check correction, and lemmatization.
[0084] In this application, to make the content in the maintenance records better recognizable, the maintenance records can be preprocessed to make them more standardized and accurate. Among them, case conversion: Since during the process of word segmentation and dictionary construction, "Replace" and "replace" will be regarded as different words, all uppercase and lowercase letters can be adjusted to the same specification. For example, all uppercase letters can be changed to lowercase to avoid duplicate words in the dictionary. Removal of special characters: It can refer to clearing non-English characters and special symbols that are useless for understanding the text. For example, delete "#" representing numbers, "&" representing the parallel relationship, "×" representing the quantitative relationship, "." representing the end of a sentence, etc. in the maintenance text. Spelling check and correction: For nouns specific to the field of excavator bucket maintenance, such as "rh" which is an abbreviation for "right-hand". These specific nouns may not be in the vocabulary of large language models, and directly using them for word vectorization conversion will result in loss of word meaning. To solve this problem, during the preprocessing process, these specific words can be converted into corresponding common words to ensure the accuracy and consistency of the data. Lemmatization: It can refer to reducing various forms of words to their common base forms. For example, the NLTK library can be used to implement lemmatization of words in short texts and lemmatize each word in the maintenance text one by one.
[0085] S303. According to the pre-trained BERT model, perform domain-specific fine-tuning on the text data in the field of drilling equipment maintenance to obtain a fine-tuned model, and use the fine-tuned model to perform vectorization representation on the text data to obtain the numerical vector corresponding to the text data.
[0086] In this application, to further facilitate machine recognition, LSA and Word2vec can also be used to perform text vectorization representation on the text data.
[0087] S304. According to the numerical vector corresponding to each maintenance record, perform clustering processing on all maintenance records to obtain maintenance record clusters.
[0088] Specifically, according to the numerical vector corresponding to each maintenance record, performing clustering processing on all maintenance records to obtain maintenance record clusters may include:
[0089] Perform clustering processing on all numerical vectors according to different initial clustering numbers and the clustering criterion functions corresponding to each initial clustering number to obtain a k-WCSS graph;
[0090] Obtain the target clustering number according to the elbow in the k-WCSS graph;
[0091] Obtain the maintenance record clusters corresponding to the target clustering number according to the target clustering number.
[0092] Among them, the initial number of clustering classes can refer to an integer value that starts from a preset minimum initial number of clustering classes and increases sequentially when performing clustering processing on all numerical vectors through K-means. For example, the minimum initial number of clustering classes can be 2, and increasing sequentially, the initial number of clustering classes such as 3, 4, 5... can be obtained.
[0093] The clustering criterion function can refer to the function value calculated after performing clustering processing according to the initial number of clustering classes. The clustering criterion function can be used as a standard to measure the quality of clustering, mainly based on the principle of small within-class distance and large between-class distance. The goal of the clustering criterion function is to make the data points within the same class as similar as possible, while the data points between different classes are as different as possible. The clustering criterion function can be the within-cluster sum of squares (WCSS): This criterion function is defined as the sum of the squares of the distances from the samples within each class to their class centers. The goal is to minimize the WCSS, that is, to minimize the within-class distance.
[0094] The k-WCSS graph can refer to the coordinate graph of the initial number of clustering classes and the clustering criterion function, where the horizontal axis is the initial number of clustering classes and the vertical axis is the clustering criterion function.
[0095] The elbow in the k-WCSS graph can refer to the point where the clustering criterion function begins to decrease in a linear manner, and this point can be identified as the target number of clustering classes.
[0096] In this application, in order to determine the optimal number of clustering classes, K-means clustering processing is performed according to a preset multiple initial numbers of clustering classes, where the K value is the initial number of clustering classes, and then the clustering results corresponding to different initial numbers of clustering classes are obtained. In order to determine the clustering quality, the clustering criterion function corresponding to each initial number of clustering classes is calculated according to the obtained clustering results. According to the corresponding relationship between the initial number of clustering classes and the clustering criterion function, a k-WCSS graph is obtained, and then according to the trend in the k-WCSS graph, the point corresponding to the elbow in the graph is determined as the target number of clustering classes, and then the clustering result corresponding to the target number of clustering classes is used as the maintenance record cluster.
[0097] S305. Determine the category label of each maintenance record according to the maintenance record cluster and the maintenance cost corresponding to each maintenance record.
[0098] Specifically, determining the category label of each maintenance record according to the maintenance record cluster and the maintenance cost corresponding to each maintenance record may include:
[0099] Obtain the maintenance cost distribution interval of each maintenance record cluster according to the maintenance cost of each maintenance record;
[0100] Obtain the category label of each maintenance record cluster according to each maintenance cost distribution interval;
[0101] Obtain the category label of each maintenance record according to the category label of each maintenance record cluster.
[0102] In this application, after obtaining the maintenance record clusters, in order to determine the maintenance cost situation of each maintenance record cluster, establish the relationship between different maintenance record clusters and maintenance costs, count the maintenance costs of the maintenance records in each maintenance record cluster, determine the maintenance cost distribution interval in each maintenance record cluster, and according to the size relationship between the maintenance cost distribution intervals, determine the category labels of different levels, classify the maintenance record clusters through the category labels, and finally determine the category labels of the maintenance clustering in the maintenance record clusters through the category labels of each maintenance record cluster.
[0103] Exemplarily, assume there are maintenance record cluster A, maintenance record cluster B, and maintenance record cluster C. According to statistics, obtain the maintenance cost distribution interval A of maintenance record cluster A, the maintenance cost distribution interval B of maintenance record cluster B, and the maintenance cost distribution interval C of maintenance record cluster C. Then compare the size relationship between the maintenance cost distribution interval A, the maintenance cost distribution interval B, and the maintenance cost distribution interval C. For example, it can be through the average value of the maintenance costs in the maintenance cost distribution interval, the maintenance cost with the largest quantity in the maintenance cost distribution interval, or the numerical range where the maintenance costs are most concentrated in the maintenance cost distribution interval, etc. Obtain the comparison result: maintenance cost distribution interval A > maintenance cost distribution interval B > maintenance cost distribution interval C. Then record the category label of maintenance record cluster A as high maintenance cost, the category label of maintenance record cluster B as medium maintenance cost, and the category label of maintenance record cluster C as low maintenance cost.
[0104] S306. Calculate the unknown parameters of the Markov process model according to the maintenance time and category label of each maintenance record to obtain the equipment maintenance cost prediction model.
[0105] Specifically, calculating the unknown parameters of the Markov process model according to the maintenance time and category label of each maintenance record to obtain the equipment maintenance cost prediction model may include:
[0106] Determine the change situation of the category label of the target model equipment with the maintenance time according to the maintenance time and category label of each maintenance record;
[0107] Determine every two adjacent category labels and the time interval between every two adjacent category labels according to the change situation;
[0108] Calculate the unknown parameters of the Markov process model according to every two adjacent category labels and the time interval between every two adjacent category labels to obtain the equipment maintenance cost prediction model.
[0109] In this application, after obtaining the category labels of each maintenance record, the time interval for each category label to change can be determined according to the correspondence between the recording time of the maintenance record and the category label. For example, suppose the category label of maintenance record A with recording time t1 is low maintenance cost, the category label of maintenance record B with recording time t2 is low maintenance cost, the category label of maintenance record C with recording time t3 is high maintenance cost, and the category label of maintenance record D with recording time t4 is high maintenance cost, and t1 - t4 increase in sequence. Then, the time interval t2 - t1 for low maintenance cost to convert to low maintenance cost, the time interval t3 - t2 for low maintenance cost to convert to high maintenance cost, and the time interval t4 - t3 for high maintenance cost to convert to high maintenance cost are obtained, a total of 3 data pairs. After obtaining all the data on category label conversion and time intervals, the unknown parameters of the Markov process model are calculated to obtain the changing rule of the category label over time, thereby obtaining the equipment maintenance cost prediction model.
[0110] Specifically, calculating the unknown parameters of the Markov process model according to every two adjacent category labels and the time interval between every two adjacent category labels to obtain the equipment maintenance cost prediction model may include:
[0111] Determine the transition probability matrix of the category label of the target model device changing over time according to every two adjacent category labels and the time interval between every two adjacent category labels;
[0112] Obtain the equipment maintenance cost prediction model according to the preset initial state and the transition probability matrix.
[0113] The transition probability matrix may refer to the concept of state transition introduced in Markov analysis. The so-called state refers to the possible states that objective things may appear or exist; state transition refers to the probability of objective things transitioning from one state to another. Markov analysis points out that in the transfer of certain factors in a system, the result of the nth time is only affected by the result of the (n - 1)th time, that is, it is only related to the current state and has nothing to do with the past state.
[0114] The determination method of the transition probability matrix can be obtained through maximum likelihood estimation.
[0115] Maximum likelihood estimation is a method for estimating the parameters of a probability model, which provides a way to evaluate unknown parameters, that is, using a known model to estimate unknown parameters. Its core idea is to reverse-infer the parameters that make the result occur with the greatest probability under the condition of known experimental results. Maximum likelihood estimation requires that all samplings are independent and identically distributed. From the perspective of classical probability, for independent samples from a population , its probability density function is , and the purpose of maximum likelihood estimation is to find a suitable parameter estimate such that the likelihood function is maximized. If the likelihood function is differentiable, the maximum likelihood estimation of the parameters can be transformed into solving the equations shown below.
[0116] Exemplarily, first, clarify all possible states of the Markov chain, such as high maintenance cost, medium maintenance cost, and low maintenance cost. Count the number of times each state transitions to other states. Divide the number of times each state transitions to other states by the total number of transitions of that state to obtain the transition probability. Ensure that the sum of the elements in each row (or column) of the transition probability matrix is equal to 1, and the transition probability matrix describing the state transition of the Markov chain is obtained.
[0117] In this application, after obtaining the transition probability matrix, it is necessary to obtain the variation law of the state probability over time, which can be expressed in the form of a linear ordinary differential equation. Solving the linear ordinary differential equation can obtain the state probability at any time. Solve it by the method of indefinite integral, which is called the matrix exponential. By calculating the matrix exponential and multiplying it by the known initial condition, it can be used to solve the linear ordinary differential equation . To calculate the matrix exponential , the knowledge of eigenvalues and eigenvectors can be used for further transformation. Let Λ be the diagonal matrix composed of the eigenvalues of the transition probability matrix, matrix S be the eigenvector matrix of the transition probability matrix, and matrix be the inverse matrix of matrix S, then . Finally, by substituting the representing the initial state vector, the state vector at the time interval t from the current moment can be calculated to complete state discrimination and prediction. The initial state vector can be, for example, [1, 0, 0], which is used to represent that the probability of class label A is 1, the probability of class label B is 0, and the probability of class label C is 0, that is, the current state is class label A.
[0118] Another model generation method provided by this application. By obtaining multiple maintenance text data of a target model device, each maintenance text data includes a maintenance record, the recording time of the maintenance record, and the maintenance cost. For each maintenance record, preprocess the maintenance record to obtain text data. The preprocessing includes case conversion, special character removal, spelling check correction, and lemmatization. Vectorize the text data according to the BERT text representation model to obtain numerical vectors. Cluster all the numerical vectors to obtain maintenance record clusters. Determine the class label of each maintenance record according to the maintenance record cluster and the maintenance cost corresponding to each maintenance record. Calculate the unknown parameters of the Markov process model according to the maintenance time and class label of each maintenance record to obtain a device maintenance cost prediction model, which improves the accuracy of the model's prediction of the maintenance cost.
[0119] Figure 4 It is a schematic flowchart of a drilling equipment maintenance cost assessment method based on maintenance text record mining provided by an embodiment of this application, as Figure 4 shown, the method includes:
[0120] S401. Determine the time to be predicted for the target model device.
[0121] Among them, the time to be predicted can refer to the time interval from the current moment. For example, assuming the current moment is 0 and it is necessary to predict the device state 30 days after the current moment, then the time to be predicted is 30 days.
[0122] S402. Input the time to be predicted and the starting state into the device maintenance cost prediction model to obtain the prediction result of the target model device at the time to be predicted. The prediction result includes each class label and the probability result corresponding to each class label. The device maintenance cost prediction model is the model obtained according to the model generation method provided by this application.
[0123] Among them, the starting state can be represented by a vector. Each position in the vector corresponds to a class label. Through the starting state, the current class label of the target model device can be determined. For example, if the current class label of the target model device is class label A, the starting state can be represented by a vector. The first position in the vector corresponds to class label A, the second position corresponds to class label B, and the third position corresponds to class label C. Then it can be determined that the starting state is the vector [1, 0, 0].
[0124] The prediction result can refer to the probabilities corresponding to different class labels. For example, the predicted probability of converting to class label A is 0.5, the probability of converting to class label B is 0.3, and the probability of converting to class label C is 0.2.
[0125] S403. Determine the state of the target model device at the time to be predicted according to the probability result.
[0126] In this application, after obtaining the probability results, the class label corresponding to the value with the highest probability is selected as the class label of the time to be predicted. For example, the probability of converting to class label A is 0.5, the probability of converting to class label B is 0.3, and the probability of converting to class label C is 0.2. It is determined that the conversion to class label A is the state of the target model device at the time to be predicted.
[0127] A maintenance cost prediction method provided by an embodiment of this application determines the time to be predicted for the target model device, inputs the time to be predicted and the starting state into the device maintenance cost prediction model, obtains the prediction result of the target model device at the time to be predicted, and determines the class label of the target model device at the time to be predicted according to the probability result. The accuracy of maintenance cost prediction is improved.
[0128] Figure 5 It is a structural schematic diagram of a model generation device provided by an embodiment of this application. The device 50 includes:
[0129] An acquisition module 501, configured to acquire multiple maintenance text data of the target model device, and each maintenance text data includes a maintenance record, a recording time of the maintenance record, and a maintenance cost;
[0130] A vectorization representation module 502, configured to perform domain-specific fine-tuning on the text data in the field of drilling equipment maintenance according to a pre-trained BERT model to obtain a fine-tuned model, and perform vectorization processing on each maintenance record through the fine-tuned model to obtain a numerical vector corresponding to each maintenance record;
[0131] A clustering module 503, configured to perform clustering processing on all maintenance records according to the numerical vectors corresponding to each maintenance record to obtain maintenance record clusters;
[0132] A determination module 504, configured to determine the class label of each maintenance record according to the maintenance record cluster and the maintenance cost corresponding to each maintenance record;
[0133] A calculation module 505, configured to calculate unknown parameters of the Markov process model according to the maintenance time and class label of each maintenance record to obtain a device maintenance cost prediction model.
[0134] In some embodiments, the vectorization representation module 502 is further configured to:
[0135] For each maintenance record, perform preprocessing on the maintenance record to obtain text data, and the preprocessing includes case conversion, special character removal, spelling check correction, and lemmatization;
[0136] According to the pre-trained BERT model, domain-specific fine-tuning is performed on the text data in the field of drilling equipment maintenance to obtain a fine-tuned model, and the numerical vector corresponding to the text data is obtained through the fine-tuned model.
[0137] In some embodiments, the clustering module 503 is further configured to:
[0138] Cluster all the numerical vectors according to different initial cluster numbers and the corresponding clustering criterion functions for each initial cluster number to obtain a k-WCSS graph;
[0139] Obtain the target cluster number according to the elbow in the k-WCSS graph;
[0140] Obtain the maintenance record clusters corresponding to the target cluster number according to the target cluster number.
[0141] In some embodiments, the determination module 504 is further configured to:
[0142] Obtain the maintenance cost distribution interval of each maintenance record cluster according to the maintenance cost of each maintenance record;
[0143] Obtain the class label of each maintenance record cluster according to each maintenance cost distribution interval;
[0144] Obtain the class label of each maintenance record according to the class label of each maintenance record cluster.
[0145] In some embodiments, the training module 505 is further configured to:
[0146] Determine the change of the class label of the target model equipment over the maintenance time according to the maintenance time and class label of each maintenance record;
[0147] Determine every two adjacent class labels and the time interval between every two adjacent class labels according to the change;
[0148] Perform unknown parameter calculation on the Markov process model according to every two adjacent class labels and the time interval between every two adjacent class labels to obtain the equipment maintenance cost prediction model.
[0149] In some embodiments, the calculation module 505 is further configured to:
[0150] Determine the transition probability matrix of the class label of the target model equipment changing over time according to every two adjacent class labels and the time interval between every two adjacent class labels;
[0151] Perform unknown parameter calculation on the Markov process model according to the preset starting state and transition probability matrix to obtain the equipment maintenance cost prediction model.
[0152] Figure 6 This is a schematic structural diagram of a drilling equipment maintenance cost assessment device provided by an embodiment of the present application. The device 60 includes:
[0153] A determination time module 601, configured to determine the time to be predicted for the target model equipment;
[0154] A prediction module 602, configured to input the time to be predicted and the starting state into the equipment maintenance cost prediction model, and obtain a prediction result of the target model equipment at the time to be predicted. The prediction result includes each category label and the probability result corresponding to each category label. The equipment maintenance cost prediction model is a model obtained according to the method of the present application;
[0155] A category label module 603, configured to determine the category label of the target model equipment at the time to be predicted according to the probability result.
[0156] Figure 7 This is a schematic structural diagram of an electronic device provided by an embodiment of the present application. As Figure 7 shown, the electronic device 70 includes:
[0157] The electronic device 70 may include a processor 701 with one or more processing cores, a memory 702 with one or more computer-readable storage media, a communication component 703, and other components. Among them, the processor 701, the memory 702, and the communication component 703 are connected through a bus 704.
[0158] In a specific implementation process, at least one processor 701 executes the computer execution instructions stored in the memory 702, so that at least one processor 701 executes the method as described above.
[0159] For the specific implementation process of the processor 701, reference may be made to the above method embodiment. Its implementation principle and technical effect are similar, and will not be elaborated here in this embodiment.
[0160] In the above Figure 7 shown embodiment, it should be understood that the processor may be a central processing unit (English: Central Processing Unit, abbreviated: CPU), and may also be other general-purpose processors, digital signal processors (English: Digital Signal Processor, abbreviated: DSP), application specific integrated circuits (English: Application SpecificIntegrated Circuit, abbreviated: ASIC), etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The steps of the method disclosed in combination with the invention can be directly embodied as being executed by a hardware processor, or executed by a combination of hardware and software modules in the processor.
[0161] The memory may include a random access memory (RAM), and may also include a non-volatile memory (NVM), such as at least one disk memory.
[0162] The bus may be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, an Extended Industry Standard Architecture (EISA) bus, or the like. The bus can be divided into an address bus, a data bus, a control bus, etc. For the sake of easy representation, the buses in the drawings of this application are not limited to only one bus or one type of bus.
[0163] In some embodiments, a computer program product is also proposed, including a computer program or instruction, and when the computer program or instruction is executed by a processor, the steps in any of the above methods are implemented.
[0164] For the specific implementation of each of the above operations, reference may be made to the previous embodiments and will not be elaborated herein.
[0165] Those of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructions, or by controlling relevant hardware through instructions. The instructions can be stored in a computer-readable storage medium and loaded and executed by a processor.
[0166] Therefore, an embodiment of this application provides a computer-readable storage medium, in which multiple instructions are stored, and the instructions can be loaded by a processor to execute the steps in any of the methods provided by the embodiments of this application.
[0167] Among them, the storage medium may include: a read-only memory (ROM), a random access memory (RAM), a disk, an optical disc, or the like.
[0168] According to one aspect of this application, a computer program product or a computer program is provided. The computer program product or the computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium.
[0169] Since the instructions stored in the storage medium can execute the steps in any of the methods provided by the embodiments of the present application, the beneficial effects achievable by any of the methods provided by the embodiments of the present application can be realized. For details, refer to the previous embodiments and will not be elaborated here.
[0170] Those skilled in the art will readily conceive of other embodiments of the present application after considering the specification and practicing the invention disclosed herein. The present application is intended to cover any variations, uses, or adaptations of the present application, which follow the general principles of the present application and include known common knowledge or conventional technical means in the technical field not disclosed in the present application. The specification and embodiments are only regarded as exemplary, and the true scope and spirit of the present application are pointed out by the following claims.
[0171] It should be understood that the present application is not limited to the exact structures already described and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the present application is only limited by the appended claims.
Claims
1. A model generation method, characterized in that, Including: Obtain multiple maintenance text data of the target model device, each maintenance text data including a maintenance record, the recording time of the maintenance record, and the maintenance cost; According to the pre-trained BERT model, perform domain-specific fine-tuning on the text data in the field of drilling equipment maintenance to obtain a fine-tuned model, and vectorize each maintenance record through the fine-tuned model to obtain a numerical vector corresponding to each maintenance record; Perform clustering processing on all maintenance records according to the numerical vector corresponding to each maintenance record to obtain a maintenance record cluster; Determine the class label of each maintenance record according to the maintenance record cluster and the maintenance cost corresponding to each maintenance record; Perform unknown parameter calculation on the Markov process model according to the maintenance time and class label of each maintenance record to obtain a device maintenance cost prediction model.
2. The method according to claim 1, characterized in that, The step of, according to the pre-trained BERT model, performing domain-specific fine-tuning on the text data in the field of drilling equipment maintenance to obtain a fine-tuned model, and vectorizing each maintenance record through the fine-tuned model to obtain a numerical vector corresponding to each maintenance record, includes: Preprocess each maintenance record to obtain text data, and the preprocessing includes case conversion, special character removal, spelling check correction, and lemmatization; According to the pre-trained BERT model, perform domain-specific fine-tuning on the text data in the field of drilling equipment maintenance to obtain a fine-tuned model, and perform vector representation on the text data through the fine-tuned model to obtain a numerical vector corresponding to the text data.
3. The method according to any one of claims 1-2, characterized in that, The step of, performing clustering processing on all maintenance records according to the numerical vector corresponding to each maintenance record to obtain a maintenance record cluster, includes: Perform clustering processing on all numerical vectors according to different initial clustering class numbers and the clustering criterion function corresponding to each initial clustering class number to obtain a k-WCSS graph; Obtain the target clustering class number according to the elbow in the k-WCSS graph; Obtain a maintenance record cluster corresponding to the target clustering class number according to the target clustering class number.
4. The method according to any one of claims 1-2, characterized in that The step of, determining the class label of each maintenance record according to the maintenance record cluster and the maintenance cost corresponding to each maintenance record, includes: Obtain the maintenance cost distribution interval of each maintenance record cluster according to the maintenance cost of each maintenance record; Obtain the class label of each maintenance record cluster according to each maintenance cost distribution interval; Obtain the class label of each maintenance record according to the class label of each maintenance record cluster.
5. The method according to any one of claims 1-2, characterized in that, The step of, performing unknown parameter calculation on the Markov process model according to the maintenance time and class label of each maintenance record to obtain a device maintenance cost prediction model, includes: Determine the change situation of the class label of the target model device with the maintenance time according to the maintenance time and class label of each maintenance record; Determine every two adjacent class labels and the time interval between every two adjacent class labels according to the change situation; Perform unknown parameter calculation on the Markov process model according to every two adjacent class labels and the time interval between every two adjacent class labels to obtain a device maintenance cost prediction model.
6. The method according to claim 5, wherein Calculating unknown parameters of the Markov process model based on every two adjacent category labels and the time interval between every two adjacent category labels to obtain a device maintenance cost prediction model, including: Determining a transition probability matrix of the category labels of the target model device changing over time according to every two adjacent category labels and the time interval between every two adjacent category labels; Calculating unknown parameters of the Markov process model according to a preset starting state and the transition probability matrix to obtain a device maintenance cost prediction model.
7. A method for evaluating the maintenance cost of a drilling device based on mining text records, characterized in that, By obtaining multiple maintenance text data of the target model device, the future maintenance cost of the drilling device can be predicted using this model, including: Determining the time to be predicted for the target model device; Inputting the time to be predicted and the starting state into the device maintenance cost prediction model to obtain a prediction result of the target model device at the time to be predicted, where the prediction result includes each category label and the probability result corresponding to each category label, and the device maintenance cost prediction model is a model obtained by the method according to any one of claims 1-6; Determining the state of the target model device at the time to be predicted according to the probability result.
8. A model generation device, characterized in that, Including: An acquisition module for acquiring multiple maintenance text data of the target model device, each maintenance text data including a maintenance record, the recording time of the maintenance record, and the maintenance cost; A vectorization representation module for performing domain-specific fine-tuning on the text data in the field of drilling equipment maintenance according to a pre-trained BERT model to obtain a fine-tuned model, and performing vectorization processing on each maintenance record through the fine-tuned model to obtain a numerical vector corresponding to each maintenance record; A clustering module for clustering all maintenance records according to the numerical vector corresponding to each maintenance record to obtain a maintenance record cluster; A determination module for determining the category label of each maintenance record according to the maintenance record cluster and the maintenance cost corresponding to each maintenance record; A calculation module for calculating unknown parameters of the Markov process model according to the maintenance time and category label of each maintenance record to obtain a device maintenance cost prediction model.
9. An electronic device, characterized in that, Including: A processor and a memory communicatively connected to the processor; The memory stores computer execution instructions; The processor executes the computer execution instructions stored in the memory to implement the method according to any one of claims 1-7.
10. A computer-readable storage medium, characterized in that, Computer execution instructions are stored in the computer-readable storage medium, and when the computer execution instructions are executed by the processor, they are used to implement the method according to any one of claims 1-7.