Maintenance cycle identification method and device based on machine learning

Abstract

The embodiment of the present invention provides a maintenance cycle identification method and device based on machine learning, which belongs to the field of equipment safety management technology. The method includes: obtaining a first operation data set of a target device, the first operation data set includes all sub-operation data of the target device, and the sub-operation data includes the operation date of the target device and the operation volume corresponding to the operation date; taking the set of sub-operation data of the target device with abnormal operation volume as the second operation data set; calculating the maintenance cycle of the target device based on the second operation data set and outputting it. The above technical solution of the present invention collects the operation data of the target device, determines whether the daily operation volume of the target device is abnormal based on the operation date and operation volume of the target device, and determines the maintenance cycle of the target device based on the screened operation data with abnormal daily operation volume and outputs it, so that the maintenance cycle of the target device can be quickly identified by only collecting the operation data of the target device, without consuming a lot of manpower.

Description

Technical Field

[0001] The present invention relates to the technical field of equipment safety management, and in particular to a maintenance cycle identification method based on machine learning and a maintenance cycle identification device based on machine learning. Background Art

[0002] Special operations at petrochemical companies are prone to accidents. Major overhauls often involve shutting down multiple units across different departments, resulting in a massive workload. Any delays in commissioning and production can lead to significant operating losses. Statistically identifying the overhaul cycles for units involved in special operations facilitates planned risk assessments for these units. However, currently, faced with this enormous amount of special operations data, companies often manually identify the start and end dates for major overhauls, requiring significant effort. Summary of the Invention

[0003] The purpose of the embodiment of the present invention is to calculate the maintenance period of a petrochemical device by collecting the operating data of the petrochemical device, so as to solve the problem of the existing technology of manually identifying the maintenance period and consuming a lot of manpower.

[0004] In order to achieve the above object, in a first aspect of the present invention, a maintenance cycle identification method based on machine learning is provided, which is characterized by comprising:

[0005] Acquire a first job data set of a target device, wherein the first job data set includes all sub-job data of the target device, and the sub-job data includes an operation date of the target device and a job volume corresponding to the operation date;

[0006] using a set of sub-job data of the target device with abnormal job volume as a second job data set;

[0007] The maintenance period of the target device is calculated based on the second operation data set and outputted.

[0008] Optionally, the taking the set of sub-job data of the target device with abnormal job volume as the second job data set includes:

[0009] Taking all sub-job data of the first job data set as input, predicting whether the workload of each sub-job data is abnormal using a preset workload abnormality prediction model; and

[0010] Taking the obtained set of sub-job data with abnormal job amounts as the second job data set;

[0011] The workload abnormality prediction model is constructed based on the Kmeans algorithm or the isolation forest algorithm.

[0012] Optionally, the calculating and outputting the maintenance period of the target device based on the second operation data set includes:

[0013] sorting all sub-job data in the second job data set in the order of job date;

[0014] Traversing all sub-job data in the second job data set, and taking a set of all sub-job data for which the difference in days between two adjacent job dates is less than a first threshold as a third job data set;

[0015] Dividing the sub-job data in the third job data set into m data blocks;

[0016] Filter out all data blocks that meet the preset conditions;

[0017] The earliest operation date to the latest operation date of all sub-operation data in each filtered data block is taken as a maintenance cycle of the target device, and the maintenance cycles corresponding to all data blocks are output in sequence.

[0018] Optionally, dividing the sub-job data in the third job data set into m data blocks includes:

[0019] sorting all sub-job data in the third job data set in the order of job date;

[0020] Among all the operation dates, the operation date whose difference with the previous operation date is not less than the first threshold is used as the dividing point;

[0021] All sub-job data in the third job data set are divided into m data blocks based on the dividing point.

[0022] Optionally, screening out all data blocks that meet a preset condition includes:

[0023] For each data block:

[0024] If the difference in number of days between the earliest job date of all sub-job data in the data block and the latest job date of all sub-job data is greater than a second threshold, and the average job volume of each sub-job data in the data block is greater than a third threshold, and the ratio of the second threshold to the number of all sub-job data in the data block is less than a fourth threshold, the data block is retained; otherwise, the data block is discarded.

[0025] Optionally, the method further includes:

[0026] Calculate the maintenance planning risk value for each maintenance cycle; and

[0027] The average of the maintenance planning risk values ​​of all maintenance cycles is used as the maintenance planning risk value of the target device.

[0028] Optionally, after taking the average of the maintenance planning risk values ​​of all maintenance cycles as the maintenance planning risk value of the target device, the method further includes:

[0029] Outputting the risk level corresponding to the maintenance plan risk value of the target device according to the maintenance plan risk value of the target device and a preset risk assessment table;

[0030] The risk assessment table includes risk levels corresponding to different intervals of the maintenance planning risk value of the target device.

[0031] Optionally, the calculating of the maintenance planning risk value of each maintenance cycle includes:

[0032] The maintenance planning risk value of each maintenance cycle is calculated according to the following formula:

[0033]

[0034] Among them, K is the maintenance plan risk value, n is the number of sub-operation data in the current maintenance cycle, x i is the workload of the ith sub-job data in the current maintenance cycle, is the mean of the workload of all sub-job data in the current maintenance cycle, and s is the standard deviation.

[0035] In a second aspect of the present invention, a maintenance cycle identification device based on machine learning is provided, comprising:

[0036] a data acquisition module configured to acquire a first job data set of a target device, wherein the first job data set includes all sub-job data of the target device, and the sub-job data includes an operation date of the target device and an operation amount corresponding to the operation date;

[0037] a workload anomaly calculation module configured to use a set of sub-job data of workload anomalies of the target device as a second job data set;

[0038] The maintenance cycle calculation module is configured to calculate and output the maintenance cycle of the target device based on the second operation data set.

[0039] Optionally, the workload abnormality calculation module includes:

[0040] a prediction unit configured to take all sub-job data of the first job data set as input and predict whether the workload of each sub-job data is abnormal using a preset workload abnormality prediction model; and

[0041] Taking the obtained set of sub-job data with abnormal job amounts as the second job data set;

[0042] The workload abnormality prediction model is constructed based on the Kmeans algorithm or the isolation forest algorithm.

[0043] Optionally, the maintenance cycle calculation module includes:

[0044] a first computing unit configured to sort all sub-job data in the second job data set in order of job date;

[0045] a second computing unit configured to traverse all sub-job data in the second job data set, and use a set of all sub-job data for which the difference in days between two adjacent job dates is less than a first threshold as a third job data set;

[0046] a third computing unit, configured to divide the sub-job data in the third job data set into m data blocks;

[0047] A fourth computing unit is configured to filter out all data blocks that meet a preset condition;

[0048] The fifth calculation unit is configured to take the earliest operation date to the latest operation date of all sub-operation data in each filtered data block as a maintenance cycle of the target device, and output the maintenance cycles corresponding to all data blocks in sequence.

[0049] Optionally, the third computing unit includes:

[0050] A sorting subunit is configured to sort all sub-job data in the third job data set in the order of job date;

[0051] The demarcation point calculation subunit is configured to use, among all the operation dates, the operation date whose difference from the previous operation date is greater than a first threshold as the demarcation point;

[0052] The data block division subunit is configured to divide all sub-job data in the third job data set into m data blocks based on the dividing point.

[0053] Optionally, the fourth computing unit includes:

[0054] The first computing subunit is configured to:

[0055] For each data block:

[0056] If the difference in number of days between the earliest job date of all sub-job data in the data block and the latest job date of all sub-job data is greater than a second threshold, and the average job volume of each sub-job data in the data block is greater than a third threshold, and the ratio of the second threshold to the number of all sub-job data in the data block is less than a fourth threshold, the data block is retained; otherwise, the data block is discarded.

[0057] Optionally, the device further comprises:

[0058] a risk calculation module configured to calculate a maintenance planning risk value for each maintenance cycle; and

[0059] The average of the maintenance planning risk values ​​of all maintenance cycles is used as the maintenance planning risk value of the target device.

[0060] Optionally, the risk calculation module is further configured to:

[0061] Outputting the risk level corresponding to the maintenance plan risk value of the target device according to the maintenance plan risk value of the target device and a preset risk assessment table;

[0062] The risk assessment table includes risk levels corresponding to different intervals of the maintenance planning risk value of the target device.

[0063] Optionally, the risk assessment calculation module includes:

[0064] The sixth calculation unit is configured to calculate the maintenance planning risk value of each maintenance cycle according to the following formula:

[0065]

[0066] Among them, K is the maintenance plan risk value, n is the number of sub-operation data in the current maintenance cycle, x i is the workload of the ith sub-job data in the current maintenance cycle, is the mean of the workload of all sub-job data in the current maintenance cycle, and s is the standard deviation.

[0067] In a third aspect of the present invention, a computer-readable medium is provided, wherein the computer-readable medium stores a computer program, and when the computer program is executed by a processor, the steps of the above method are implemented.

[0068] In a fourth aspect of the present invention, a terminal device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the computer program.

[0069] The above technical solution of the present invention collects the operating data of the target device, determines whether the daily operating volume of the target device is abnormal based on the operating date and operating volume of the target device, and determines and outputs the maintenance cycle of the target device based on the screened operating data with abnormal daily operating volume. Therefore, it is only necessary to collect the operating data of the target device to quickly identify the maintenance cycle of the target device without consuming a lot of manpower.

[0070] Other features and advantages of the embodiments of the present invention will be described in detail in the subsequent detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] The accompanying drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the following detailed description, they are used to explain the embodiments of the present invention, but do not constitute a limitation of the embodiments of the present invention. In the accompanying drawings:

[0072] Figure 1 This is a flow chart of a maintenance cycle identification method based on machine learning provided in a preferred embodiment 1 of the present invention;

[0073] Figure 2 This is a schematic block diagram of the structure of a maintenance cycle identification device based on machine learning provided in a preferred embodiment 2 of the present invention;

[0074] Figure 3 This is the terminal device provided by the preferred embodiment 2 of the present invention.

[0075] Description of Reference Numerals

[0076] 10-terminal device, 100-processor, 101-memory, 102-computer program, 210-data acquisition module, 220-workload abnormality calculation module, 230-maintenance cycle calculation module, 240-risk calculation module. DETAILED DESCRIPTION

[0077] The following describes the specific embodiments of the present invention in detail with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only used to illustrate and explain the present invention and are not intended to limit the present invention.

[0078] Example 1

[0079] like Figure 1 As shown, in a first aspect of this embodiment, a maintenance cycle identification method based on machine learning is provided, characterized by comprising:

[0080] S100: Acquire a first job data set of a target device, where the first job data set includes all sub-job data of the target device, where the sub-job data includes an operation date of the target device and a job volume corresponding to the operation date;

[0081] S200, using a set of sub-job data of abnormal job load of the target device as a second job data set;

[0082] S300: Calculate and output the maintenance period of the target device based on the second operation data set.

[0083] In this way, this embodiment collects the operating data of the target device, determines whether the daily operating volume of the target device is abnormal based on the operating date and operating volume of the target device, and determines and outputs the maintenance cycle of the target device based on the filtered operating data of abnormal daily operating volume. Therefore, it is only necessary to collect the operating data of the target device to quickly identify the maintenance cycle of the target device without consuming a lot of manpower.

[0084] Specifically, special operations in petrochemical enterprises are accident-prone areas, and workers must strictly follow the work ticket when performing operations. The work ticket often includes data such as the operation date, operation volume, work ticket number, operation task, operation sequence, device location, and device number. This data can often be directly read through the petrochemical enterprise's enterprise operation permit management system, or manually input from paper documents. This is not limited in this embodiment. In step S100, in order to accurately filter the operation data corresponding to the petrochemical device, the obtained work ticket data needs to be filtered. In this embodiment, if the operation is a petrochemical device, the work ticket data includes device location data. If the operation is a non-petrochemical device, the device location data is empty. Therefore, after reading the work ticket data, it is necessary to filter out the work ticket data with an empty device location and discard the filtered work ticket data. In this way, the retained work ticket data are all work ticket data related to the petrochemical device operation. Taking the maintenance cycle identification of any target device as an example, all job ticket data with the same device number are obtained, and the device is taken as the target device. Further, all operation dates and the workload corresponding to the operation dates of the target device are obtained. Each operation date and the workload corresponding to the operation date constitute a sub-operation data. For example, in a sub-operation data, it includes [2012-02-29, 12], where 2012-02-29 is the operation date and 12 is the workload corresponding to the operation date; all sub-operation data constitute the first operation data set of the target device. In step S200, a predetermined method is used to determine whether any sub-operation data contains abnormal workloads, i.e., sub-operation data with abnormally high workloads. Since maintenance of a petrochemical plant involves multiple operations, which can significantly increase the workload, if sub-operation data with abnormally high workloads exists, the target plant may be undergoing maintenance. For example, if the sub-operation data [2012-02-29, 12] contains an abnormally high workload of 12, this indicates that the target plant performed maintenance on the operation date of 2012-02-29. A second operation dataset is constructed using the collection of all sub-operation data with abnormally high workloads. Based on this second operation dataset, the maintenance cycle of the target plant can be calculated and output.

[0085] This embodiment uses unsupervised machine learning to determine abnormal workload. Therefore, in step S200, a set of sub-job data of abnormal workload of the target device is used as the second job data set, including:

[0086] Taking all sub-job data of the first job data set as input, the preset workload abnormality prediction model is used to predict whether the workload of each sub-job data is abnormal; and the set of all sub-job data with abnormal workloads is used as the second job data set; the workload abnormality prediction model is constructed based on the Kmeans algorithm or the isolation forest algorithm.

[0087] Taking the Isolation Forest algorithm as an example, the Isolation Forest algorithm is an ensemble-based anomaly detection method with linear time complexity, high accuracy, and fast processing of large data sets. It currently has a wide range of industrial applications, such as attack detection in network security, fraud detection in financial transactions, disease detection, and noisy data filtering. The Isolation Forest algorithm is pre-trained using historical sub-job data from a target device. This historical sub-job data includes the workload of the target device during normal operation and the workload of the target device during maintenance operations, where the workload of the target device during maintenance operations is significantly greater than the workload of the target device during normal operation. This constructs a workload anomaly prediction model. Taking all sub-job data in the first job dataset as input, the workload anomaly prediction model outputs sub-job data with abnormally high workloads, thereby using the set of sub-job data with abnormally high workloads as the second job dataset. The Isolation Forest algorithm is a state-of-the-art technique, and its specific training and anomaly detection processes are not detailed here.

[0088] Since the abnormally high workload of the sub-job data may be caused by temporary maintenance or other operations, in order to eliminate the interference of temporary maintenance or other operations on the identification of the maintenance cycle of the target device, in step S300, the maintenance cycle of the target device is calculated based on the second job data set and output, including:

[0089] S301. Sort all sub-job data in the second job data set in order of job date. In this embodiment, the sub-job data are arranged in order from earliest to latest job date.

[0090] S302. All sub-operation data in the second operation data set are traversed, and the set of all sub-operation data for which the difference in the number of days between two adjacent operation dates is less than a first threshold value T1 is used as the third operation data set. Since abnormally high operation volume in a petrochemical plant is often reflected continuously in the maintenance cycle along with the operation dates, if the difference in the number of days between two adjacent operation dates with abnormally high operation volume is not less than T1, it is considered that the abnormally high operation volume is caused by temporary maintenance, and the data needs to be discarded. For example, sub-operation data 1 [2012-02-29, 12], sub-operation data 2 [2012-03-01, 16], sub-operation data 3 [2012-03-03, 17] and sub-operation data 4 [2012-03-17, 16] are adjacent sub-operation data. Let T1 = 3, then sub-operation data 1, sub-operation data 2, and sub-operation data 3 meet the conditions, indicating that sub-operation data 1, sub-operation data 2 and sub-operation data 3 belong to the same maintenance cycle, and sub-operation data 3 and Sub-operation data 4 does not meet the conditions, indicating that sub-operation data 4 and sub-operation data do not belong to the same maintenance cycle; if there is sub-operation data 5 [2012-03-18, 16] adjacent to sub-operation data 4, it means that sub-operation data 4 and sub-operation data 5 belong to the same maintenance cycle. If sub-operation data 5 is [2012-03-25, 16], it means that sub-operation data 4 and sub-operation data 5 do not belong to the same maintenance cycle. Sub-operation data 4 is discarded, and so on, to obtain the third operation data set.

[0091] S303: Divide the sub-job data in the third job data set into m data blocks. Dividing the sub-job data in the third job data set into m data blocks includes: sorting all sub-job data in the third job data set in order of job date; using the job date whose difference from the previous job date is not less than a first threshold as a demarcation point; and dividing all sub-job data in the third job data set into m data blocks based on the demarcation point. Referring to step S302, if sub-job data 1, sub-job data 2, and sub-job data 3 are consecutive and adjacent sub-job data, then job data 1, sub-job data 2, and sub-job data 3 are considered a single data block. If sub-job data 6 [2012-03-27, 16] and sub-job data 7 [2012-03-28, 14] also exist in the third job data set, then sub-job data 6 and sub-job data 7 are considered a single data block. Similarly, the sub-job data are divided into m data blocks. Simultaneously, based on A list Lm is generated for each data block, where m represents the mth data block. Each list Lm includes the sub-job data in the mth data block, and the sub-job data in Lm are arranged in descending order by job date. For example, data block 1 includes sub-job data 1, sub-job data 2, and sub-job data 3, and the generated list is L1. L1 then includes [2012-02-29, 12], [2012-03-01, 16], and [2012-03-03, 17]. Similarly, L2, ..., and Lm are obtained.

[0092] S304. Filter out all data blocks that meet the preset conditions. Specifically, for each data block: if the difference in days between the earliest operation date of all sub-operation data in the data block and the latest operation date of all sub-operation data is greater than the second threshold value T2, and the average operation volume of each sub-operation data in the data block is greater than the third threshold value V1, and the ratio of the second threshold value to the number of all sub-operation data in the data block is less than the fourth threshold value R, then retain the data block; otherwise, discard the data block. Since the maintenance cycle of petrochemical equipment in actual production is generally 7 days or 1 month, T2 can be set to 6. For L1 to Lm obtained in step S303, if L1 includes [2012-02-29, 12], [2012-03-01, 16], and [2012-03-03, 17], since the earliest operation date in L1 is 2012-02-29 and the latest operation date is 2012-03-0 3, then the difference between the dates is 3. Since it does not meet the condition of being greater than T2, it is determined that L1 is not the maintenance period of the target device and the data of L1 is discarded; if L1 includes [2012-02-29, 12], [2012-03-01, 16], [2012-03-03, 17], [2012-03-04, 14], [2012-03-06, 16], [2012-03-08, 12], The earliest operation date in L1 is 2012-02-29, and the latest operation date is 2012-03-08. The difference between the two is greater than T2. ​​Therefore, L1 is determined to be a maintenance cycle for the target device. Furthermore, since the target device has a high workload during the maintenance cycle, to further improve the accuracy of maintenance cycle screening, a third threshold V1 for the maintenance cycle workload is set. The mean workload of Lm is compared with V1. V1 can be determined based on actual conditions. For example, V1 can be the minimum average daily workload of historical maintenance cycles. For example, if the target device has Lm maintenance cycles and the minimum average daily workload of each maintenance cycle is 12, V1 can be set to 12. In this embodiment, the third threshold V1 is set to 14. The average workload of each sub-operation data in L1 is (12+16+17+14+16+12) / 6=14.5, resulting in an average workload of L1 of 14.5, is greater than V1, therefore, it is considered that the workload of L1 meets the workload of the maintenance cycle; at the same time, since petrochemical plants usually need to operate continuously during the maintenance cycle, the ratio of the second threshold T2 to the list length of the maintenance cycle Lm, that is, the number of sub-operation data in Lm, should be less than a set value, such as the fourth threshold R. If the ratio is greater than R, it means that the target device has fewer operating dates in the cycle and does not meet the maintenance cycle conditions; for example, if L1 includes [2012-02-29, 12], [2012-03-01, 16], [2012-03-03, 17], [2012-03-04, 14], [2012-03-06, 16], [2012-03-08, 12], T2 = 6, R = 0.8, and the list length of L1 is 6, then 6 / 6 = 1, which is greater than R. Therefore, it is considered that there are fewer operation dates in L1 and it does not meet the continuous operation condition of the maintenance cycle. The data of L1 is discarded. If L1 includes [2012-02-29, 12], [2012-03-01, 16], [2012-03-02, 17], [2012-03-03, 12], [2012-03-04, 14], [2012-03-06, 16], [2012-03-07, 12], and [2012-03-08, 12], the list length of L1 is 8. 6 / 8 = 0.75, which is less than R. Therefore, it is considered that L1 meets the maintenance cycle condition. The data of L1 is retained. Similarly, step S304 is executed for L1 to Lm, and finally the maintenance cycle Lm' of the target device is screened out.

[0093] S305. Take the earliest operation date to the latest operation date in all the sub-operation data in each filtered data block as a maintenance cycle of the target device, and output the maintenance cycles corresponding to all data blocks in sequence; take the earliest operation date to the latest operation date in the sub-operation data in the maintenance cycle Lm' of the target device obtained in step S304 as a maintenance cycle of the target device, and output the maintenance cycles L1~Lm' in sequence, so as to realize the identification of the maintenance cycle of the target device.

[0094] After the maintenance cycle of the target device is identified, in order to further perform risk assessment on the target device, the method of this embodiment further includes:

[0095] S400, calculating the maintenance planning risk value of each maintenance cycle; and taking the average of the maintenance planning risk values ​​of all maintenance cycles as the maintenance planning risk value of the target device; In step S400, the maintenance planning risk value of each maintenance cycle is calculated according to the following formula: Among them, K is the maintenance plan risk value, n is the number of sub-operation data in the current maintenance cycle, xi is the operation volume of the i-th sub-operation data in the current maintenance cycle, is the mean of the workload of all sub-job data in the current maintenance cycle, and s is the standard deviation.

[0096] S500. Based on the planned maintenance risk value of the target device and a preset risk assessment table, the risk level corresponding to the planned maintenance risk value of the target device is output, wherein the risk assessment table is shown in Table 1; the risk assessment table includes the risk levels corresponding to different intervals of the planned maintenance risk value of the target device. For the obtained planned maintenance risk value, the level of the planned maintenance risk of the target device is assessed according to the data in the risk assessment table, so as to facilitate subsequent improvements by the staff, which is beneficial to the safety management of petrochemical production and can effectively avoid the occurrence of safety accidents.

[0097]

[0098]

[0099] Table 1

[0100] The present embodiment is described below with examples:

[0101] For a large-scale refinery A, there are two units, namely coking unit I and coking unit II. In step S100, the operation ticket data is exported through the enterprise operation permit management system. Part of the operation ticket data and the fields included are shown in Table 2:

[0102] Serial number License Type Job start time Invoicing Department Installation location 1 Fire 2017-01-01 Security Group I coking 2 Fire 2017-01-02 Security Group Ⅱ Coking 3 Working at heights 2017-01-02 Technical Team

[0103] Table 2

[0104] Among them, according to the national "Safety Specifications for Special Operations in Chemical Production Units", there are eight types of work permits, including fire, entering confined spaces, blind plate plugging, high-altitude operations, lifting, temporary power use, earthwork and circuit breaking. The start time of the operation is specified to the day, the invoicing department is the work ticket approval department, the device location is I coking, II coking, and may also be empty.

[0105] In step S200, the data with empty values ​​for the location of the cleaning device is counted according to the device and date. The groupby function in pandas can be used to count according to the device and date, and then a column is added to store the numeric date. The results of the coking device 1 are shown in Table 3:

[0106] Serial number Job start time Working time Daily workload 1 2012-03-03 40971 13 2 2012-03-04 40972 12 3 2012-03-05 40973 8

[0107] Table 3

[0108] For the two coking units I and II, taking coking unit I as an example, two series of data, daily operation volume and operation time of the unit, are selected as input. An unsupervised machine learning algorithm, such as the isolation forest algorithm, is used. The algorithm parameters can be set to the default values. The abnormal operation volume prediction model training is started. The above data is then input into the trained model to calculate whether the daily operation volume of coking unit I is abnormally excessive. In this way, whether the daily operation volume of the two units is abnormally excessive can be obtained. The calculation results of the coking unit I are shown in Table 4:

[0109]

[0110]

[0111] Table 4

[0112] In step S300, for coking units I and II, taking coking unit I as an example, data indicating abnormal daily operating volume for the unit is screened. All dates in the data are looped through, and the number of days between two adjacent dates is calculated. If this value is less than a set threshold value T1 = 7, the date and the corresponding number of operations are output to list Lm, where m is the mth maintenance cycle. Conversely, if the number of days between the earliest and latest dates in the list is less than a set threshold value T2 = 7, and the mean of all operation quantities in list Lm is greater than a set threshold value V1 = 2.6, and the ratio of T2 to the length of list Lm is less than a set threshold value R, then list Lm is output as a maintenance cycle for the unit, m is incremented by one, and the next list is screened. If the above three conditions are not met, the Lm data is cleared. The final identified maintenance data for the unit is L = {Lm}, where m∈{0,1--,N}, and N is the number of maintenance cycles. Repeating the above steps can identify maintenance data for two sets of units. Among them, the I coking unit has L1 = [[2012-02-29, 12], ..., [2012-03-31, 8]], L2 = [[2016-03-11, 17], ..., [2016-04-01, 26]], L3 = [[2018-10-09, 5], ..., [2018-11-02, 4]], a total of three inspection and maintenance cycles, while the II coking unit has no inspection and maintenance cycle.

[0113] Example 2

[0114] like Figure 2 As shown, this embodiment provides a maintenance cycle identification device based on machine learning, including:

[0115] The data acquisition module 210 is configured to acquire a first job data set of a target device, the first job data set including all sub-job data of the target device, the sub-job data including the job date of the target device and the job volume corresponding to the job date;

[0116] The workload anomaly calculation module 220 is configured to use a set of sub-job data of the workload anomaly of the target device as a second job data set;

[0117] The maintenance cycle calculation module 230 is configured to calculate and output the maintenance cycle of the target device based on the second operation data set.

[0118] Optionally, the workload abnormality calculation module 220 includes:

[0119] a prediction unit configured to take all sub-job data of the first job data set as input and predict whether the workload of each sub-job data is abnormal using a preset workload abnormality prediction model; and

[0120] Taking the obtained set of sub-job data with abnormal job amounts as the second job data set;

[0121] The workload anomaly prediction model is built based on the Kmeans algorithm or the isolation forest algorithm.

[0122] Optionally, the maintenance cycle calculation module 230 includes:

[0123] a first computing unit configured to sort all sub-job data in the second job data set in order of job date;

[0124] The second computing unit is configured to traverse all sub-job data in the second job data set, and use a set of all sub-job data whose difference in days between two adjacent job dates is less than a first threshold as a third job data set;

[0125] a third computing unit, configured to divide the sub-job data in the third job data set into m data blocks;

[0126] A fourth computing unit is configured to filter out all data blocks that meet a preset condition;

[0127] The fifth calculation unit is configured to use the earliest operation date to the latest operation date of all sub-operation data in each filtered data block as a maintenance cycle of the target device, and sequentially output the maintenance cycles corresponding to all data blocks.

[0128] Optionally, the third computing unit includes:

[0129] A sorting subunit is configured to sort all sub-job data in the third job data set in the order of job date;

[0130] The demarcation point calculation subunit is configured to use, among all the operation dates, the operation date whose difference from the previous operation date is greater than a first threshold as the demarcation point;

[0131] The data block division subunit is configured to divide all sub-job data in the third job data set into m data blocks based on the dividing point.

[0132] Optionally, the fourth computing unit includes:

[0133] The first computing subunit is configured to:

[0134] For each data block:

[0135] If the difference in number of days between the earliest job date of all sub-job data in the data block and the latest job date of all sub-job data is greater than the second threshold, and the average job volume of each sub-job data in the data block is greater than the third threshold, and the ratio of the second threshold to the number of all sub-job data in the data block is less than the fourth threshold, the data block is retained; otherwise, the data block is discarded.

[0136] Optionally, the device further comprises:

[0137] The risk calculation module 240 is configured to calculate the maintenance planning risk value of each maintenance cycle; and

[0138] The average of the maintenance planning risk values ​​of all maintenance cycles is used as the maintenance planning risk value of the target device.

[0139] Optionally, the risk calculation module 240 is further configured to:

[0140] Output the risk level corresponding to the maintenance plan risk value of the target device based on the maintenance plan risk value of the target device and the preset risk assessment table;

[0141] The risk assessment table includes the corresponding risk levels when the maintenance planning risk value of the target device belongs to different intervals.

[0142] Optionally, the risk assessment calculation module includes:

[0143] The sixth calculation unit is configured to calculate the maintenance planning risk value of each maintenance cycle according to the following formula:

[0144]

[0145] Among them, K is the maintenance plan risk value, n is the number of sub-operation data in the current maintenance cycle, x i is the workload of the ith sub-job data in the current maintenance cycle, is the mean of the workload of all sub-job data in the current maintenance cycle, and s is the standard deviation.

[0146] This embodiment further provides a computer-readable medium, which stores a computer program. When the computer program is processed and executed, the steps of the above method are implemented.

[0147] like Figure 3 As shown, this embodiment further provides a terminal device, the terminal device 10 comprising: a processor 100, a memory 101, and a computer program 102 stored in the memory 101 and executable on the processor 100. When the processor 100 executes the computer program 102, the steps in the above method embodiment are implemented, for example, Figure 1 Alternatively, when the processor 100 executes the computer program 102, the functions of the modules / units in the above-mentioned device embodiments are realized.

[0148] The terminal device 10 may be a computing device such as a desktop computer, a notebook, a PDA, or a cloud server. The terminal device 10 may include, but is not limited to, a processor 100 and a memory 101. Those skilled in the art will understand that Figure 3 It is only an example of the terminal device 10 and does not constitute a limitation of the terminal device 10. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, the terminal device may also include input and output devices, network access devices, buses, etc.

[0149] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0150] The above describes in detail the optional embodiments of the present invention in conjunction with the accompanying drawings. However, the embodiments of the present invention are not limited to the specific details in the above embodiments. Within the technical concept of the embodiments of the present invention, various simple modifications can be made to the technical solutions of the embodiments of the present invention, and these simple modifications all fall within the scope of protection of the embodiments of the present invention.

[0151] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any appropriate manner without contradiction. To avoid unnecessary repetition, the embodiments of the present invention will not further describe various possible combinations.

[0152] Those skilled in the art will appreciate that all or part of the steps in the methods of the aforementioned embodiments can be accomplished by instructing the relevant hardware through a program, which is stored in a storage medium and includes a number of instructions for causing a single-chip microcomputer, chip, or processor to execute all or part of the steps in the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0153] In addition, the various embodiments of the present invention may be arbitrarily combined, and as long as they do not violate the concept of the embodiments of the present invention, they should also be regarded as the contents disclosed in the embodiments of the present invention.

Claims

1. A maintenance cycle identification method based on machine learning, characterized in that: include: Acquire a first job data set of a target device, wherein the first job data set includes all sub-job data of the target device, and the sub-job data includes an operation date of the target device and a job volume corresponding to the operation date; Taking all sub-job data of the first job data set as input, a preset workload abnormality prediction model is used to predict whether the workload of each sub-job data is abnormal; as well as The obtained set of sub-job data with abnormal workload is used as the second job data set, and the workload abnormality prediction model is constructed based on the Kmeans algorithm or the isolation forest algorithm; sorting all sub-job data in the second job data set in the order of job date; Traversing all sub-job data in the second job data set, and taking a set of all sub-job data for which the difference in days between two adjacent job dates is less than a first threshold as a third job data set; Dividing the sub-job data in the third job data set into m data blocks; Filter out all data blocks that meet the preset conditions; The earliest operation date to the latest operation date of all sub-operation data in each filtered data block is taken as a maintenance cycle of the target device, and the maintenance cycles corresponding to all data blocks are output in sequence.

2. The maintenance cycle identification method based on machine learning according to claim 1 is characterized in that: The step of dividing the sub-job data in the third job data set into m data blocks includes: sorting all sub-job data in the third job data set in the order of job date; Among all the operation dates, the operation date whose difference with the previous operation date is not less than the first threshold is used as the dividing point; All sub-job data in the third job data set are divided into m data blocks based on the dividing point.

3. The maintenance cycle identification method based on machine learning according to claim 1 is characterized in that: The method of filtering out all data blocks that meet the preset conditions includes: For each data block: If the difference in number of days between the earliest job date of all sub-job data in the data block and the latest job date of all sub-job data is greater than a second threshold, and the average job volume of each sub-job data in the data block is greater than a third threshold, and the ratio of the second threshold to the number of all sub-job data in the data block is less than a fourth threshold, the data block is retained; otherwise, the data block is discarded.

4. The maintenance cycle identification method based on machine learning according to claim 1 is characterized in that: The method further comprises: Calculate the maintenance planning risk value for each maintenance cycle; and The average of the maintenance planning risk values ​​of all maintenance cycles is used as the maintenance planning risk value of the target device.

5. The maintenance cycle identification method based on machine learning according to claim 4 is characterized in that: After taking the average of the maintenance planning risk values ​​of all maintenance cycles as the maintenance planning risk value of the target device, the method further includes: Outputting the risk level corresponding to the maintenance plan risk value of the target device according to the maintenance plan risk value of the target device and a preset risk assessment table; The risk assessment table includes risk levels corresponding to different intervals of the maintenance planning risk value of the target device.

6. The maintenance cycle identification method based on machine learning according to claim 4 is characterized in that: The calculation of the maintenance planning risk value of each maintenance cycle includes: The maintenance planning risk value of each maintenance cycle is calculated according to the following formula: ； Among them, K is the maintenance plan risk value, n is the number of sub-operation data in the current maintenance cycle, x i is the workload of the ith sub-job data in the current maintenance cycle, is the mean of the workload of all sub-job data in the current maintenance cycle, and s is the standard deviation.

7. A maintenance cycle identification device based on machine learning, characterized in that: include: a data acquisition module configured to acquire a first job data set of a target device, wherein the first job data set includes all sub-job data of the target device, and the sub-job data includes an operation date of the target device and an operation amount corresponding to the operation date; The workload abnormality calculation module includes: a prediction unit configured to take all sub-job data of the first job data set as input and predict whether the workload of each sub-job data is abnormal using a preset workload abnormality prediction model; and The obtained set of sub-job data with abnormal workload is used as the second job data set, and the workload abnormality prediction model is constructed based on the Kmeans algorithm or the isolation forest algorithm; Maintenance cycle calculation module, including: a first computing unit configured to sort all sub-job data in the second job data set in order of job date; a second computing unit configured to traverse all sub-job data in the second job data set, and use a set of all sub-job data for which the difference in days between two adjacent job dates is less than a first threshold as a third job data set; a third computing unit, configured to divide the sub-job data in the third job data set into m data blocks; A fourth computing unit is configured to filter out all data blocks that meet a preset condition; The fifth calculation unit is configured to take the earliest operation date to the latest operation date of all sub-operation data in each filtered data block as a maintenance cycle of the target device, and output the maintenance cycles corresponding to all data blocks in sequence.

8. The maintenance cycle identification device based on machine learning according to claim 7 is characterized in that: The third computing unit includes: A sorting subunit is configured to sort all sub-job data in the third job data set in the order of job date; The demarcation point calculation subunit is configured to use, among all the operation dates, the operation date whose difference from the previous operation date is greater than a first threshold as the demarcation point; The data block division subunit is configured to divide all sub-job data in the third job data set into m data blocks based on the dividing point.

9. The maintenance cycle identification device based on machine learning according to claim 7, characterized in that: The fourth computing unit includes: The first computing subunit is configured to: For each data block: If the difference in number of days between the earliest job date of all sub-job data in the data block and the latest job date of all sub-job data is greater than a second threshold, and the average job volume of each sub-job data in the data block is greater than a third threshold, and the ratio of the second threshold to the number of all sub-job data in the data block is less than a fourth threshold, the data block is retained; otherwise, the data block is discarded.

10. The maintenance cycle identification device based on machine learning according to claim 7, characterized in that: The device further comprises: a risk calculation module configured to calculate a maintenance planning risk value for each maintenance cycle; and The average of the maintenance planning risk values ​​of all maintenance cycles is used as the maintenance planning risk value of the target device.

11. The maintenance cycle identification device based on machine learning according to claim 10, characterized in that: The risk calculation module is further configured to: Outputting the risk level corresponding to the maintenance plan risk value of the target device according to the maintenance plan risk value of the target device and a preset risk assessment table; The risk assessment table includes risk levels corresponding to different intervals of the maintenance planning risk value of the target device.

12. The maintenance cycle identification device based on machine learning according to claim 10, characterized in that: The risk calculation module includes: The sixth calculation unit is configured to calculate the maintenance planning risk value of each maintenance cycle according to the following formula: ； Among them, K is the maintenance plan risk value, n is the number of sub-operation data in the current maintenance cycle, x i is the workload of the ith sub-job data in the current maintenance cycle, is the mean of the workload of all sub-job data in the current maintenance cycle, and s is the standard deviation.

13. A computer-readable medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

14. A terminal device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

Images