WO2023151191A1 - 机械设备画像生成方法 - Google Patents

机械设备画像生成方法 Download PDF

Info

Publication number
WO2023151191A1
WO2023151191A1 PCT/CN2022/090147 CN2022090147W WO2023151191A1 WO 2023151191 A1 WO2023151191 A1 WO 2023151191A1 CN 2022090147 W CN2022090147 W CN 2022090147W WO 2023151191 A1 WO2023151191 A1 WO 2023151191A1
Authority
WO
WIPO (PCT)
Prior art keywords
data
label
equipment
value
tag
Prior art date
Application number
PCT/CN2022/090147
Other languages
English (en)
French (fr)
Inventor
王新梦
王宗文
李海龙
Original Assignee
烟台杰瑞石油服务集团股份有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 烟台杰瑞石油服务集团股份有限公司 filed Critical 烟台杰瑞石油服务集团股份有限公司
Priority to US18/160,876 priority Critical patent/US20230259862A1/en
Publication of WO2023151191A1 publication Critical patent/WO2023151191A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
    • G06Q10/06395—Quality analysis or management
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/04—Manufacturing
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
    • Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
    • Y02P90/30—Computing systems specially adapted for manufacturing

Definitions

  • the present application relates to the technical field of intelligent management of mechanical equipment, and more specifically, to a method for generating an image of mechanical equipment.
  • Portrait technology is widely used in the Internet and other industries, mainly using portraits, which can be used to mine user characteristics, grasp user preferences, and deeply and thoroughly understand and master the portrayed objects.
  • large-scale mechanical equipment can also be portrayed like a human being, and feature extraction and comprehensive description can be completed through portrait technology.
  • a large amount of monitoring and monitoring data are generated during the delivery, production and maintenance of large-scale mechanical equipment. These data can directly reflect the remarkable characteristics of the equipment, but it is impossible to dig out its characteristics intuitively and deeply only through a large amount of data. Therefore, the construction of mechanical equipment portraits can accurately excavate and extract the characteristics of various aspects of mechanical equipment, and present them comprehensively, so that we can keep abreast of the equipment status and operating status.
  • the present application provides a method for generating mechanical equipment portraits, the method includes: obtaining all data information related to mechanical equipment; performing data fusion on data information related to mechanical equipment to obtain a multi-source data information set; based on The multi-source data information collection acquires the target attribute data and target state data of the mechanical equipment; generates the attribute label of the mechanical equipment based on the acquired target attribute data; inputs the current target state data of the mechanical equipment into the trained neural network model , to obtain the current state label of the mechanical equipment, wherein the current state label includes the state label value of the mechanical equipment; through the obtained attribute label value of the attribute label and the obtained state label value of the current state label, a device portrait of the mechanical equipment is generated.
  • the mechanical equipment portrait generation method also includes: initializing the neural network model based on the target state data of the mechanical equipment and the number of target state label values of the target state data; generating multiple statistical indicators based on the target state data; and including multiple The data matrix of the state classification label data of the statistical indicators and the target state data is input into the initialized neural network model, and the neural network model is trained.
  • acquiring the target attribute data and target state data of the mechanical equipment based on the multi-source data information set includes: acquiring initial attribute data and initial state data based on the multi-source data information set; to obtain the target attribute data; normalize the initial state data to obtain the target state data.
  • the equipment portrait of the mechanical equipment including: constructing all tags based on the attribute tag value and the current status tag value of the mechanical equipment Value set and device set; based on the tag value set and device set, construct a co-occurrence matrix; based on the co-occurrence matrix, obtain the label cluster of the attribute label and the current state label through the clustering algorithm; and obtain the current status of the device based on the label cluster Key tag values to generate equipment portraits of mechanical equipment.
  • the current key tag values of the equipment are obtained based on the tag clusters, and the equipment portraits of the mechanical equipment are generated, including: comparing all the tag values of each device with the tag values contained in the acquired tag clusters; The label value covers the first label cluster with the largest number of label value categories in the label cluster, and the corresponding label value in the first label cluster is set as the current key label value of the device.
  • the method for generating a mechanical equipment portrait further includes: visualizing the generated equipment portrait as a picture word cloud.
  • the method for generating a mechanical equipment portrait further includes: comparing the status label value of the current target status data of the mechanical equipment with a set value to obtain the current status label of the mechanical equipment.
  • the state label value of the current target state data of the mechanical equipment is compared with the set value to obtain the current state label of the mechanical equipment.
  • data fusion is performed on data information related to mechanical equipment to obtain a multi-source data information set, including: setting unique identifiers that are different from each other for each equipment; A second library table associated with the library table; and the first library table and the second library table are associated with each other with a unique identifier.
  • the neural network model includes a backpropagation neural network model.
  • the statistical indicators include: mean value, standard deviation, square root amplitude, effective value, peak value, skewness, kurtosis, maximum value, margin value, form factor, and pulse index.
  • status tag values include: normal, fault, high, medium and low.
  • a computer device including a memory and a processor, the memory stores a computer program that can run on the processor, and the processor implements the above-mentioned mechanical device when executing the computer program Steps of the image generation method.
  • a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the steps of the above-mentioned method for generating an image of a mechanical device are realized.
  • the label system constructed in this application includes not only static attributes and instantaneous parameters, but also parameter state evaluation and maintenance state information, and provides a method for constructing equipment current state labels based on neural networks, so as to facilitate the acquisition of equipment Labels such as the current health status can finally obtain a portrait of the equipment, which is helpful for the detection of plunger pump failures, and can grasp the current operating status of the equipment in a timely and accurate manner, which is convenient for controlling the entire production process.
  • Fig. 1 shows a flow chart of a method for generating a mechanical equipment portrait according to a preferred embodiment of the present application
  • Fig. 2 shows a schematic diagram of fusion processing of multi-source data related to mechanical equipment according to an embodiment of the present application
  • FIG. 3 schematically shows an example labeling system of sample data constructed according to the present application
  • FIG. 4 shows a schematic diagram of visualizing a portrait of a mechanical device as a word cloud of a portrait according to a preferred embodiment of the present application
  • Fig. 5 shows a schematic flowchart of a method for generating a mechanical equipment portrait according to a specific embodiment of the present application.
  • a method for generating a mechanical equipment portrait including: acquiring all data information related to the mechanical equipment (S101); performing data fusion on the data information related to the mechanical equipment , to obtain a multi-source data information set (S102); obtain the target attribute data and target state data of the mechanical equipment based on the multi-source data information set (S103); generate the attribute label of the mechanical equipment based on the acquired target attribute data (S104 ); the current target state data of the mechanical equipment is input into the trained neural network model to obtain the current state label of the mechanical equipment, wherein the current state label comprises the state label value (S105) of the mechanical equipment;
  • the attribute tag value and the obtained status tag value of the current status tag are used to generate a device image of the mechanical device ( S106 ).
  • a relatively complete image label system can be constructed for mechanical equipment, such as plunger pump equipment, which helps to grasp the dynamics and operating conditions of the equipment in a timely and accurate manner during the production process.
  • Mechanical equipment is widely used.
  • the plunger pump in the oil and gas equipment service industry is one of the main large-scale mechanical equipment used in the oil and gas industry. It can be used for high-intensity operations such as cementing, acidizing, and fracturing.
  • a large amount of production data and operation and maintenance data are generated in the entire life cycle of mechanical equipment from production landing to on-site operation to maintenance and scrapping.
  • the data involved in the operation of mechanical equipment is diverse and scattered, and different types of data are scattered in major equipment management systems, which need to be sorted out, mainly including (but not limited to the following data): equipment file attribute data , Equipment production operation data, equipment after-sales maintenance data, equipment real-time monitoring system various sensor data.
  • the above four types of data belong to the main data range involved in the plunger pump. Therefore, the construction of the plunger pump portrait index system needs to be realized based on the above data sources.
  • S102 Perform data fusion on the data information related to the mechanical equipment to obtain a multi-source data information set.
  • the data of mechanical equipment-related business systems are scattered in various business systems from multiple sources, and the data is relatively isolated. Data fusion is required to realize the serialization of multi-source data information.
  • the above obtained data information is fused by the following method.
  • a unique identifier is determined as the unique identity symbol of each device.
  • the unique identifier is different for different devices, and may (but not limited to) be the "device number" when the device leaves the factory.
  • the data information here should be all the database tables in the underlying structured database of each business system.
  • the database table structure that is, the fields contained in the database table
  • the The first library table with the unique identifier of the device and the second library table that can be associated with the first library table through other fields are summarized.
  • sort out these two types of database tables for example, sort out how many first database tables with device unique identifiers and second database tables that can be associated with the first database table through other fields for each data source information , and, what is each library table and what fields it contains, and so on.
  • the first library table has a field for the unique device identifier
  • the second library table does not have a field for the unique device identifier, but other fields in the second library table can be associated with the first library table.
  • first library table and second library table can be set in multiples, and the "first library table” and “second library table” are used to distinguish whether the library table has a unique device identifier.
  • Figure 2 shows the fusion processing of multi-source data related to mechanical equipment according to an embodiment of the present application schematic diagram.
  • the "piston pump equipment number” is used as the unique identifier.
  • systems and database tables related to the "piston pump equipment number” are: fracturing truck operation monitoring system, plunger pump intelligent detection system, Cracked car after-sales maintenance management system and equipment production file management library.
  • the equipment operation parameter table in the fracturing vehicle after-sales maintenance management system has 23 fields; the equipment vibration signal record table in the plunger pump intelligent detection system has 9 fields; the equipment maintenance record table in the fracturing vehicle after-sales maintenance management system has 12 fields
  • the above four tables are associated, and duplicate fields are removed. After data fusion, the result table contains 61 fields in total.
  • the required fields can be filtered according to different label requirements and placed in the associated result table, and the fields contained in the above data fusion table can be deleted and queried according to the label calculation needs.
  • the fused device multi-source data information set indexed by the unique device identifier is obtained.
  • device tags are divided into two categories, which are device status hot tags and device attribute cold tags.
  • the device attribute cold tag refers to the tag value that is an inherent attribute of the device or the attribute tag class that will not change once it is generated
  • the device status hot tag refers to the tag value that changes periodically or randomly according to the device status kind. All labels for machinery and equipment can be classified into the above two categories.
  • FIG. 3 it schematically shows an example labeling system of sample data constructed according to the present application.
  • the equipment attribute cold label can include: equipment number, equipment type, equipment production date, equipment production date, equipment rated power, equipment stroke, equipment plunger specification, equipment maximum pressure, equipment maximum displacement, equipment dimensions, equipment weight wait.
  • Equipment status hot labels can include: equipment health status label (label value is normal or failure), equipment current pressure status label (label value includes: high, medium, low), equipment failure risk level, etc.
  • the initial attribute data can include (but not limited to): equipment number, equipment type, equipment production date, equipment commissioning date, equipment rated power, equipment stroke, equipment plunger specification , the maximum pressure of the equipment, the maximum displacement of the equipment, the overall dimensions of the equipment, the weight of the equipment...
  • the standardization processing method can be based on the experience of experts in the field, sort the initial attribute data by attention and importance, and perform a comprehensive score, the score range is between 0 and 1; and remove the comprehensive score between 0.8
  • attribute data with all expert comprehensive scores greater than 0.8 are used as target attribute data.
  • the target attribute data that is, the initial attribute data with an expert score greater than 0.8 according to expert experience, may include (but not limited to): equipment number, equipment type, equipment production Date, date of commissioning of equipment, rated power of equipment, stroke of equipment, specifications of plunger of equipment.
  • the state of the plunger pump can be (but not limited to) the health state (normal or faulty) of the hydraulic end pump valve of the plunger pump, and the corresponding state data can include (but Not limited to): historical hydraulic end single cylinder vibration amplitude data, historical hydraulic end single cylinder temperature value data, historical hydraulic end single cylinder pressure value data.
  • the normalization processing includes performing MIN-MAX normalization processing on the initial state data, so as to obtain the target state data.
  • S104 Generate an attribute label of the equipment based on the acquired target attribute data of the mechanical equipment.
  • the structure is directly named based on the field name of the device target attribute data; and the attribute label data is stored in a standard structured label database.
  • the attribute labels can include (but not limited to): equipment number label, equipment type label, equipment production date label, equipment commissioning date label, equipment rated power label, and equipment stroke label , Equipment plunger specification label.
  • S105 Input the current target state data of the mechanical equipment into the trained neural network model to obtain the current state label of the mechanical equipment.
  • a backpropagation (BP) neural network is used.
  • BP backpropagation
  • other neural networks can also be used.
  • the historical target state data of the device and the number of label values of the state of the device initialize the number of network layers, the number of neurons in each layer, the nonlinear activation function of each layer, the connection weight, the number of single training samples, the number of sample training cycles, and the learning rate , loss function, optimizer and other neural network initial values.
  • the initial network parameters of the neural network are set as follows:
  • parameter name parameter value Network layers 10 The number of neurons in each layer 20-50 pieces, the last layer is 2 pieces non-linear activation function Sigmod connection weight fixed random number Single training sample number 100 Number of sample training loops 800 learning rate 0.001 loss function MSE optimizer Adam
  • the above-mentioned BP neural network is trained through the training data.
  • the status label of the equipment to be classified can be (but not limited to) the health status label of the pump valve at the liquid end of the equipment, and the label value is normal or faulty.
  • the health status data of the pump and valve at the hydraulic end of the equipment includes three indicators: the historical single cylinder vibration amplitude at the hydraulic end, the historical single cylinder temperature value at the hydraulic end, and the historical single cylinder pressure value at the hydraulic end.
  • the example sample data is the 10-hour operating data of the target equipment (that is, the hydraulic end pump and valve of the equipment) under two historical normal and fault states.
  • the original data is divided into 36,000 segments at 1s intervals, and each index is statistically calculated for each segment.
  • Data that is, historical target state data, including but not limited to the historical liquid end single cylinder vibration amplitude, historical liquid end single cylinder temperature value, historical liquid end single cylinder pressure value
  • the 11 statistical index data of the above-mentioned vibration amplitude, the 11 statistical index data of the single-cylinder temperature value, the 11 statistical index data of the single-cylinder pressure value, and the failure or normal data label value of the pump valve at the liquid end of the equipment is used as the column of the matrix, here, there are 11*3+1 columns in total; the 36000 sample data obtained at the interval of 1s under the normal state and the fault state are used as the matrix
  • the data matrix obtained above is divided into a training data set and a testing data set according to a ratio of 9:1.
  • test accuracy is obtained by calculating the AUC value based on the test result output based on the test data and its corresponding real state label. If the test accuracy is less than the set threshold, the model test is completed. If the test accuracy is greater than the threshold, the sample data is re-selected to perform the above-mentioned model training and testing again.
  • the example sample data is based on the above-mentioned experimental data and neural network training steps, and the current health status of the pump valve at the hydraulic end of the target equipment is calculated as failure.
  • the health status of other equipment components or status labels of other equipment such as failure risk levels can also be calculated by the above method.
  • the current status label of the device can also be obtained in a predefined way, that is, for the current target status data of the device, based on the experience of historical experts, the classification of the current status of the device is output as the current status tag of the device. And store the current state label of the equipment into the standard structured label database.
  • the current status label of the equipment can be obtained in a predefined way, which can be but not limited to the current pressure status label of the equipment.
  • the label values include: high, medium, and low.
  • ⁇ 1 and ⁇ 1 are preset as the upper and lower limits of the threshold, and the following label values are set in this predefined way:
  • the label system shown in Figure 3 can be constructed.
  • the tag system may include the following tags: transient parameter tags, such as instantaneous pressure tags, instantaneous temperature tags, instantaneous Saber tags, instantaneous speed tags, etc.; working status tags, such as the current pressure status of equipment , the current temperature state of the equipment, the current Shabi state of the equipment, etc.; vibration feature tags, such as vibration signal time domain amplitude mean, vibration signal time domain amplitude absolute mean, vibration signal time domain amplitude variance, etc.
  • Feature tags such as vibration signal time-domain amplitude mean value, vibration signal time-domain amplitude absolute mean value, vibration signal time-domain amplitude variance and other liquid end vibration feature tags.
  • S106 Generate a device image of the mechanical device by using the generated attribute tag value of the attribute tag and the obtained status tag value of the current status tag.
  • a device portrait is constructed based on a tag value clustering method.
  • the co-occurrence matrix example is as follows:
  • clustering algorithm such as k-medoid, etc., based on the above-mentioned label-device co-occurrence matrix, cluster various labels, and finally obtain the attribute label of the target attribute data and the multi-class label clustering of the current state label.
  • the number of clustering categories can be set as required.
  • the first category contains 10 labels such as the pump and valve health status (fault) label, the crosshead health status (health) label, and the pump and valve pressure status (high) label.
  • the second category contains another 15 labels.
  • Device A has a total of There are 100 current tags. In order to highlight the key points, some tags in the portrait of device A are selected. If the 100 tags of device A include 20 tags of the first category, that is, all 10 tags of the first category, and only 5 tags of the second category, then the tag value of the first category It is the most concentrated and most important type of tags on device A at the current moment, that is, the above-mentioned current key tags. Therefore, a portrait of device A is generated with this first type of label.
  • the device portrait is visualized as a picture word cloud map, as shown in FIG. 4 .
  • the operation status of the equipment can be easily obtained from the figure, so as to facilitate the grasp of the dynamics, operation status and health status of the equipment.
  • Fig. 5 shows a schematic flowchart of a method for generating a mechanical equipment portrait according to a specific embodiment of the present application.
  • data fusion is performed on the acquired equipment portrait data sources, that is, all data information related to mechanical equipment, so as to obtain a multi-source data information set.
  • the obtained multi-source data information set classify the device tags, for example, the above-mentioned device attribute tags and device status tags.
  • preprocessing is performed on the obtained equipment attribute label data and equipment status label data to obtain equipment target attribute labels and equipment target status labels.
  • a device portrait is generated based on the device target attribute label and the device target status label.
  • the current status tag of the device can be obtained through the above-built neural network model, and of course the current status tag of the device can also be obtained through a preset threshold according to the experience of experts in this field. According to actual needs, the above two methods can be selected, or the two methods can be used in combination.
  • a device portrait is constructed by tag value clustering, and these data are presented through a visualization method (for example, a word cloud map).
  • the label system constructed in this application not only includes static attributes and instantaneous parameters, but also includes parameter status evaluation and maintenance status information, and provides a method for constructing equipment current status labels based on neural networks, so as to facilitate the acquisition of equipment Labels such as the current health status can finally obtain a portrait of the equipment, which is helpful for the detection of plunger pump failures, and can grasp the current operating status of the equipment in a timely and accurate manner, which is convenient for controlling the entire production process.
  • Nonvolatile memory can include read only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory.
  • Volatile memory can include random access memory (RAM) or external cache memory.
  • RAM random access memory
  • RAM is available in many forms such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDRSDRAM), Enhanced SDRAM (ESDRAM), Synchronous Chain Synchlink DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Human Resources & Organizations (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Strategic Management (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Economics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • General Business, Economics & Management (AREA)
  • Tourism & Hospitality (AREA)
  • Marketing (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Educational Administration (AREA)
  • Development Economics (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Primary Health Care (AREA)
  • Game Theory and Decision Science (AREA)
  • Manufacturing & Machinery (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Testing And Monitoring For Control Systems (AREA)

Abstract

本发明提供一种机械设备画像生成方法，该方法包括：获取所有与机械设备相关的数据信息（S101）；对机械设备相关的数据信息进行数据融合，以获得多源数据信息集合（S102）；基于多源数据信息集合获取机械设备的目标属性数据和目标状态数据（S103）；基于所获取的所述目标属性数据生成机械设备的属性标签（S104）；将机械设备的当前目标状态数据输入到训练好的神经网络模型中，获得机械设备当前状态标签，其中，当前状态标签包括机械设备的状态标签值（S105）；通过所获取的属性标签的属性标签值和所获得的当前状态标签的状态标签值，生成机械设备的设备画像（S106）。根据本申请的方法，有助于实现柱塞泵故障检测，并能及时准确地掌握设备当前的运行状况，便于掌控整个生产过程。

Description

机械设备画像生成方法
相关申请的引用
本申请要求于2022年2月14日向中华人民共和国国家知识产权局提交的第202210132467.6号中国专利申请的权益，在此将其全部内容以援引的方式整体并入本文中。
技术领域
本申请涉及机械设备智能化管理技术领域，更具体地说，涉及一种机械设备画像生成方法。
背景技术
画像技术被广泛应用于互联网等多行业领域，主要以用画像为主，可以用来挖掘用户特征、掌握用户偏好、以及深入彻底的了解和掌握被刻画对象。同样，大型机械设备也可以像人一样被刻画，也可以通过画像技术进行特征提取和完成全面描述刻画。大型机械设备出厂、生产和维修过程都产生了大量的监控、监测数据，这些数据可以直接反应出设备的显著特征，但是只是通过大量的数据无法直观深入的挖掘出其特征。所以构建机械设备画像，可以准确的挖掘提取机械设备各方面特征、并将其全面的呈现出来，便于我们及时了解掌握设备状态和运行现状。
发明内容
要解决的技术问题
目前为止，对于机械设备画像的研究相对较少，对于油气行业大型机械服务设备的相关研究更少。所以创建完整的机械设备画像是亟待解决的问题。随着设备数据的持续积累和广泛应用，各类大型设备画像将被创建，从而帮助在生产作业过程中及时准确掌握设备动态和运行状况。
技术方案
为了实现上述目的，本申请提供一种机械设备画像生成方法，该方法包括：获取所有与机械设备相关的数据信息；对机械设备相关的数据信息进行数据融合，以获得多源数据信息集合；基于多源数据信息集合获取机械设备的目标属性数据和目标状态数据；基于所获取的所述目标属性数据生成机械设备的属性标签；将机械设备的当前目标状态数据输入到训练好的神经网络模型中，获得机械设备当前状态标签，其中，当前状态标签包括机械设备的状态标签值；通过所获取的属性标签的属性标签值和所获得的当前状态标签的状态标签值，生成机械设备的设备画像。
进一步地，机械设备画像生成方法还包括：基于机械设备的目标状态数据以及目标状态数据的目标状态标签值数量，对神经网络模型进行初始化；基于目标状态数据生成多个统计指标；以及将包括多个统计指标以及目标状态数据的状态分类标签数据的数据矩阵输入到初始化的所述神经网络模型中，对神经网络模型进行训练。
进一步地，基于多源数据信息集合获取机械设备的目标属性数据和目标状态数据，包括：基于多源数据信息集合获取初始属性数据和初始状态数据；基于对初始属性数据的关注度和/或重要性，获取目标属性数据；对初始状态数据进行归一化处理，获取目标状态数据。
进一步地，通过所获取的属性标签的属性标签值和所获得的当前状态标签的状态标签值，生成机械设备的设备画像，包括：基于机械设备的属性标签值和当前状态标签值，构建所有标签值集合和设备集合；基于标签值集合和设备集合，构建共现矩阵；基于共现矩阵，通过聚类算法获取属性标签和当前状态标签的标签聚类簇；以及基于标签聚类簇获取设备当前重点标签值，生成机械设备的设备画像。
进一步地，基于标签聚类簇获取设备当前重点标签值，生成机械设备的设备画像，包括：将各设备所有标签值与所获取的标签聚类簇所含标签值进行比较；以及获取各设备所含标签值覆盖标签聚类簇中标签值类别数最多的第一标签聚类簇，并将第一标签聚类簇中所对应的标签值设为设备当前重点标签值。
进一步地，机械设备画像生成方法还包括：将所生成的设备画像可视化为画像词云图。
进一步地，机械设备画像生成方法还包括：将机械设备的当前目标状态数据的状态标签值与设定值进行比较，获得机械设备的当前状态标签。
进一步地，将机械设备的当前目标状态数据的状态标签值与设定值进行比较，获得机械设备的当前状态标签。
进一步地，对机械设备相关的数据信息进行数据融合，以获得多源数据信息集合，包括：为各设备设定相互不同的唯一标识；获取各设备具有唯一标识的第一库表以及与第一库表相关联的第二库表；以及以唯一标识使得第一库表和第二库表相互关联。
进一步地，神经网络模型包括反向传播神经网络模型。
进一步地，统计指标包括：均值、标准差、方根幅值、有效值、峰值、偏度、峭度、最大值、裕度值、波形因子、脉冲指数。
进一步地，状态标签值包括：正常、故障、高、中和低。
根据本申请的又一方面，提供一种计算机设备，包括存储器及处理器，所述存储器上存储有可在处理器上运行的计算机程序，所述处理器执行所述计算机程序时实现上述机械设备画像生成方法的步骤。
根据本申请的再一方面，提供一种计算机可读存储介质，其上存储有计算机程序，所述计算机程序被处理器执行时实现上述机械设备画像生成方法的步骤。
有益效果
根据本申请，通过将客户画像技术应用到柱塞泵设备中，构建起了较为完善的柱塞泵设备画像标签体系。
并且，本申请中所构建的标签体系既包含了静态属性和瞬时参量，又包含了参量状态评价和维护状态信息，并提供了基于神经网路进行设备当前状态标签的构建方法，从而便于获得设备当前健康状态等标签，最终获得关于设备的画像，有助于实现柱塞泵故障检测，并能及时准确地掌握设备当前的运行状况，便于掌控整个生产过程。
附图说明
构成本申请的一部分的说明书附图用来提供对本发明的进一步理解，本发明的示意性实施例及其说明用于解释本发明，并不构成对本发明的不当限定。在附图中：
图1示出了根据本申请的一优选实施例的机械设备画像生成方法的流程图；
图2示出了根据本申请的一实施例的机械设备相关多源数据的融合处理示意图；
图3示意性示出了根据本申请的所构建的样本数据的示例标签体系；
图4示出了根据本申请的一优选实施例的将机械设备的画像可视化为画像词云图的示意图；
图5示出了根据本申请的一具体实施例的机械设备画像生成方法的流程示意图。
具体实施方式
为了使本申请的目的、技术方案及优点更加清楚明白，以下结合附图及实施例，对本申请进行进一步详细说明。应当理解，此处描述的具体实施例仅仅用以解释本申请，并不用于限定本申请。
根据本申请的一实施例，如图1所示，提供一种机械设备画像生成方法，该方法包括：获取所有与机械设备相关的数据信息(S101)；对机械设备相关的数据信息进行数据融合，以获得多源数据信息集合(S102)；基于多源数据信息集合获取机械设备的目标属性数据和目标状态数据(S103)；基于所获取的所述目标属性数据生成机械设备的属性标签(S104)；将机械设备的当前目标状态数据输入到训练好的神经网络模型中，获得机械设备当前状态标签，其中，当前状态标签包括机械设备的状态标签值(S105)；通过所获取的属性标签的属性标签值和所获得的当前状态标签的状态标签值，生成机械设备的设备画像(S106)。
根据本申请能够为机械设备，诸如，柱塞泵设备，构建起较为完善的画像标签体系，有助于在生产过程中及时准确掌握设备的动态和运行状况。
下面，通过详细的描述来说明机械设备画像生成方法的具体过程。
S101：获取所有与机械设备相关的数据信息。
机械设备应用广泛，例如：油气设备服务行业的柱塞泵就是油气行业主要应用的大型机械设备之一，可以利用其进行固井、酸化、压裂等高强度作业。机械设备从生产落地到现场作业再到维修、报废，整个生命周期中产生大量的生产数据和运维数据。
并且，机械设备运行期间涉及数据是多元的且分散的，不同类型的数据分散在各大设备管理***中，需要对其进行梳理，主要包括(但不仅限于下述数据)：设备档案属性类数据、设备生产作业运行数据、设备售后维保数据、设备实时监控***各类传感器数据。通过对包含上述数据的设备相关业务***底层数据库表、库表字段进行梳理，并以此作为机械设备画像数据源，从而获得所有与各机械设备相关的数据信息。
下面，以压裂井场压裂车载柱塞泵为例来说明获取与该设备相关的数据信息。
对与柱塞泵相关的多源数据以及涉及数据源***进行梳理，获得如下所示的数据库表：
Figure PCTCN2022090147-appb-000001
以上四类数据属于柱塞泵涉及的主要数据范围，因此，构建柱塞泵画像指标体系需要基于上述数据源来实现。
S102：对机械设备相关的数据信息进行数据融合，以获得多源数据信息集合。
机械设备相关业务***数据多源分散在各业务***，数据相对孤立，需要对其进行数据融合，实现多源数据信息的串并。
根据本申请的一优选实施例，通过以下方法对上述获得数据信息进行融合。
优选地，确定一个唯一标识作为每台设备的唯一身份象征，该唯一标识对于不同设备之间互不相同，可以(但不仅限于)是设备出厂时的“设备编号”。
梳理如上所述获得的数据信息中设备相关的所有数据表，该处数据信息应该是各业务***底层结构化数据库中的所有数据库表，根据库表结构(即库表中所含字段)，将具有设备唯一标识的第一库表以及与该第一库表通过其他字段可进行关联的第二库表进行归纳。
具体地，把这两类数据库表进行梳理，例如，梳理每个数据源信息有多少个具有设备唯一标识的第一库表以及与该第一库表通过其他字段可进行关联的第二库表，并且，每个库表分别是什么并且包含哪些字段等等。
第一库表中具有设备唯一标识的字段，第二库表中没有设备唯一标识的字段，但是第二库表中的其他字段可以与第一库表进行关联。
上面所说的第一库表和第二库表均可以设置有多个，使用“第一库表”和“第二库表”用于区分库表是否具有设备唯一标识。
开通上面所涉及业务***数据库(例如第一库表和第二库表等)彼此之间的访问权限，以设备唯一标识作为关联字段，通过数据库语句，例如通过结构化查询语言(sql)语句，对上述中梳理出来的各***数据库表创建关联关系，从而实现以设备唯一标识为索引的设备多源数据有效融合。
以压裂井场压裂车载柱塞泵为例来说明对机械设备相关的数据信息进行数据融合的过程，图2示出了根据本申请的一实施例的机械设备相关多源数据的融合处理示意图。
如图2所示，将“柱塞泵设备编号”作为唯一标识，与“柱塞泵设备编号”相关的***和库表示例为：压裂车作业监控***、柱塞泵智能检测***、压裂车售后维护管理***以及设备生产档案管理库。例如，压裂车售后维护管理***中设备运行参数表共23个字段；柱塞泵智能检测***中设备振动信号记录表共9字段；压裂车售后维护管理***中设备维保记录表共12个字段以及设备生产档案管理库中设备档案信息表共25个字段，基于唯一标识“柱塞泵设备编号”关联上述四张表，除去重复字段，数据融合后，结果表中共含61个字段。
当然，可以根据不同标签需求筛选需要的字段放入关联结果表中，并且可以根据标签计算需要对上述数据融合表所含字段进行删减查询等。
通过如上所述的操作，获得以设备唯一标识为索引的融合后的设备多源数据信息集合。
本申请中，设备标签分为两大类，分别是设备状态热标签类和设备属性冷标签类。其中，设备属性冷标签指的是标签值是设备固有属性或者一旦生成一般不会发生变化的属性标签类；设备状态热标签指的是标签值跟随设备状态发生周期性或不定时随机变化的标签类。机械设备所有标签都可以归入上述两大类中。
例如，如图3所示，其示意性示出了根据本申请的所构建的样本数据的示例标签体系。
例如，设备属性冷标签可以包括：设备编号、设备类型、设备生产日期、设备投产日期、设备额定功率、设备冲程、设备柱塞规格、设备最高压力、设备最大排量、设备外形尺寸、设备重量等。
设备状态热标签可以包括：设备健康状态标签(标签值为正常或故障)、设备当前压力状态标签(标签 值包括：高、中、低)、设备故障风险等级等。
S103：基于上述获得的多源数据信息集合获取机械设备的目标属性数据和目标状态数据。
从上述获得的多源数据信息集合中获取设备的初始属性数据，即，以设备唯一标识为索引，检索上述获得的多源数据信息集合中设备所有相关属性字段，作为初始属性数据。
以压裂井场压裂车载柱塞泵为例，初始属性数据可以包含(但不仅限于)：设备编号、设备类型、设备生产日期、设备投产日期、设备额定功率、设备冲程、设备柱塞规格、设备最高压力、设备最大排量、设备外形尺寸、设备重量……。
对初始属性数据进行标准化处理获得目标属性数据。根据本申请一优选实施例，标准化处理方法可以基于本领域专家经验，对初始属性数据进行关注度和重要性排序，并进行综合评分，评分范围在0与1之间；并且去除综合评分在0.8以下的属性数据，将所有专家综合评分大于0.8的属性数据作为目标属性数据。
以压裂井场压裂车载柱塞泵为例，目标属性数据，即，根据专家经验，专家评分大于0.8分的初始属性数据，可以包括(但不仅限于)：设备编号、设备类型、设备生产日期、设备投产日期、设备额定功率、设备冲程、设备柱塞规格。
从上述获得的多源数据信息集合中获取设备初始状态数据，即以设备唯一标识作为索引，检索上述获得的多源数据信息集合中设备所有与设备状态相关的字段数据，作为初始状态数据。
以压裂井场压裂车载柱塞泵为例，柱塞泵状态可以(但不仅限于)是柱塞泵液力端泵阀健康状态(正常或者故障)，该状态对应状态数据可以包含(但不仅限于)：历史液力端单缸振动幅值数据、历史液力端单缸温度值数据、历史液力端单缸压力值数据。
并且，对初始状态数据进行标准化处理。根据本申请一优选实施例，该标准化处理包括对初始状态数据进行MIN-MAX归一化处理，从而获得目标状态数据。
S104：基于所获取的机械设备的目标属性数据生成设备的属性标签。具体地，例如，基于设备目标属性数据字段名称直接命名构建；并将属性标签数据存入标准结构化标签数据库。
以压裂井场压裂车载柱塞泵为例，属性标签可以包括(但不仅限于)：设备编号标签、设备类型标签、设备生产日期标签、设备投产日期标签、设备额定功率标签、设备冲程标签、设备柱塞规格标签。
S105:将机械设备的当前目标状态数据输入到训练好的神经网络模型中，获得机械设备当前状态标签。
根据本申请的一优选实施例，使用反向传播(BP)神经网络。当然，也可以使用其他神经网络。
根据设备历史目标状态数据以及设备该状态标签值数量，初始化网络层数、各层神经元个数、各层非线性激活函数、连接权值、单次训练样本数、样本训练循环数、学习速率、损失函数、优化器等神经网络初始值。
根据本申请的一优选实施例，神经网络的初始网络参数设置如下：
参数名称 参数值
网络层数 10
各层神经元个数 20-50个，最后一层为2个
非线性激活函数 Sigmod
连接权值 固定随机数
单次训练样本数 100
样本训练循环数 800
学习速率 0.001
损失函数 MSE
优化器 Adam
通过训练数据对上述BP神经网络进行训练。
以压裂井场压裂车载柱塞泵为例说明训练过程。
待分类设备状态标签可以(但不仅限于)是设备液力端泵阀健康状态标签，标签值为正常或者故障。设备液力端泵阀健康状态数据包括：历史液力端单缸振动幅值、历史液力端单缸温度值、历史液力端单缸压力值三个指标。
示例样本数据为目标设备(即，设备液力端泵阀)历史正常和故障两种状态下的各10小时的运行数据，分别以1s为间隔划分原始数据为36000段，统计计算各指标每段数据(即，历史目标状态数据，包括但不限于历史液力端单缸振动幅值、历史液力端单缸温度值、历史液力端单缸压力值)的均值、标准差、方根幅值、有效值、峰值、偏度、峭度、最大值、裕度值、波形因子、脉冲指数11个统计指标数据。
将上述振动幅值的11个统计指标数据、单缸温度值的11个统计指标数据、单缸压力值的11个统计指标数据以及设备液力端泵阀故障或正常数据标签值(即，从设备历史目标状态数据中获得的状态分类标签数据)作为矩阵的列，此处，共11*3+1列；将正常和故障两种状态下，以1s为间隔，获取的36000样本数据作为矩阵的行，总共36000*2行，从而获得(36000*2,11*3+1)大小的数据矩阵。
将上述获得的数据矩阵，按照9：1比例分为训练数据集和测试数据集。
将训练数据集输入构建好的BP神经网络模型，数据在各层网络进行计算后经过激活函数输出到下一层神经网络层，最后一层输出计算结果，将计算结果与真实数据输入到损失函数，损失函数计算损失值，当损失值大于设定阈值时，优化函数根据损失值反向传播值在梯度方向上朝着降低损失值的方向更新各层网络连接权值；当损失函数值小于设定阈值时，神经网络训练结束，保存网络结构和各级神经元信息，从而，得到训练好的神经网络。
上述设定阈值越小越好，优选地，至少要小于0.05。
将上述测试数据集中的当前目标状态数据输入训练好的神经网络模型中，输出测试结果(即，设备当前状态预测标签)和测试精度。测试精度根据基于测试数据输出的测试结果和它对应的真实状态标签，通过计算AUC值获得。如果测试精度小于设定阈值则完成模型测试，如果测试精度大于阈值则重新选择样本数据重新进行上述的模型训练和测试。
基于上述完成训练的神经网络模型，输入设备当前目标状态数据，预测输出设备当前状态分类，作为设备当前状态标签。最后，将设备当前状态标签存入标准结构化标签数据库。
以压裂井场压裂车载柱塞泵为例，示例样本数据基于上述实验数据和神经网络训练步骤，计算得到目标设备当前液力端泵阀健康状态标签为故障。
另外，其他设备部件健康状态或者其他设备的诸如故障风险等级等状态标签，亦可通过上述方法进行计算。
此外，设备当前状态标签还可以通过预定义的方式获取，即针对设备当前目标状态数据，基于历史专家经验，输出设备当前状态分类，作为设备当前状态标签。并将该设备当前状态标签存入标准结构化标签数据库。
以压裂井场压裂车载柱塞泵为例，通过预定义的方式获取设备当前状态标签，可以但不仅限于是设备当前压力状态标签，标签值包括：高、中、低。例如，将α1,β1预设为阈值的上下限，通过该预定义的方式设定如下标签值：
高：设备当前压力>α1
中：β1<＝设备当前压力<＝α1
低：设备当前压力<β1。
另外，以压裂井场压裂车载柱塞泵为例，基于上述画像标签构建方法，可以构建如图3所示的标签体系。如图所示，该标签体系可以包括如下标签：瞬态参数标签类，诸如，瞬时压力标签、瞬时温度标签、瞬时沙比标签、瞬时转速标签等；作业状态标签类，诸如，设备当前压力状态、设备当前温度状态、设备当前沙比状态等；振动特征类标签，例如，诸如振动信号时域幅值均值、振动信号时域幅值绝对均值、振动信号时域幅值方差等的动力端振动特征类标签；诸如振动信号时域幅值均值、振动信号时域幅值绝对均值、振动信号时域幅值方差等的液力端振动特征类标签等等。
S106：通过所生成属性标签的属性标签值和所获得的当前状态标签的状态标签值，生成机械设备的设备画像。
根据本申请的一优选实施例，基于标签值聚类的方法来构建设备画像。
获取机械设备的所有属性标签值和状态标签值，创建所有标签值集合X以及目标区域范围内所有设备集合Y，构建共现矩阵(X,Y)，如果设备Y i具有X j类标签值，则矩阵中对应位置值为1，否则为0，该处， 1，0只是作为各设备和各标签值的对应情况标记，根据各设备与标签值的对应关系作为聚类特征，从而将具有相似对应关系的标签聚为一类，并将每类标签作为具有该类标签的设备的画像。
以压裂井场压裂车载柱塞泵为例，共现矩阵示例如下所示：
Figure PCTCN2022090147-appb-000002
利用诸如k-medoid等的聚类算法，基于上述标签-设备共现矩阵，对各类标签进行聚类，最终获得目标属性数据的属性标签和当前状态标签的多类标签聚类簇，其中，聚类类别数量可以根据需要设定。
对各设备所有标签值与各类标签簇所含标签值进行比较，获取各设备所含标签值覆盖标签值类别数最多的标签聚类簇，并取对应标签聚类簇下标签值作为该设备当前重点标签值，生成该设备的设备画像。
例如，第一类包含：泵阀健康状态(故障)标签、十字头健康状态(健康)标签、泵阀压力状态(高)标签等10个标签，第二类包含另外15个标签，设备A一共具有100个当前标签，为了突出重点，选用设备A的画像中的部分标签。如果设备A的100个标签中包含第一类的20个标签，也就是包含第一类的所有的10个标签，而只包含第二类中的5个标签，那么第一类中的标签值就是设备A当前时刻最集中和最重要的一类标签，即，上面所说的当前重点标签。因此，以该第一类标签生成设备A的画像。
并且，根据本申请的一优选实施例，将设备画像可视化为画像词云图，如图4所示。如图所示，从该图中可以容易获得设备的运行状况，从而便于掌握设备的动态、运行状况及健康状况。
图5示出了根据本申请的一具体实施例的机械设备画像生成方法的流程示意图。
如图所示，根据上面所述的方法获取所有与机械设备相关的数据信息，从而构建设备画像数据源。
接着，根据上面所述的方法，对所获取的设备画像数据源，即，所有与机械设备相关的数据信息，进行数据融合，从而获得多源数据信息集合。
根据所获得的多源数据信息集合，对设备标签进行分类，例如，上面提到的设备属性标签和设备状态标签。
并且，对所获得设备属性标签数据和设备状态标签数据进行预处理，以获得设备目标属性标签和设备目标状态标签。
基于设备目标属性标签和设备目标状态标签来生成设备画像。
在包括属性标签和状态标签的标签中，可以通过上述构建好的神经网络模型来获得设备当前状态标签，当然也可以根据本领域的专家经验，通过预设的阈值来获得设备当前状态标签。根据实际需要，可以对上述两种方法进行选择，或者将两种方法结合使用。
最终，通过所获得的设备目标属性标签和设备目标状态标签，例如，通过标签值聚类的方法来构建设备画像，并且通过可视化方法(例如，词云图)将这些数据呈现出来。
根据本申请，通过将客户画像技术应用到柱塞泵设备中，构建起了较为完善的柱塞泵设备画像标签体系。
并且，本申请中所构建的标签体系即包含了静态属性和瞬时参量，又包含了参量状态评价和维护状态信息，并提供了基于神经网路进行设备当前状态标签的构建方法，从而便于获得设备当前健康状态等标签，最终获得关于设备的画像，有助于实现柱塞泵故障检测，并能及时准确地掌握设备当前的运行状况，便于掌控整个生产过程。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程，是可以通过计算机程序来指令相关的硬件来完成，所述的计算机程序可存储于一非易失性计算机可读取存储介质中，该计算机程序在 执行时，可包括如上述各方法的实施例的流程。其中，本申请所提供的各实施例中所使用的对存储器、存储、数据库或其它介质的任何引用，均可包括非易失性和/或易失性存储器。非易失性存储器可包括只读存储器(ROM)、可编程ROM(PROM)、电可编程ROM(EPROM)、电可擦除可编程ROM(EEPROM)或闪存。易失性存储器可包括随机存取存储器(RAM)或者外部高速缓冲存储器。作为说明而非局限，RAM以多种形式可得，诸如静态RAM(SRAM)、动态RAM(DRAM)、同步DRAM(SDRAM)、双数据率SDRAM(DDRSDRAM)、增强型SDRAM(ESDRAM)、同步链路(Synchlink)DRAM(SLDRAM)、存储器总线(Rambus)直接RAM(RDRAM)、直接存储器总线动态RAM(DRDRAM)、以及存储器总线动态RAM(RDRAM)等。
以上所述仅为本发明的优选实施例而已，并不用于限制本发明，对于本领域的技术人员来说，本发明可以有各种更改和变化。凡在本发明的精神和原则之内，所作的任何修改、等同替换、改进等，均应包含在本发明的保护范围之内。

Claims (13)

  1. 一种机械设备画像生成方法，其特征在于，所述机械设备画像生成方法包括：
    获取所有与所述机械设备相关的数据信息；
    对所述机械设备相关的数据信息进行数据融合，以获得多源数据信息集合；
    基于所述多源数据信息集合获取所述机械设备的目标属性数据和目标状态数据；
    基于所获取的所述目标属性数据生成所述机械设备的属性标签；
    将所述机械设备的当前目标状态数据输入到训练好的神经网络模型中，获得所述机械设备的当前状态标签，其中，所述当前状态标签包括所述机械设备的状态标签值；
    通过所获取的所述属性标签的属性标签值和所获得的所述当前状态标签的状态标签值，生成所述机械设备的设备画像。
  2. 根据权利要求1所述的机械设备画像生成方法，其特征在于，所述机械设备画像生成方法还包括：
    基于所述机械设备的目标状态数据以及所述目标状态数据的目标状态标签值数量，对神经网络模型进行初始化；
    基于所述目标状态数据生成多个统计指标；以及
    将包括所述多个统计指标以及所述目标状态数据的状态分类标签数据的数据矩阵输入到初始化的所述神经网络模型中，对所述神经网络模型进行训练。
  3. 根据权利要求1所述的机械设备画像生成方法，其特征在于，基于所述多源数据信息集合获取所述机械设备的目标属性数据和目标状态数据，包括：
    基于所述多源数据信息集合获取初始属性数据和初始状态数据；
    基于对所述初始属性数据的关注度和/或重要性，获取所述目标属性数据；
    对所述初始状态数据进行归一化处理，获取所述目标状态数据。
  4. 根据权利要求1所述的机械设备画像生成方法，其特征在于，通过所获取的所述属性标签的属性标签值和所获得的所述当前状态标签的状态标签值，生成所述机械设备的设备画像，包括：
    基于所述机械设备的所述属性标签值和所述当前状态标签值，构建所有标签值集合和设备集合；
    基于所述标签值集合和所述设备集合，构建共现矩阵；
    基于所述共现矩阵，通过聚类算法获取所述属性标签和所述当前状态标签的标签聚类簇；以及
    基于所述标签聚类簇获取设备当前重点标签值，生成所述机械设备的设备画像。
  5. 根据权利要求4所述的机械设备画像生成方法，其特征在于，基于所述标签聚类簇获取设备当前重点标签值，生成所述机械设备的设备画像，包括：
    将各设备所有标签值与所获取的所述标签聚类簇所含标签值进行比较；以及
    获取各设备所含标签值覆盖所述标签聚类簇中标签值类别数最多的第一标签聚类簇，并将所述第一标签聚类簇中所对应的标签值设为所述设备当前重点标签值。
  6. 根据权利要求1至5中任一项所述的机械设备画像生成方法，其特征在于，所述机械设备画像生成方法还包括：
    将所生成的所述设备画像可视化为画像词云图。
  7. 根据权利要求1至5中任一项所述的机械设备画像生成方法，其特征在于，所述机械设备画像生成方法还包括：
    将所述机械设备的当前目标状态数据的状态标签值与设定值进行比较，获得所述机械设备的当前状态标签。
  8. 根据权利要求1所述的机械设备画像生成方法，其特征在于，对所述机械设备相关的数据信息进行数据融合，以获得多源数据信息集合，包括：
    为各设备设定相互不同的唯一标识；
    获取各设备具有唯一标识的第一库表以及与所述第一库表相关联的第二库表；以及
    以所述唯一标识使得所述第一库表和所述第二库表相互关联。
  9. 根据权利要求1所述的机械设备画像生成方法，其特征在于，所述神经网络模型包括反向传播神经网络模型。
  10. 根据权利要求2所述的机械设备画像生成方法，其特征在于，所述统计指标包括：均值、标准差、方根幅值、有效值、峰值、偏度、峭度、最大值、裕度值、波形因子、脉冲指数。
  11. 根据权利要求7所述的机械设备画像生成方法，其特征在于，所述状态标签值包括：正常、故障、高、中和低。
  12. 一种计算机设备，包括存储器及处理器，所述存储器上存储有可在处理器上运行的计算机程序，其特征在于，所述处理器执行所述计算机程序时实现权利要求1至11中任一项所述方法的步骤。
  13. 一种计算机可读存储介质，其上存储有计算机程序，其特征在于，所述计算机程序被处理器执行时实现权利要求1至11中任一项所述的方法的步骤。
PCT/CN2022/090147 2022-02-14 2022-04-29 机械设备画像生成方法 WO2023151191A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US18/160,876 US20230259862A1 (en) 2022-02-14 2023-01-27 Method for generating portrait of mechanical equipment

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202210132467.6A CN114595943A (zh) 2022-02-14 2022-02-14 机械设备画像生成方法
CN202210132467.6 2022-02-14

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US18/160,876 Continuation US20230259862A1 (en) 2022-02-14 2023-01-27 Method for generating portrait of mechanical equipment

Publications (1)

Publication Number Publication Date
WO2023151191A1 true WO2023151191A1 (zh) 2023-08-17

Family

ID=81805454

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/090147 WO2023151191A1 (zh) 2022-02-14 2022-04-29 机械设备画像生成方法

Country Status (2)

Country Link
CN (1) CN114595943A (zh)
WO (1) WO2023151191A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117116407A (zh) * 2023-10-23 2023-11-24 之江实验室 基于电子病历的患者画像生成方法、装置及存储介质

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115544144B (zh) * 2022-10-14 2024-05-31 中电金信软件有限公司 一种标签数据的处理方法及装置

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107423442A (zh) * 2017-08-07 2017-12-01 火烈鸟网络(广州)股份有限公司 基于用户画像行为分析的应用推荐方法及***，储存介质及计算机设备
CN108416620A (zh) * 2018-02-08 2018-08-17 杭州浮云网络科技有限公司 一种基于大数据的画像数据的智能社交广告投放平台
CN112364008A (zh) * 2020-11-20 2021-02-12 国网江苏省电力有限公司营销服务中心 一种面向电力物联网智能终端的设备画像构建方法
US20210216813A1 (en) * 2020-01-15 2021-07-15 Accenture Global Solutions Limited Data clustering

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107423442A (zh) * 2017-08-07 2017-12-01 火烈鸟网络(广州)股份有限公司 基于用户画像行为分析的应用推荐方法及***，储存介质及计算机设备
CN108416620A (zh) * 2018-02-08 2018-08-17 杭州浮云网络科技有限公司 一种基于大数据的画像数据的智能社交广告投放平台
US20210216813A1 (en) * 2020-01-15 2021-07-15 Accenture Global Solutions Limited Data clustering
CN112364008A (zh) * 2020-11-20 2021-02-12 国网江苏省电力有限公司营销服务中心 一种面向电力物联网智能终端的设备画像构建方法

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117116407A (zh) * 2023-10-23 2023-11-24 之江实验室 基于电子病历的患者画像生成方法、装置及存储介质
CN117116407B (zh) * 2023-10-23 2024-02-23 之江实验室 基于电子病历的患者画像生成方法、装置及存储介质

Also Published As

Publication number Publication date
CN114595943A (zh) 2022-06-07

Similar Documents

Publication Publication Date Title
WO2023151191A1 (zh) 机械设备画像生成方法
Cheng et al. Data and knowledge mining with big data towards smart production
CN113723632A (zh) 一种基于知识图谱的工业设备故障诊断方法
US20190286983A1 (en) Machine learning-based semiconductor manufacturing yield prediction system and method
Liao et al. A case-based reasoning system for identifying failure mechanisms
CN107273924B (zh) 基于模糊聚类分析的多数据融合的电厂故障诊断方法
CN104573845A (zh) 信息***设备状态检修辅助决策方法
Vathsala et al. Prediction model for peninsular Indian summer monsoon rainfall using data mining and statistical approaches
US20230259862A1 (en) Method for generating portrait of mechanical equipment
WO2021142622A1 (zh) 确定不良原因的方法、电子设备、存储介质及***
Vychuzhanin et al. Analysis and structuring diagnostic large volume data of technical condition of complex equipment in transport
CN109507992A (zh) 一种机车制动***部件的故障预测方法、装置及设备
CN111144433A (zh) 基于svm模型油井工况智能诊断分析方法及装置
Meroño-Peñuela et al. Detecting and reporting extensional concept drift in statistical linked data
Li et al. A distributed parallel alarm management strategy for alarm reduction in chemical plants
CN117689256A (zh) 一种铝合金熔铸产品质量溯源方法
CN116245212A (zh) 基于pca-lstm的电力数据异常检测与预测方法及***
CN116149895A (zh) 大数据集群性能预测方法、装置和计算机设备
CN108052087A (zh) 基于信息熵的制造过程多元质量诊断分类器
CN112784129A (zh) 一种泵站设备运维数据监管平台
Guo et al. A Novel Financial Risk Analysis and Early Warning Method based on Data Mining
JP2024500175A (ja) 工業プラントのプロセスデータベースシステムから時系列データセットを検索する方法
CN118035249B (zh) 核工程复杂关系数据建模与监控方法和***
CN112633527B (zh) 企业维修网点管理方法、***、计算机设备和存储介质
CN117422312B (zh) 一种企业经营风险的评估方法、介质及装置

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22925551

Country of ref document: EP

Kind code of ref document: A1